<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Takara TLDR - Daily AI Papers</title>
    <link>https://tldr.takara.ai</link>
    <description>Daily AI research papers from Takara.ai</description>
    <lastBuildDate>Fri, 20 Nov 2026 12:21:28 +0000</lastBuildDate>
    <atom:link href="https://tldr.takara.ai/api/papers" rel="self" type="application/rss+xml"></atom:link>
    <item>
      <title>Twice-Told Critiques: Claim-Level Homogenization of Peer Review in the LLM Era</title>
      <link>https://tldr.takara.ai/p/2611.26286</link>
      <description><![CDATA[Peer review relies on several readers finding different problems. We measure critique overlap in conference and journal reviews. At ICLR, measured overlap rose from 12.0% in 2026 to 21.3% in 2027, in reviews from the same stage of the review process, with year-on-year rises at NeurIPS, ICML and the F1000 and PLOS journals. A third ICLR review of a paper added 5.46 new problems to the other two in 2027, against 6.12 in 2026. We apply MELD, a detector of LLM-generated text, to all reviews. Within a paper, two reviews that MELD flags overlap more than two unflagged reviews: by 3.0 points at ICLR in 2024-2025, by 4.5 points in 2026-2027, and by 5.4 points across five journal sources. These results provide evidence that LLM-flagged reviews of one paper are less distinct from each other than unflagged reviews, and suggest that the diversity of critiques among reviews of a single paper is falling.]]></description>
      <pubDate>Thu, 19 Nov 2026 15:45:51 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.26286</guid>
    </item>
    <item>
      <title>Direct3D-V3: Latent-Free Diffusion Transformers in the Native Voxel Domain</title>
      <link>https://tldr.takara.ai/p/2611.26511</link>
      <description><![CDATA[Native 3D generation has largely been built around latent diffusion, with recent advances concentrating on stronger variational autoencoders (VAE) and increasingly elaborate latent representations, from vecset and triplanes to structured sparse voxels. Despite their architectural differences, these methods all share the same premise: geometry is first compressed into a latent space and only then modeled by the generator. We take a different route and propose Direct3D-V3, a latent-free diffusion transformer (DiT) framework that discards the VAE altogether and operates in the native voxel domain, directly modeling raw SDF voxel grids during both training and sampling. To make this direct voxel formulation geometrically reliable, we introduce an implicit manifold regularization over neighboring voxels to suppress non-watertight surfaces, and convert predicted SDF grids into meshes via Marching Cubes so that rendered normal losses can supervise the model directly from its predicted SDF. To improve alignment between the conditioning image and generated geometry, Direct3D-V3 also predicts the camera parameters of the conditioning image as an auxiliary task, encouraging the DiT to learn the mapping from image observations to 3D shape. This latent-free design removes the reconstruction bottleneck, brings training under an objective in the native voxel domain, and yields meshes with substantially richer local detail. The predicted camera and native SDF voxel output further make Direct3D-V3 extensible to generative reconstruction, where diffusion is interleaved with normal-guided refinement to improve alignment with the observed view while the generative prior completes unseen regions. Project page: https://dreamtechai.github.io/Direct3D-V3.]]></description>
      <pubDate>Thu, 19 Nov 2026 17:58:20 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.26511</guid>
    </item>
    <item>
      <title>Copilot or Culprit? How AI Is Changing Open-Source Bug Reporting and Maintenance</title>
      <link>https://tldr.takara.ai/p/2611.25605</link>
      <description><![CDATA[AI-generated bug reports have drawn criticism for unreliable claims and faulty reproducers that burden maintainers. However, as LLM capabilities improve, some maintainers have begun to express more positive views on AI use in bug reporting and maintenance. Understanding this evolution requires examining whether AI-involved reports are becoming more reliable, which uses of AI are associated with valid findings, and how maintainers investigate these reports and act on them. To this end, we conduct the first large-scale study of 120,918 bug reports from 101 open-source projects between January 2024 and September 2026, tracing AI involvement from discovery and reporting through validation and resolution. Our findings show that AI-involved reports become more complete over time, with a growing share of reports confirmed as valid. In bug reporting, AI for reproduction and diagnosis are associated with higher validity, whereas report writing is not. Extends into maintenance, maintainers increasingly use AI to investigate code and recommend fixes. However, they devote more effort to AI-assisted reports of human-discovered bugs than to reports of AI-discovered bugs. These findings highlight both AI's growing participation in bug investigation and the continuing need for verification and follow-up. Drawing on these findings, we provide practical guidance for turning AI-generated findings into verifiable reports: emphasizing reproducibility, verification, and accountability. We followed this guidance when submitting 53 reports on AI-discovered bugs, of which 69.81% were accepted by maintainers as of writing.]]></description>
      <pubDate>Thu, 19 Nov 2026 06:53:32 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.25605</guid>
    </item>
    <item>
      <title>One Prior, Two Dynamics: Repurposing Video Diffusion Transformers for Self-Supervised Latent Action Learning</title>
      <link>https://tldr.takara.ai/p/2611.26617</link>
      <description><![CDATA[Videos reveal how the world changes, but rarely the actions that cause those changes. Latent action learning recovers such supervision from visual transitions, yet often learns the underlying dynamics largely from scratch. We ask whether a pretrained video generator can instead provide the prior for both inverse and forward dynamics. We introduce O$^{2}$LaF, built on a simple principle: one prior, two dynamics. We initialize both models from the same pretrained video DiT and adapt them to their distinct roles: a feed-forward transition encoder with learnable action queries for inverse dynamics, and a causal, action-conditioned video generator for forward dynamics. By controlling information access and optimizing both dynamics end-to-end through a compact bottleneck, O$^{2}$LaF learns transferable actions from unlabeled videos. Across navigation and robotic manipulation, O$^{2}$LaF sets a new state of the art in few-shot action probing and enables stronger zero-shot action transfer across visual contexts and embodiments. Beyond generative video priors, we further extend our framework to predictive representation priors such as V-JEPA2, achieving competitive performance. Our findings suggest a broader paradigm for latent-action learning: repurposing pretrained video models to discover transferable actions and predict their consequences.]]></description>
      <pubDate>Thu, 19 Nov 2026 18:58:14 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.26617</guid>
    </item>
    <item>
      <title>When Safety Rationales Hallucinate: Diagnosing and Grounding Multimodal Safety Guards</title>
      <link>https://tldr.takara.ai/p/2611.25793</link>
      <description><![CDATA[Multimodal safety guards increasingly generate rationales alongside safety predictions, yet whether these rationales are reliable remains largely unexplored. We systematically study *rationale hallucination* from two perspectives: visual rationale hallucination, where rationales contain unsupported visual facts, and safety attribution hallucination, where grounded content is assigned an incorrect safety interpretation. Across rationale-generation methods, multiple vision-language backbones, and safety datasets, we find that rationale hallucination is widespread and does not consistently diminish with higher classification accuracy or increased model scale. Further analysis of the Think-with-Image paradigm shows that hallucination is concentrated in fine-grained attributes and relations, while visual tools are often invoked infrequently or without sufficient spatial focus. Based on these findings, we propose *Tool-Contrastive Grounding Reinforcement Learning* (TCG-RL), which rewards tool use only when it improves rationale grounding over a tool-free reference. TCG-RL substantially reduces rationale hallucination while maintaining competitive safety prediction performance. By revealing the gap between prediction correctness and rationale reliability, our study motivates a shift in multimodal safety guarding from outcome-only evaluation toward safety decisions that are both accurate and verifiably grounded.]]></description>
      <pubDate>Thu, 19 Nov 2026 09:22:27 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.25793</guid>
    </item>
    <item>
      <title>SoK: Assessing the Evolution of LLM Capability in Software Vulnerability Research</title>
      <link>https://tldr.takara.ai/p/2611.26601</link>
      <description><![CDATA[LLM-based agents are becoming increasingly capable at discovering, validating and patching vulnerabilities in real software, and vendors and researchers report that this is already changing how vulnerabilities are found and fixed. These claims, however, are scattered across papers, vendor reports and disclosures and are rarely comparable. We introduce a framework that rates these capabilities on an autonomy scale modeled on the levels of driving automation, and qualifies each claim by its correctness, its availability and the strength of its evidence. We use this framework to systematize the published evidence on LLM capability in vulnerability research. We find that agents discovered, validated, and patched vulnerabilities in real-world software without human intervention before the restricted frontier models of 2026, and that open-weight models subsequently attained comparable autonomy for discovery and validation. We then measure the consequences of evolving LLM security capabilities in a decade of public vulnerability records, advisories and exploitation data. In 2026 the number of disclosed vulnerabilities more than doubled, with an increasing share rated critical or high, and severe bugs that had survived a decade or more in widely used code were reported several times more often than before. Yet the time between the stages of a vulnerability's life, from report to fix and from disclosure to attack, has not shifted noticeably so far. We grade the mitigations proposed in the literature with an explicit rubric and draw guidance for each group of stakeholders from the best-supported ones.]]></description>
      <pubDate>Thu, 19 Nov 2026 18:54:19 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.26601</guid>
    </item>
    <item>
      <title>Actor-Predictor Architecture for Reinforcement Learning with Formal Specification Objectives</title>
      <link>https://tldr.takara.ai/p/2611.26313</link>
      <description><![CDATA[One fundamental challenge in reinforcement learning (RL) is enforcing temporal safety constraints, such as satisfying formal specifications. Existing methods often transform logic formulas into automata for state augmentation and reward shaping. However, this approach assumes that automaton construction is both feasible and tractable, which limits its applicability to small specifications and linear temporal logic (LTL). This paper proposes Actor-Predictor, a framework that learns a surrogate representation of logic progress without explicitly constructing an automaton. The predictor continuously estimates whether a partial trajectory will satisfy the formal specification in a later step, and its learned features are provided as auxiliary inputs to the actor, which is then optimized using GRPO and RWR. We show theoretically that, under sufficient predictive accuracy, the learned representation is equivalent to the automaton state. Experiments on grid world and robotic tasks in Isaac Lab demonstrate that Actor-Predictor outperforms automaton-based baselines on LTL-constrained tasks. It also performs well on non-regular formal specifications, such as $(a^n b^n)$, for which finite-automaton-based approaches are not applicable.]]></description>
      <pubDate>Thu, 19 Nov 2026 16:03:02 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.26313</guid>
    </item>
    <item>
      <title>Sirius-Tiamat: How Small Can a One-Pass Decision Encoder Get Before Its Decisions Change?</title>
      <link>https://tldr.takara.ai/p/2611.26339</link>
      <description><![CDATA[A one-pass decision encoder reads a text, a question and several named options, and returns a probability for each option in a single forward pass. We ask how far Laya, a 421M-parameter encoder of this kind, can be compressed before its decisions change. We count decisions that differ from Laya's own full-precision (fp32) answer, and track calibration (how well confidence matches accuracy, measured by expected calibration error, ECE), on 1,221 synthetic items with the training schemas, 273 with an unseen schema, and 800 out-of-family questions from ARC, MMLU and HellaSwag. Layer pruning of a related model family is known; we add flips, calibration, held-out data and retraining. bf16 changes 0.2% of in-family decisions but 2.2% out of family; 6-bit grouped rounding 1.3% and 5.9%; deleting 6 layers 0.9% and 9.1%, while ECE rises from 5.9% to 39.2%. One fitted temperature, needing no labels, restores it to 3.6%. One scale per tensor fails below six bits, and CPU dynamic int8 changes 30% of decisions, nearly all through one layer. Total variation tracks flips for rounding but not for deletion, and an early exit saves layers over a static cut only if its confidence gate is first recalibrated. Distilling a 16-layer student cuts its in-family flips from 44% to 9% (45% out of family). At equal stored size, rounding flips fewer decisions than deletion. Data are synthetic and LLM-labelled, for one architecture.]]></description>
      <pubDate>Thu, 19 Nov 2026 16:17:13 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.26339</guid>
    </item>
    <item>
      <title>Power Setpoint Tracking for Large-Scale EV Charging: Transformer-Constrained Charging Policy Learning with Self-Tuning Constraint Prices</title>
      <link>https://tldr.takara.ai/p/2611.25965</link>
      <description><![CDATA[In power setpoint tracking (PST), an operator follows a purchased charging profile without overloading shared transformers or undercharging vehicles. PST is cast as a constrained Markov decision process that minimizes tracking error subject to overload and mean user-satisfaction constraints. Transformer-Constrained Charging Policy Learning (TC-CPL) extends soft actor-critic with proportional-integral Lagrange-multiplier updates from measured violations, a residual on the incumbent dispatcher that is warm-started from its demonstrations, and a transformer-aware encoder shared by all ports. It observes no departure times or states of charge. Simulations (five seeds, 100 paired days) compare it with the incumbent, Round-Robin. On a 600-port, 122-substation district, raw overload falls from 3.59 to 0.0017 kWh/day (overloaded seed-days from 77% to 0.4%), and absolute tracking error falls by 27%. Delivered energy rises by 16%, and satisfaction rises from 0.959 to 0.989. On a 42-kW transformer with household load, overload falls from 1.86 to 0.0014 kWh/day and satisfaction rises from 0.974 to 0.985, but absolute tracking error is 22% higher. Against hand-set fixed multipliers, self-tuning is statistically indistinguishable on the district. On the single transformer, it raises satisfaction by 0.011 at similar overload, but a short sweep found a better-serving fixed pair. The overload multipliers stay below their cap.]]></description>
      <pubDate>Thu, 19 Nov 2026 11:51:56 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.25965</guid>
    </item>
    <item>
      <title>Diagnosis is not a policy: a state-transition audit of caoability-conditioned test-time inference</title>
      <link>https://tldr.takara.ai/p/2611.25566</link>
      <description><![CDATA[Test-time actions can satisfy retrieval, state-realization, program-consistency, or rendering objectives without correcting an answer. Test-time Error Response and Realizability Audit (TERRA) is an evaluation/auditing protocol, not a new inference algorithm. It uses a unified decomposition of intervention outcomes into applicability, target-state realization, endpoint translation, harm, and resource/inference cost. Paired comparisons fix the incoming state and evaluation conditions; cross-model analyses vary the receiving configuration. Executable financial reasoning experiments yield three findings: the same context expansion can harm one configuration and benefit another; sufficient evidence does not ensure correctness because configurations differ in evidence-to-answer translation; and diagnostic headroom---accuracy attainable with idealized diagnostic information over the measured baseline---plus local consistency does not ensure practical gains. Action value is $V(a\mid x,m)$, not diagnosis alone; diagnosis is not a deployable policy $π(a\mid x,m)$.]]></description>
      <pubDate>Thu, 19 Nov 2026 06:09:20 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.25566</guid>
    </item>
    <item>
      <title>Grid Cells as Spectral Successor Features: A Reusable Basis for Continual Reinforcement Learning</title>
      <link>https://tldr.takara.ai/p/2611.26504</link>
      <description><![CDATA[Continual reinforcement learning requires agents to adapt when the task changes while reusing what they know about the environment. Successor features separate dynamics from task-specific reward, but their effectiveness depends on the choice of base features. Taking the identification of entorhinal grid cells with eigenvectors of the successor representation as a design principle, we build this basis from eigenvectors of the transition matrix of a fixed reference policy. With such a basis, the successor features at a state-action pair are a linear combination of the base features at that same pair, which is not true in general, and the coefficients are known in closed form: each successor feature is its base feature multiplied by $θ_i = 1/(1-γλ_i)$, where $λ_i$ is its eigenvalue. The long-term return is thus predicted from the same features as the immediate reward, no successor-feature network or replay buffer is needed, and adapting to a new task reduces to a linear regression on the reward. The agent acts greedily on the action values of the reference policy, which is one step of policy improvement over it and suffices to reach a single goal. In a two-goal gridworld, with the uniform random walk as reference policy, whose eigenvectors are the low-frequency Laplacian eigenvectors, agents built this way maintain near-maximal return across every change of goal, whereas SimpleSF and DQN lose most of their return at each change.]]></description>
      <pubDate>Thu, 19 Nov 2026 17:54:04 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.26504</guid>
    </item>
    <item>
      <title>SteerTrace: Tracing How Target Attacks Steer World Action Models</title>
      <link>https://tldr.takara.ai/p/2611.25719</link>
      <description><![CDATA[World action models (WAMs) predict future videos and actions with one backbone, and their imagined future is often seen as a safety signal. We introduce SteerTrace, a targeted universal attack on WAMs that also traces its effect inside the model. One $\ell_\infty$-bounded (4/255) perturbation per model, shared across frames, tasks, and camera views, pushes the executed action toward an attacker-chosen target. On eight WAMs from four families (FastWAM, Faster-WAM, OpenWAM and LingBot-VA) on LIBERO and RoboTwin, it drives task success close to zero and moves the attacked action toward the target. Same-budget noise and BadWAM produce little action shift in the target direction. We use the signed targeted effect to measure how much of the effect each intervention removes or reproduces. Under the standard attack, restoring future tokens removes 61-94% of the targeted effect in the five models whose action experts attend to them. When the attack is constrained to keep the predicted future close to the clean prediction, the targeted effect can become more dependent on current-frame tokens. In three of these five models, restoring current-frame tokens removes 87-99% of the effect, compared with 9-15% for future tokens. In six models, restoring just 0.36% of the FFN neurons in layers 10-29 of the video expert removes 72-96% of the targeted effect. Residual directions 2-16 hold only 11-18% of the shift energy, yet removing them cancels most of the targeted effect.]]></description>
      <pubDate>Thu, 19 Nov 2026 08:32:45 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.25719</guid>
    </item>
    <item>
      <title>SkillLens: Action-Guided Dense Visual Dynamics for Proficiency Estimation</title>
      <link>https://tldr.takara.ai/p/2611.25712</link>
      <description><![CDATA[Human proficiency estimation is challenging because skill is a high-level and often subjective property that is not captured by action identity alone. Existing approaches leverage multimodal observations such as ego-exo video, gaze, pose, and inertial signals, but the gap between sensor-level evidence and abstract proficiency labels remains difficult to bridge. We present SkillLens, a proficiency estimation framework that uses atomic action descriptions as temporal guidance, rather than direct semantic supervision, to localize proficiency-relevant windows for dense visual-dynamics extraction. Each window is summarized into an action-dynamics feature and scored by a high-low prototype margin against higher- and lower-proficiency executions from the same activity-action category. Fusing these ego/exo PDM cues with a global ego-exo-gaze baseline through CORAL ordinal calibration, our method achieves 0.542 accuracy on Ego-Exo4D. Our project page is available at https://skill-lens.net/]]></description>
      <pubDate>Thu, 19 Nov 2026 08:24:26 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.25712</guid>
    </item>
    <item>
      <title>One Framework to Diagnose Them All: A Stratified Benchmark for Bengali LLM Hallucination</title>
      <link>https://tldr.takara.ai/p/2611.26336</link>
      <description><![CDATA[Hallucination research in low-resource languages fragments easily: each newly named failure mode prompts its own dataset, spending scarce annotators on incomparable resources. We argue for consolidation and test it at community scale. Alikbachan is a Bengali hallucination-detection benchmark stratified along three crossed axes: a human-validated cultural-distance band, six task types, and a five-family, 32-mode hallucination taxonomy. It comprises a harvested 2,516-item test set (Phase 1) and a quota-constructed, gold-labeled 5,000-item set (Phase 2). The stratification is diagnostic. Four zero-shot LLM judges vary by over 30 macro-F1 points across strata, and the item-level predictions of 99 Phase-1 and 32 Phase-2 competition systems reproduce the profile: discourse-level omission defeats judges and most teams alike, whereas only a minority over-flag negative controls. The judges' verdict probabilities rank items well (ROC-AUC 0.74-0.95) but are overconfident (ECE 0.04-0.29). An open competition on Alikbachan drew 222 teams and 3,566 submissions, and a single-prompt 32B judge reaches the 57th percentile. Of the 25 teams in both phases, 24 regress on Phase 2 (mean $-17.6$ points) while every zero-shot judge improves, and Phase-1 rank barely predicts Phase-2 rank (Spearman $ρ=0.35$), so a Phase-1 leaderboard alone would have crowned a different winner.]]></description>
      <pubDate>Thu, 19 Nov 2026 16:15:38 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.26336</guid>
    </item>
    <item>
      <title>Attention Head Specialization: Phase Transitions in Sample Complexity</title>
      <link>https://tldr.takara.ai/p/2611.26228</link>
      <description><![CDATA[Attention heads specialize into distinct functions during training, often emerging in stages together with sharp changes in model behavior. While this phenomenon has mostly been studied as a function of training time, we ask how head specialization depends on the amount of available training data. Empirically, we find that heads specialized to syntactic relations in BERT emerge as the dataset size increases. We then study this phenomenon in a solvable high-dimensional model of multi-head attention and establish that, in this model, head specialization is a genuine phase transition in sample complexity, corresponding to a non-analyticity in the test error and other observables. Depending on data structure and regularization, specialization can follow uninformative head differentiation or occur through a cascade of phase transitions that progressively recover weaker latent features, providing a statistical form of simplicity bias. Finally, we relate these statistical transitions to the stagewise emergence of specialized heads during online learning, disentangling statistical from dynamical origins of head specialization.]]></description>
      <pubDate>Thu, 19 Nov 2026 15:02:51 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.26228</guid>
    </item>
    <item>
      <title>SheetAlign: Grounded Interaction and Stage-Isolated Learning for Spreadsheet Agents</title>
      <link>https://tldr.takara.ai/p/2611.26069</link>
      <description><![CDATA[Spreadsheets are widely used for data analysis and supporting decision-making. Automating spreadsheet analysis while protecting sensitive data and controlling deployment costs motivates locally deployed open-weight large language models (LLMs). Meanwhile, complex spreadsheet tasks require both task reasoning and workbook understanding, while predefined operation interfaces impose additional tool-selection and invocation demands. We introduce SheetAlign, a spreadsheet manipulation framework with stage-isolated policy learning, organized around where to operate, what to change, and how to implement transformations. Task-guided inspection and API grounding anchor code generation in workbook observations and available operations, while stage-isolated learning optimizes program generation and review using edit-progress rewards. At inference, execution consistency selects programs through agreement among workbook states. On SpreadsheetBench, our grounded task interface improves Qwen3-4B's accuracy from 13.5% to 29.2% on average without training. With learning and execution consistency, SheetAlign achieves 32.8% accuracy on average on SpreadsheetBench and 30.1% on Domain-Spreadsheet, outperforming comparable baselines. The code will be made publicly available upon publication.]]></description>
      <pubDate>Thu, 19 Nov 2026 13:13:12 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.26069</guid>
    </item>
    <item>
      <title>Safe Subgame Resolving in Multiplayer Games</title>
      <link>https://tldr.takara.ai/p/2611.25981</link>
      <description><![CDATA[Poker and other games with hidden information are too large to solve in one pass. Strong systems compute a complete blueprint strategy offline and refine it in real time inside the reached subgame. In 2p0s games, a safety theorem guarantees that subgame resolving never increases how much an adversary can gain by exploiting the resolved strategy relative to the blueprint strategy. With three or more players acting independently, no such guarantee exists yet, and heuristic resolving can increase the profile's best-response gap. We prove the first conditional safety theorem for uncorrelated multiplayer subgame resolving, a post-hoc certificate for any delivered update. The best-response gap of the resolved profile exceeds that of the blueprint by at most four terms. Two concern the resolver: the subgame value the blueprint promised without realizing ($A_i^{S}$) and the resolver's subgame solve error ($\varepsilon_i$). The other two are the opponents' best-response gains inside the subgame over their promised values ($\sum_q M_{q}$) and a structural tax ($Θ$), which measures how much the players' total utility can change. The tax is zero in constant-sum games, including every rake-free poker variant. Constructions attain each term with equality. On three-player Leduc poker and on Kuhn poker with up to six players, resolving without a certificate raises the gap beyond a small tolerance in 12% to 100% of the updates. The certificate flags every such update and accepts 41% (Kuhn) and 86% (Leduc) of the standard resolving updates. A resolver that keeps the blueprint whenever the certificate exceeds the tolerance delivers no harmful update. The certificate needs only one best-response pass per opponent inside the subgame, so real-time multiplayer solvers that refine a blueprint can attach it as a safety check. Code is available at https://github.com/lbn187/MultiPlayerSolver]]></description>
      <pubDate>Thu, 19 Nov 2026 12:04:23 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.25981</guid>
    </item>
    <item>
      <title>Evaluating safeguarding in multi-turn conversations with AI assistants</title>
      <link>https://tldr.takara.ai/p/2611.26009</link>
      <description><![CDATA[Conversational AI assistants are now used for personal and emotional support. Developers report improvements in how new models respond to sensitive conversations, yet users and civil society continue to raise concerns about their behaviour in real-world, repeated interactions that safety evaluations rarely capture. Often, evaluations present signs of distress clearly and in a single message, even though people may reveal their struggles gradually in otherwise ordinary conversations. We worked with three community mental health support workers to test if models provide the same level of safeguarding when cues are shared all at once or gradually over time. We designed nine scenarios in which a user shares cues related to depression, psychosis, or domestic abuse. For each scenario, the support workers established a gold standard response, ranging from signposting support to pausing the everyday task to address safeguarding. We presented five cues either together in one message or dispersed across a conversation. Across more than 7,400 conversations, we evaluated six widely used models from five developers. GPT-5.5, Grok 4.3, DeepSeek V4, and Llama 4 Maverick met the support workers' standards in just 26-59% of conversations when cues were concentrated, and only 3-6% when dispersed. Claude Sonnet 5 and Claude Opus 5 showed a smaller gap, meeting the standards in 68-97% of concentrated conversations and 59-87% of dispersed conversations. Even the best-performing model, Claude Opus 5, fell short of the gold standard in 13% of conversations when cues were dispersed. Safety evaluations centred on explicit, legible concerns can overstate how well safeguards protect people when signs of distress emerge gradually or indirectly through ordinary conversations. For claims of improved safety to be taken seriously, developers must design and assess safeguards around how people actually communicate.]]></description>
      <pubDate>Thu, 19 Nov 2026 12:28:53 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.26009</guid>
    </item>
    <item>
      <title>WHU-Road3D: A dataset and benchmark towards fine-grained road surface mapping</title>
      <link>https://tldr.takara.ai/p/2611.25873</link>
      <description><![CDATA[Accurately extracting and mapping road surface elements is crucial for applications such as high-definition maps, road asset management, and intelligent transportation systems. This study first provides a comprehensive review of the benchmark datasets for extracting and mapping road elements, encompassing overhead-to-perspective data, large-to-local scales, and coarse-to-fine details. Subsequently, we thoroughly review the relevant approaches for extracting road surface elements. However, we argue that existing datasets still require enhancements in semantic richness, data modality diversity, and scenario complexity. To bridge these gaps, this paper presents WHU-Road3D, a large-scale multimodal dataset for fine-grained road surface mapping to address these shortcomings. WHU-Road3D provides approximately 100 km of diverse Chinese highways and urban streets. It features a rich combination of MLS point clouds, associated panoramic images, vehicle trajectories, detailed vectorized annotations, and pointwise instance annotations for road surface elements. The dataset carefully annotates 15 distinct categories of road elements, including common markings like road boundaries, zebra crossings, lane dividers, stop lines, as well as 10 types of direction symbols. To our knowledge, WHU-Road3D stands as the first multimodal dataset for mapping such a comprehensive range of road surface elements, designed to facilitate cutting-edge research in instance segmentation, full-element extraction and mapping on road surfaces. Beyond these immediate applications, WHU-Road3D also provides a valuable source for hierarchical road topology reasoning, map updating, and street scene generation. The dataset, along with the code for our benchmark methods, will be made publicly available at \href{https://github.com/WHU-USI3DV/WHU-Road3D.git}{https://github.com/WHU-USI3DV/WHU-Road3D.git}.]]></description>
      <pubDate>Thu, 19 Nov 2026 10:30:10 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.25873</guid>
    </item>
    <item>
      <title>STARK: Moving towards Personalized Agents That Truly Learn and Remember</title>
      <link>https://tldr.takara.ai/p/2611.25705</link>
      <description><![CDATA[Long-term memory is essential for personalized agentic systems that interact with users over long time periods, yet current architectures only capture what happened while leaving the reasoning of why it happened to be derived later by underlying LLMs.%, and silo knowledge within individual users. These events can, however, convey not only what a user experienced but also how that experience shaped their behavioral patterns. We introduce STARK, a dual-graph memory architecture which treats these events not just as nodes but also as edges between behavioral insights. STARK maintains an Episodic Graph which records interactions as temporally, causally, and semantically linked episodes; and an Insight Graph which represents LLM-extracted behavioral signals scored by how many events support them, connected by the very events that led to them. At inference, behavioral knowledge is retrieved directly from insight nodes grounded by event-as-edges, eliminating repeated episodic traversal. The event-as-edge structure also naturally enables projecting the knowledge graph into a relational embedding space where cross-user transfer becomes possible under shared contextual circumstances, allowing the system to infer missing insights for cold-start users. Evaluations show that STARK improves by 1.48 judge score on LongMemEval information extraction and 0.93 on LoCoMo open domain questions over state of the art. On the public LongLaMP abstract-generation benchmark, cross-user transfer improves open-ended judge scores by 0.7 points and ROUGE-L by 3 at 5% graph sparsity.]]></description>
      <pubDate>Thu, 19 Nov 2026 08:19:28 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.25705</guid>
    </item>
    <item>
      <title>ResVision: Spectrum-Matched Resizing and Context-Preserving Inspection for Visual Reasoning</title>
      <link>https://tldr.takara.ai/p/2611.26318</link>
      <description><![CDATA[Agentic visual reasoning improves vision-language models by selectively enlarging evidence without adding new scene content, but the benefit depends on whether fine structure remains accessible and whether local observations stay grounded in the full scene. We identify three recurring bottlenecks: high-frequency detail can be attenuated during visual-token formation; spatially distributed evidence can remain encoded yet difficult for the language model to retrieve; and isolated crops can weaken whole-detail correspondence through scale and positional mismatch. ResVision addresses these limitations with two complementary components. Spectrum-Matched Resizing (SMR) estimates image complexity and selects an adaptive global sampling scale, avoiding unnecessary enlargement on simple inputs while allocating more visual tokens to detail-rich images. Context-Preserving Inspection (CPI) returns each inspected crop together with the complete working view, preserving local detail and global context across tool turns. We evaluate Direct and Agentic variants on four frontier VLMs and 29 benchmarks spanning six visual-reasoning families. Complete ResVision (Agentic Base plus SMR plus CPI) is best for every model, improving by 3.36 to 4.94 points over Direct Base and 1.87 to 2.94 points over Agentic Base. These results show that adaptive visual sampling and context-preserving inspection address distinct, complementary constraints in multimodal reasoning.]]></description>
      <pubDate>Thu, 19 Nov 2026 16:04:30 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.26318</guid>
    </item>
    <item>
      <title>Exact Bayesian layers via characteristic functions</title>
      <link>https://tldr.takara.ai/p/2611.26426</link>
      <description><![CDATA[Bayesian neural networks provide principled uncertainty estimates, but training them with Monte Carlo sampling can be expensive. Sampling-free variational inference algorithms make training Bayesian neural networks feasible, but existing approaches assume a Gaussian output distribution, which can be inaccurate in applied problems. Here, we introduce CF-Prop, an uncertainty propagation method that uses characteristic functions to compute the exact output distribution of a Bayesian ReLU layer, without any sampling, even in non-Gaussian settings. Because CF-Prop propagates characteristic functions, the evidence lower bound---the objective of variational training---can be computed in closed form for a broad class of likelihoods, including the Gaussian, Poisson, and Gamma. This allows to implement an efficient resulting variational inference algorithm, CF-VI. On CIFAR-100, CF-VI achieves the lowest calibration error of any tested method, improving on the next best method by about 35%. On a heavy-tailed insurance claim dataset, CF-VI accurately captures tail uncertainty that is severely underestimated by both other sampling-free methods and Monte Carlo methods.]]></description>
      <pubDate>Thu, 19 Nov 2026 17:04:30 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.26426</guid>
    </item>
    <item>
      <title>Deviations from the NTK Regime in CNNs under Joint Depth-Width Scaling</title>
      <link>https://tldr.takara.ai/p/2611.25624</link>
      <description><![CDATA[The concentration and training-time constancy of the neural tangent kernel (NTK) depend on the interplay between network width, depth, and initialization. While joint depth-width scaling has been studied in fully-connected, residual, and recurrent networks, its effects in convolutional neural networks (CNNs) remain less understood. We investigate these effects numerically in CNNs, examining NTK dispersion at initialization and kernel rotation during training. Let depth $L$ be the number of convolutional layers, and width $C$ the number of channels per layer. We find that NTK dispersion increases with $L/C$ in the critical and chaotic initialization regimes, but remains small in the ordered regime. At critical initialization, dispersion curves approximately collapse as a function of $L/C$, suggesting a nontrivial joint scaling regime. During training, kernel rotation generally increases with $L/C$, even in the ordered regime where initial dispersion is small. Together, these results suggest that $L/C$ may govern NTK deviations from the infinite-width limit in certain regimes, and that NTK concentration at initialization and stability during training exhibit different scaling behaviors.]]></description>
      <pubDate>Thu, 19 Nov 2026 07:09:49 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.25624</guid>
    </item>
    <item>
      <title>MoGloss: Joint Motion and Pseudo-Gloss Pretraining for American Sign Language Understanding</title>
      <link>https://tldr.takara.ai/p/2611.25333</link>
      <description><![CDATA[Sign language translation and recognition have been shown to benefit from various pretraining approaches. We propose a novel two-stage pretraining approach for the motion encoder, which first learns to reconstruct masked motion, followed by joint masked prediction of pseudo-gloss units and motion. Pseudo-glosses serve as an intermediate structured representation of natural language text and are obtained from the captions by a large language model. We curate a corpus of about 1,100 hours of American Sign Language from five public datasets, estimate 2D pose for all clips and define an evaluation protocol. We then reproduce state-of-the-art translation methods and evaluate them using the same train and test splits. We propose a translation model built on our pretraining strategy, and this model outperforms the previous state-of-the-art methods Uni-Sign and SHuBERT, as well as the YouTube-ASL baseline, on the How2Sign, ASL STEM Wiki and OpenASL datasets. Furthermore, the gain from the joint pretraining using motion and pseudo-gloss grows as translation data shrinks. A classifier built on the same encoder also achieves state-of-the-art accuracy on ASL Citizen isolated sign recognition and outperforms our reproduction of Uni-Sign on WLASL2000.]]></description>
      <pubDate>Thu, 19 Nov 2026 01:16:07 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.25333</guid>
    </item>
    <item>
      <title>QUBO-based feature selection for MRI undersampling pattern design</title>
      <link>https://tldr.takara.ai/p/2611.25625</link>
      <description><![CDATA[We formulate undersampling mask design for Cartesian magnetic resonance imaging (MRI) as a feature subset selection problem. Each sampling line is represented by a binary feature indicating whether it is included in each candidate mask. Using candidate masks and their training-set evaluations, we estimate mutual information (MI) and conditional mutual information (CMI). These estimates and a sampling-count penalty define a quadratic unconstrained binary optimization (QUBO) model. The model is solved using simulated annealing and quantum annealing to determine a feature subset and its corresponding sampling mask. We assess the selected subsets on the test set through zero-filled reconstruction in Fourier undersampling simulations using fastMRI knee magnitude reference images, with patient-disjoint training and test sets. At all six tested target sampling rates from 15% to 40%, the resulting reconstructions have lower mean root mean square error than those obtained with four conventional sampling methods. They also have higher mean structural similarity than those obtained using univariate mutual-information ranking at all tested rates. These results support information-theoretic feature selection as an approach to MRI undersampling mask design under the tested zero-filled simulation settings.]]></description>
      <pubDate>Thu, 19 Nov 2026 07:10:59 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.25625</guid>
    </item>
    <item>
      <title>RL Post-Training in Diffusion Transformers Is Activation Steering</title>
      <link>https://tldr.takara.ai/p/2611.26627</link>
      <description><![CDATA[Reinforcement learning (RL) post-training substantially improves the visual quality, text rendering, and compositional accuracy of diffusion transformers. Because capabilities such as rendering requested text or composing multiple objects are inherently prompt-dependent, a natural hypothesis is that improving them requires learning complex, input-dependent transformations. To isolate how much of these gains an input-independent shift explains, we probe each RL-tuned model with a deliberately simple parameterization: a fixed set of steering vectors added to the block outputs of the frozen base model. On SD3.5-M, these vectors hold 65x fewer parameters than the RL-tuned LoRA (0.29M against 18.8M). We fit them to match the velocity predictions of public RL-tuned checkpoints, with no reward in the objective. Although the vectors are identical for every prompt, token, and sampling seed, they recover 90.0-96.3% of the reward gains of six RL-tuned image models, and 89.8% and 111.9% on two RL-tuned video models. Initialized at zero and trained with online RL, the vectors alone exceed the reward gains of weight-based RL on text rendering (OCR) and human preference (PickScore) on SD3.5-M. A constant shift thus suffices for most of the gain of policy-gradient RL post-training: without new input-dependent transformations, it steers the pretrained generator toward what it can already compute.]]></description>
      <pubDate>Thu, 19 Nov 2026 18:59:53 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.26627</guid>
    </item>
    <item>
      <title>Evolve Where It Matters: Coverage-Guided Self-Evolution of Agent Skills</title>
      <link>https://tldr.takara.ai/p/2611.25429</link>
      <description><![CDATA[Agent skills package procedural knowledge as natural-language instructions, and a skill helps only when its instructions are correct and properly followed by the agent. Self-evolving agents revise their own skills from the runs they make, but existing methods judge each run as a whole, by its outcome or by where it went wrong, and cannot tell whether a failure comes from a wrong instruction or from one the agent did not follow. Instruction-level coverage was introduced to test skills and labels each instruction of a run as follow, not-follow, or not covered, which separates these causes. We leverage it for skill self-evolution with SkillPlumber, a framework in which an agent evolves its own skills where coverage shows they fall short. A persistent root cause analysis agent turns each label into a revision question and checks each revision against the next run, while a maintainer distills investigations across tasks into a shared strategy library. Across three benchmarks and three executors, SkillPlumber solves 64.7% to 70.5% of the tasks, 25.6 to 30.4 percentage points higher than the initial skill and 8.7 to 14.0 percentage points higher than the strongest baseline, and removing coverage reduces recovered failures by 20.8% on SkillsBench with GPT-5.5.]]></description>
      <pubDate>Thu, 19 Nov 2026 03:11:24 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.25429</guid>
    </item>
    <item>
      <title>SiliconGym: A Versatile Reinforcement Learning Environment for Agentic EDA</title>
      <link>https://tldr.takara.ai/p/2611.25658</link>
      <description><![CDATA[Reinforcement learning (RL) with electronic design automation (EDA) feedback is advancing large language models for hardware design and verification. Extending these gains to hardware-engineering agents requires environments that span tasks, preserve state across tool calls, and support large hierarchies alongside parallel rollouts. We present SiliconGym, the first versatile, cross-task and cross-toolchain RL environment for hardware design and verification. By decoupling versioned workspaces, budgeted execution, and acceptance-gated scoring, SiliconGym scales from module-level RTL, testbench, and PPA tasks to multi-round repository repair and full-hierarchy SoC flows. Streaming dispatch overlaps tool checks with generation, finishing a 256-answer RL batch 21.9% faster than native ChipSeek. Using interchangeable open-source and commercial backends, we show that open-source-trained functional correctness transfers cleanly to commercial simulation, while paired netlist evaluation reveals when open-source PPA gains survive commercial synthesis.]]></description>
      <pubDate>Thu, 19 Nov 2026 07:40:52 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.25658</guid>
    </item>
    <item>
      <title>Efficient Latent Reasoning for Vision-Language Model via Dynamics-Aware Quantization and Distillation</title>
      <link>https://tldr.takara.ai/p/2611.25670</link>
      <description><![CDATA[Efficient chain-of-thought (CoT) reasoning in vision-language models (VLMs) requires reducing both intermediate reasoning steps and their execution cost. Latent CoT methods compress textual rationales into continuous thoughts, yet their full-precision (FP) models retain substantial storage and computation costs. Lowering numerical precision could reduce these remaining costs, yet direct quantization severely degrades reasoning accuracy, particularly with 4-bit weights and activations. Operator and perturbation analyses reveal key sensitivity and direction-dependent predictive effects of continuous-thought errors, motivating preservation of information retrieval and subsequent reasoning beyond reconstruction accuracy. We therefore introduce QL-VLM, a dynamics-aware Quantization framework for efficient Latent reasoning in Vision-Language Models (VLMs). First, Dynamics-Guided Quantization (DGQ) uses nonuniform codebooks that preserve rotary-key structure, mitigating quantization-induced distortion in vision-language information retrieval. Second, Dynamics-Preserving Distillation (DPD) aligns continuous thoughts, latent-query readouts, and predictions with a full-precision teacher, maintaining predictive consistency through subsequent low-precision reasoning. On ScienceQA-IMG with Qwen2-VL-2B-Instruct, 4-bit QL-VLM achieves 90.43% accuracy and improves baseline by 40.33%, with 16.64x fewer bit-operations and 6.40x lower measured latency than explicit CoT.]]></description>
      <pubDate>Thu, 19 Nov 2026 07:48:38 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.25670</guid>
    </item>
    <item>
      <title>From AI Surrogate to Engineering Tool: Toward Production Use of Fiber-Orientation Prediction in Crash CAE</title>
      <link>https://tldr.takara.ai/p/2611.25991</link>
      <description><![CDATA[Crash simulations of glass-fiber-reinforced plastic parts are highly sensitive to the fiber-orientation field supplied to the material model. In conventional workflows, that field comes from resin-flow analysis, which requires molding-specialist support and takes about 25 hours of work per part. We previously proposed a surrogate model that addresses this bottleneck by predicting the orientation field in about 3 minutes. This paper reports the implementation, validation, and operational design that turned the model into an engineering tool for Honda design workflows. We present staged validation from 2D plates to production parts, evaluation against a predefined engineering angular-error criterion, and the expert review that approved design use in March 2025; the system has since been transferred to the crash-safety group and is in initial, limited use in directed fracture-prediction scenarios. The developed system combines injection-gate setting support for cases in which gates are unknown (90% gate-count accuracy, mean position error 11% of flow length), voxelization, surrogate inference, automatic mapping to the crash CAE model, and report generation in a single GUI. The complete orientation-preparation workflow takes about 0.1 hours, a 99.6% reduction from the conventional 25-hour flow-analysis route, and can be run directly by crash CAE engineers without molding-specialist support. As evidence of practical impact, component-level crash analysis using the predicted orientation field reduced the absolute peak-force error from 19.0% with an isotropic baseline to 11.4%, while confining predicted fracture to the test-observed region. The deployment lesson is that a surrogate model alone is not enough for practical adoption: gate-setting support, automatic data conversion, and shareable reporting are what make the model usable as a design tool.]]></description>
      <pubDate>Thu, 19 Nov 2026 12:12:45 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.25991</guid>
    </item>
    <item>
      <title>MemMend: A Selective Revision Framework for Memory Question Answering</title>
      <link>https://tldr.takara.ai/p/2611.25721</link>
      <description><![CDATA[Revising every answer from a long-term memory can consume substantial inference budget and damage facts that were already correct. We introduce MemMend, a modular selective revision framework that routes potentially incomplete answers to an editor, represents corrections as bounded exact-text replacements, and accepts only edits with resolvable source coordinates and valid edit operations. The editor preserves the current answer when no legal repair is proposed; accepted edits require no additional semantic-verification call. On 1,986 LoCoMo development questions, MemMend achieves 67.17 conversation-macro F1 and 88.35 mean Judge score, improving over a verified revision chain by 0.82 and 1.24 points while using 62.37% fewer revision calls. Routing uses 54.15% fewer revision calls than ungated local review, with similar F1 but a 0.67-point lower Judge score. Applied directly to native PropMem, Mem0, and SimpleMem answers, the editing interface raises F1 by 3.79, 6.44, and 0.42 points. It outperforms same-gate rewriting and saves 49.85-83.72% of editing calls relative to ungated editing, which retains higher F1 in this adaptation. A source-by-editing factorial supports local editing for F1 but leaves an independent quality benefit from source obligations unresolved. Transfer to 451 LongMemEval-V2 Small questions also favors revision efficiency, while an external PerLTQA dialogue subset reveals a completeness gap. These results support selective editing as a reusable quality-cost operating point.]]></description>
      <pubDate>Thu, 19 Nov 2026 08:34:26 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.25721</guid>
    </item>
    <item>
      <title>EgoCaptioner: Towards Comprehensive and Fine-Grained Egocentric Video Captioning</title>
      <link>https://tldr.takara.ai/p/2611.26380</link>
      <description><![CDATA[Egocentric video captioning requires fine-grained detail and accurate timing, yet current multimodal large language models (MLLMs) often miss brief actions, confuse their order, or misinterpret the wearer's motion. Existing caption formats also offer limited support for combining detailed narration with precise timing. We present EgoCaptioner, which pairs a video-level overview of the setting, participants, and the wearer's overall activity with timed steps whose narration adapts to varying perceptual demands and marks key moments at specific timestamps. We develop a coarse-to-fine annotation pipeline that establishes event structure before expanding step descriptions, yielding 50,000 training examples from eight egocentric video sources. Complementary rewards assess factual accuracy and coverage, measure step-level temporal alignment, and penalize fragmented or redundant narration, guiding caption refinement through Group Relative Policy Optimization (GRPO). To improve the reliability of factual supervision, we supplement caption references with pose-derived motion facts when sensor data are available. Across five egocentric benchmarks, EgoCaptioner achieves leading performance among evaluated open-source models in caption quality, caption-conditioned question answering, and temporal alignment. It also surpasses Gemini 3.1 Pro on PhysCapBench and all evaluated proprietary baselines on Minerva-Ego under the caption-to-QA protocol. Ablations support the complementary roles of factual and temporal rewards and show that coarse-to-fine annotation provides more effective supervision than single-pass annotation.]]></description>
      <pubDate>Thu, 19 Nov 2026 16:39:59 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.26380</guid>
    </item>
    <item>
      <title>Seer: Guiding Small Deep-Research Agents with Shared Decision Models</title>
      <link>https://tldr.takara.ai/p/2611.26423</link>
      <description><![CDATA[Deep-research agents repeatedly decide what to search, read, and verify, often using large cloud-hosted language models for action proposal. However, privacy and local control motivate running agents on users' devices, where limited memory and compute favor smaller models and complicate on-device training. Yet inference-time guidance faces three obstacles: exploring alternatives through real execution adds tool calls, prompted self-evaluation is bounded by the agent's judgment, and model-specific guidance risks repeated adaptation costs. We introduce Seer, an inference-time framework providing local guidance that separates the agent's action proposal from learned consequence prediction and value estimation. From a shared corpus of real research trajectories, we train two models: a tool-response simulator on observed call--response pairs and a value estimator on histories labeled with discounted terminal correctness. The simulator and estimator are trained offline and reused across agents without further training. Moreover, Seer executes only the selected action and adds only real tool responses to the committed history, avoiding tool calls on discarded candidates at additional inference cost. Across multiple deep-research benchmarks, Seer substantially improves agent performance for different language models. For instance, GAIA accuracy rises from 13.6% to 41.7% for Qwen3.5-4B and from 22.3% to 50.5% for Qwen3.5-9B. Furthermore, our analyses suggest that broader candidate coverage can compensate for imperfect simulation, while prediction and scoring errors limit the benefits of wider imagined search. These findings highlight the potential of reusable models for prediction and evaluation to strengthen small research agents through extra local computation without the real execution of the discarded candidates.]]></description>
      <pubDate>Thu, 19 Nov 2026 17:03:24 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.26423</guid>
    </item>
    <item>
      <title>Client Exits - A Way to Increase LLM Transparency on the Client Side</title>
      <link>https://tldr.takara.ai/p/2611.26309</link>
      <description><![CDATA[Textual recommendations generated by large language models (LLMs) can be difficult to inspect. This paper introduces \emph{Client Exits}, a lightweight mechanism for enriching LLM responses while separating semantic intent from client-specific presentation. The model inserts predefined, machine-interpretable markers into its output, and the receiving client decides whether and how to realize the corresponding enrichments. This allows supplementary information to be presented as graphics, text, speech, or other modalities without prescribing a concrete interface. We implement a prototype combining a finetuned Qwen model with a modified LibreChat client. Using the Question--Option--Criteria (QOC) approach including three XML-based client exits. We evaluate two QOC-oriented model variants trained on essentially the same examples, with client-exit annotations added for one variant. Across 100 test questions, the client-exit model includes all three markers in every response. An LLM judge, informed about the mechanism without being instructed to favor it, prefers this model's answers in 66\% of the comparisons. These results demonstrate successful marker generation on the evaluated test set and support the feasibility of client-controlled enrichment.]]></description>
      <pubDate>Thu, 19 Nov 2026 15:58:09 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.26309</guid>
    </item>
    <item>
      <title>Beyond Error Mitigation: Using Boosting for Noisy Quantum Machine Learning</title>
      <link>https://tldr.takara.ai/p/2611.25752</link>
      <description><![CDATA[Quantum error mitigation (QEM) can recover information about ideal quantum computations from noisy circuit executions, but may require a large sampling overhead. Fundamental lower bounds show that recovering noiseless observables without problem-specific information can require superpolynomially many samples, with exponential overhead for broad classes of noisy circuits. However, problem-specific information can reduce the number of samples needed to obtain the required answer. We investigate this possibility in supervised quantum machine learning, where labeled examples provide information about the function to be learned. Binary classification provides such a setting, since correct prediction depends on the sign and margin of the classifier output rather than on accurate reconstruction of every noiseless expectation value. We develop a theoretical and empirical framework for comparing boosting of noisy parameterized quantum circuits with probabilistic error cancellation (PEC), requiring both methods to reach the same learning target. For the specified circuit architectures under Pauli noise, we establish a common hypothesis-class complexity bound: matching empirical margin loss gives the same upper bound on expected classification risk. We derive resource criteria and evaluate AdaBoostR and Real AdaBoost in finite-shot teacher-student simulations across circuit depth and physical noise strength. Under the same classification-accuracy target, boosting requires fewer circuit executions as depth and noise increase. Boosting also achieves higher test accuracy than the corresponding reference. For the supervised-learning problems studied, these results demonstrate a task-dependent resource advantage: exploiting information retained in noisy outputs can avoid the cost of ideal-output reconstruction without sacrificing observed classification performance.]]></description>
      <pubDate>Thu, 19 Nov 2026 08:56:10 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.25752</guid>
    </item>
    <item>
      <title>Calibrate in Context: Efficient Data-Free Quantization from Small Synthetic Pools</title>
      <link>https://tldr.takara.ai/p/2611.25711</link>
      <description><![CDATA[We enable accurate low-bit recovery from small synthetic pools by improving both image synthesis and student preparation. A grid with lower local response error can incur greater error in the student's block context. Functional Grid Calibration (FGC) selects activation scales and integer offsets against full-precision responses constructed from the student's current block input. With weights fixed during calibration, FGC evaluates candidates only at their consuming operators, bringing block context to calibration at nearly local-scoring cost. For DeiT and ViT, progressive inversion updates the same images through increasingly deep teacher prefixes. With 1,000 images, the complete pipeline exceeds MimiQ's reported accuracy in all 18 W4A4/W5A5 settings across nine backbones. With 256 images, DeiT-S and ViT-S reach 68.20% and 66.90% ImageNet top-1 accuracy at W4A4 in 26-27 minutes on one RTX 5090. The pipeline outperforms strengthened 1K-image MimiQ by 13.03 and 24.36 percentage points at lower preparation cost and the same recovery-update count. The same low-budget checkpoints retain their advantage on ImageNetV2.]]></description>
      <pubDate>Thu, 19 Nov 2026 08:24:19 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.25711</guid>
    </item>
    <item>
      <title>How to Effectively Manage the Energy in Residential Systems: Development and Comparison of Methods</title>
      <link>https://tldr.takara.ai/p/2611.26412</link>
      <description><![CDATA[The transition from fossil fuels presents significant challenges in energy management. To ensure a sustainable energy future, it is crucial to reduce energy consumption when possible and to synchronize electricity consumption and storage with the production from sources with low greenhouse gas emissions. This thesis explores a variety of energy management methods used to manage batteries, heat pumps, and other energy-related systems. The thesis presents three simulation cases alongside one real-world experiment to compare energy management done through simple rules, optimization, and machine learning methods. Furthermore, this thesis introduces TreeC, a novel machine learning-based method that learns understandable yet efficient energy management strategies, applicable to many cases.
  The three key contributions of this thesis can be summarized as follows: 1) The developed TreeC method demonstrates competitive performance compared to a variety of energy management approaches, such as model predictive control and reinforcement learning, across a diverse set of energy management problems. The understandability of TreeC's energy management strategies greatly improves the transparency and trustworthiness of machine learning-based energy management. 2) The real-world experiment highlights the need to validate energy management methods on actual hardware and not solely on simulations. Both TreeC and model predictive control performed worse than in simulations, due to, respectively, insufficiently representative training data and the real-time input data not being reliable enough. 3) Energy management algorithms reduce costs in electrified Belgian dwellings, particularly in poorly insulated dwellings, electric vehicle chargers, and photovoltaic installations. However, on average, these cost savings do not yet outweigh the current costs and effort required to implement such algorithms.]]></description>
      <pubDate>Thu, 19 Nov 2026 16:59:58 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.26412</guid>
    </item>
    <item>
      <title>Generalist Meets Specialists: An Evolving Multimodal Agent with Persistent Tool Orchestration for Traffic Scenario Understanding</title>
      <link>https://tldr.takara.ai/p/2611.25868</link>
      <description><![CDATA[Traffic scenario understanding requires combining fine-grained visual evidence with spatial and temporal reasoning. Tool-augmented vision-language models (VLMs) can use specialized perception tools, but effective collaboration faces two challenges: terminal rewards provide limited guidance for intermediate decisions and error recovery, while repeated planning adds unnecessary cost across similar tasks. Echoing the fast-slow dichotomy of human cognition, we propose StarTV-Agent, which combines policy learning with experience evolution to support adaptive reasoning when needed and efficient reuse of prior experience when possible. During training, Role-Sensitive Process Optimization (RSPO) combines process advantages with terminal advantages, while Weakness-Driven Demonstration Learning (WDL) provides validated teacher demonstrations targeting the student's weak questions. During inference, Lightweight Adaptive Workflow Reuse (LAWR) enables a fast path by selecting suitable workflows and binding task-specific arguments and resource references, while a slow ReAct-based path handles tasks without suitable workflows. We further design label-free feedback based on execution completion, workflow adoption, and cost savings to guide workflow reuse and memory updates. Across three challenging benchmarks, StarTV-Agent achieves up to 8.0% relative performance improvement, while reducing token consumption by up to 52.1%.]]></description>
      <pubDate>Thu, 19 Nov 2026 10:20:37 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.25868</guid>
    </item>
    <item>
      <title>Demonstration-Free Skill Composition for Navigation and Manipulation</title>
      <link>https://tldr.takara.ai/p/2611.25462</link>
      <description><![CDATA[Hierarchical reinforcement learning enables sequential tasks to be represented as compositions of separately trained skills. This paper evaluates demonstration-free hierarchical pipelines for MiniGrid navigation and robotic manipulation, comparing their training requirements and task performance with flat reinforcement learning baselines. The evaluation concerns complete pipelines with task-specific observation, reward and control structures. In navigation, two parameterised skills are reused across unseen FourRooms layouts under manually specified high-level plans. In manipulation, reaching, grasp alignment and lifting are trained separately, with downstream skills initialised from states collected during predecessor rollouts. Across unseen navigation layouts, the hierarchical pipeline substantially outperforms the evaluated flat DQN baselines while requiring fewer training interactions. In manipulation, the hierarchical chain requires fewer training steps than flat SAC and exhibits lower observed variation across seeds. The frozen Pick and Lift skills are further reused for stacking by training an additional Place skill.]]></description>
      <pubDate>Thu, 19 Nov 2026 03:47:02 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.25462</guid>
    </item>
    <item>
      <title>SegAnyFruit: Learning Generalizable and Edge Deployable Fruit Instance Segmentation at Scale</title>
      <link>https://tldr.takara.ai/p/2611.25475</link>
      <description><![CDATA[Deploying fruit instance segmentation in real agricultural settings requires generalizing across unseen species, orchards, and imaging conditions under strict onboard compute and annotation budgets. We introduce SegAnyFruit (SAF), a data construction and progressive training framework that distills foundation-model priors into lightweight, prompt-free segmenters. SAF combines generic mask pre-training, broad mid-training on automatically annotated web imagery from our curated WebFruits dataset, and human-verified agricultural refinement post-training on PaperFruits, together spanning 733K images and 9.66M whole-fruit masks across 165 fruit types. Across seven benchmarks, zero-shot SAF-RF-DETR-XL reaches 69.06 average mask AP50:95 (+4.76 over SAM 3) while remaining deployable on edge devices. SAF initialization also boosts few-shot to full-shot adaptation across five agricultural domains while mitigating catastrophic forgetting. Finally, we systematically analyze training stages, data scale, model capacity, and generative augmentation, and validate our pipeline's transferability to underwater, cell, and human segmentation. We will release our datasets, code, and SAF models.]]></description>
      <pubDate>Thu, 19 Nov 2026 04:03:32 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.25475</guid>
    </item>
    <item>
      <title>Mesh-Free Keypoint Learning for Category-Level 6D Object Pose Estimation</title>
      <link>https://tldr.takara.ai/p/2611.25513</link>
      <description><![CDATA[Category-level 6D pose estimation relates partial RGB-D observations of unseen instances to a canonical frame shared across a category. Object-aware reconstruction helps adaptive keypoints retain the geometry of the visible surface. Common reconstruction targets select observed points by their distance to a registered instance surface. We present MORSEPose, which derives this support directly from standard pose and metric-size annotations. We map observed 3D points to normalized object coordinates and retain those inside a tolerancedilated pose-and-size envelope. This construction is consistent across camera frames and metric scales. When the annotated extent bounds the object surface, the envelope conservatively contains the support of a conventional surface-distance selector. We integrate this target with multi-level RGB-D features, adaptive keypoints, and learnable category context. Experiments on REAL275 and HouseCat6D show that the envelope closely matches surface-based supervision at both point and task levels, remains stable under boundary perturbations and viewpoint changes, and supports competitive category-level pose accuracy across benchmarks.]]></description>
      <pubDate>Thu, 19 Nov 2026 05:03:40 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.25513</guid>
    </item>
    <item>
      <title>Efficient and Robust Drafting with Multi-Token Prediction</title>
      <link>https://tldr.takara.ai/p/2611.25813</link>
      <description><![CDATA[Speculative decoding has emerged as a critical technique for accelerating Large Language Model (LLM) inference by decoupling generation into a drafting phase and a parallel verification phase. A crucial challenge is therefore designing a drafter that is both lightweight and accurate enough to approximate the predictions of the target model. We present a highly optimized Multi-Token Prediction (MTP) architecture that extends existing backbone LLMs to efficiently produce high-quality drafts. Built around a single shared drafting head, our approach features three core architectural optimizations: target activation injection to anchor semantic planning, zero-copy KV-cache cross-attention to bypass prefill compute and reduce memory overhead, and an efficient clustered vocabulary projection to reduce the cost of full-vocabulary decoding. Testing these improvements with the Gemma 4 model family, our G-MTP shows out-of-the-box speedups for both large server-side setups (up to 3.1$\times$) and smaller models targeting memory-constrained on-device environments (up to 2.3$\times$ on H100 and up to 2.2$\times$ on a smartphone GPU). In addition, we present parameter-efficient fine-tuning recipes for G-MTP that keep the drafter aligned with the backbone LLM under domain shift, requiring up to 1.83$\times$ fewer target forward passes than tuning only the backbone.]]></description>
      <pubDate>Thu, 19 Nov 2026 09:33:39 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.25813</guid>
    </item>
    <item>
      <title>Online Mistake Detection in Instruction-Conditioned Procedural Video</title>
      <link>https://tldr.takara.ai/p/2611.25567</link>
      <description><![CDATA[An assistant monitoring a person performing a procedure is most useful when it can detect mistakes while there is still time to recover. This requires judging actions against task instructions as the video unfolds, identifying the attempted step, locating it in time, and assessing its execution. Existing online detectors do not use the instruction text, while instruction-conditioned methods operate offline. In this paper, we study online, instruction-conditioned mistake detection in egocentric videos. Using only observed video, task instructions, and prior predictions, our model identifies and localizes completed action attempts and classifies them as correct, modified, slip, corrections, or additions; omitted steps are determined after the video ends. We compare a modular pipeline, with separate step identification, temporal localization, and execution verification, to a joint model that predicts all outputs in a single step. To support supervised learning from mistakes, we repartition EgoPER to include erroneous videos in all splits. The approaches achieve similar step-segmentation F1@0.5 (93.6% and 92.1%), while the joint model achieves higher binary mistake-detection F1@0.5 (84.1% vs. 80.7%), with its largest gain on slips (56.4% vs. 37.5%). Component-level analysis suggests that error propagation between stages contributes to the modular approach's lower detection score. On test videos containing mistakes, both designs outperform released AEM and ESTANet checkpoints under our evaluation protocol.]]></description>
      <pubDate>Thu, 19 Nov 2026 06:09:24 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.25567</guid>
    </item>
    <item>
      <title>Statistical Inference with Ill-Conditioned Random Weight Neural Network Sieves</title>
      <link>https://tldr.takara.ai/p/2611.26410</link>
      <description><![CDATA[Random weight neural networks (RWNNs) are shallow networks whose hidden-layer parameters are randomly generated and then kept fixed, while only the output coefficients are estimated. Conditional on the sampled parameters, RWNN regression is a linear sieve problem with random approximation space. We show that such random sieves cannot remain uniformly well-conditioned as their dimension grows. For parametrized feature families satisfying polynomial covering and Lipschitz conditions, the eigenvalues of the population covariance matrix are proven to have polynomial upper bounds determined by the covering geometry of the parameter space, leading to a vanishing smallest eigenvalue. We develop approximation, estimation, and inference theory that explicitly treats the ill-conditioning. Under a $β$-Hölder source condition, we obtain both faster-than-Monte-Carlo $L^2$ approximation rates and new $L^\infty$ approximation bounds based on stratified sampling. We then derive improved consistency guarantees for ridge-regularized least squares driven by ridge effective dimension, spectral, and leverage-based stability conditions. For inference, we establish asymptotic linearity and Gaussian limits for linear functional directions while allowing the smallest eigenvalue of the population covariance matrix to vanish. Nondegeneracy is directional: we characterize it with a ridge-normalized Rayleigh quotient and extend it to smooth functionals via Riesz representation. Consistency of ridge-adjusted variance estimators is proved, yielding feasible studentized inference. Numerical experiments illustrate the deterioration of random-feature conditioning and the finite-sample behavior of the resulting estimation and inference procedures.]]></description>
      <pubDate>Thu, 19 Nov 2026 16:59:34 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.26410</guid>
    </item>
    <item>
      <title>MorphTwin: Enriching Deformable Material Responses from Videos for Morphable and Interactive Physical Twins</title>
      <link>https://tldr.takara.ai/p/2611.25963</link>
      <description><![CDATA[Reconstructing interactive physical twins from video requires accurately capturing how objects deform while maintaining efficient simulation. We observe that reproducing motion in a video does not guarantee correct responses to new forces when the physical representation restricts how an object can deform. In spring-based twins, coupling between volume and shape responses creates such a limitation. We present MorphTwin, which enriches local deformation modes to capture and independently control these responses while retaining standard spring forces and explicit simulation. To support interpretable material editing and transfer, we use offline calibration to map shared relative response changes to stiffness updates on each reconstructed graph. Under identical inverse settings on synthetic scenes, MorphTwin reduces the mean absolute error of Poisson's ratio by 75.8% and the mean relative error of Young's modulus by 22.1% compared to PhysTwin. We introduce an egocentric stereo and RGB-D benchmark covering interactions with objects of varying stiffness. Real-world experiments demonstrate improved reconstruction accuracy and up to 14.4% lower future tracking error than PhysTwin at interactive speeds.]]></description>
      <pubDate>Thu, 19 Nov 2026 11:51:16 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.25963</guid>
    </item>
    <item>
      <title>Reward misalignment as an opportunity for materials discovery</title>
      <link>https://tldr.takara.ai/p/2611.26186</link>
      <description><![CDATA[Reward misalignment is usually regarded as a failure of artificial intelligence. In materials discovery with generative AI models, however, disagreement between predefined rewards and expert judgment may reveal materials that the rewards overlook. We investigate this possibility using electronic flat bands as a prime example. Experts mostly agree with a numerical bandwidth score, but their disagreements are systematic and cannot be explained by an estimated 5% rate of mistaken labels. We learn the experts' preferences from pairwise comparisons and use them to guide a crystal diffusion model. The learned preferences and numerical score agree on the flattest bands but diverge elsewhere. For example, the preferences favor KCo2As2 and CoGa, reported compounds with partially flat bands that the score would discard. The framework also extends to synthesizability and phonon properties. Our findings suggest that reward misalignment can expose opportunities for materials discovery beyond what fixed rewards are designed to find.]]></description>
      <pubDate>Thu, 19 Nov 2026 14:36:15 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.26186</guid>
    </item>
    <item>
      <title>Dynamics of Protected State Evolution</title>
      <link>https://tldr.takara.ai/p/2611.25541</link>
      <description><![CDATA[States with identical protected observations can admit different future interventions. Each transition specifies its successor, response increment, required initial reserve and net consumption. Composing these records in execution order distinguishes protocols with equal net consumption and different initial reserve requirements. At a fixed state, a weighted Gram operator determines the linear response ellipsoid and a quotient identifies directions attainable only through cooperation between blocks. Nonlinear systems with identical complete affine data can have incomparable finite-budget response sets. In regular charts, the first correction to quadratic minimum cost is a covariant cubic form involving the third derivative of cost and the Hessians of protection and response. At singular gates, sequential feasibility depends on the peak reserve along the path. A history-weighted operator detects responses outside the old channel's range closure. Controlled congruence characterizes exact reductions of deterministic labelled dynamics, and a two-cut value identity separates the effect of paid verification from that of subsequent action selection.
  On Qwen2.5-7B Base, two projected policies preserved 2,048 reference top tokens under a per-position KL ceiling of $2\times10^{-5}$. Over 200 arrivals they recovered 391 of 396 initial rejections. Each yielded over 1,200 times the held-out cross-entropy reduction of historical-channel-only AdamW. Under an expanded 2,560-position contract, fixed continuation and neutral renewal began on the same protected logit fiber in distinct structural strata. Both admitted all 200 subsequent writes. Fixed continuation yielded 1.74 times the held-out loss reduction per unit squared write cost. Renewal had a smaller KL excess on the first candidate tested at every arrival and used fewer full verification scans.]]></description>
      <pubDate>Thu, 19 Nov 2026 05:44:57 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.25541</guid>
    </item>
    <item>
      <title>Harnessing Medical Visual-Language Models with Agentic Evidence Management</title>
      <link>https://tldr.takara.ai/p/2611.25840</link>
      <description><![CDATA[Medical vision-language agents increasingly rely on segmentation tools, and recent work has focused on acquiring the right mask, often treating a good mask as the end point of tool use. But do these agents use the resulting evidence effectively? To answer this question, we need a benchmark that separates evidence use from segmentation quality, spans multiple medical image types, and allows the same mask to be presented in different ways. Existing medical VQA datasets do not jointly provide these properties. We therefore construct Mask-Evidence MedVQA, a 2.5K-example benchmark spanning four image types, where each question is paired with an answer-relevant ground-truth mask and multiple possible renderings. Our study reveals two limitations of direct tool use. First, evidence presentation matters. Different renderings of the same fixed mask can substantially change model behavior. Second, masks introduce an evidence trade-off. Highlighting a relevant region can expose diagnostic evidence the model previously missed, but the added visual cue can also induce incorrect associations. As a result, even a ground-truth mask can correct some answers while turning initially correct ones into incorrect ones. Together, these findings show that acquiring the right mask is not enough. An agent must also decide how to present the evidence and whether it should change the answer. We therefore introduce TUNE (Tool Use is Not Enough), an evidence-management harness that explicitly manages what evidence to acquire, how to present it, and whether to accept the resulting answer update. Across three public medical VQA benchmarks, evidence management improves the fine-tuned Qwen3.5-9B backbone's mean accuracy by 4.0 percentage points. Full TUNE improves over direct tool use by 9.1 points on average across seven benchmarks and achieves competitive results against existing medical VLMs and agents.]]></description>
      <pubDate>Thu, 19 Nov 2026 09:55:33 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.25840</guid>
    </item>
    <item>
      <title>The Stability Gap in Continual Reinforcement Learning</title>
      <link>https://tldr.takara.ai/p/2611.26252</link>
      <description><![CDATA[Continual reinforcement learning (CRL) aims to enable agents to adapt to changing tasks while preserving previously acquired capabilities. Evaluation often relies on final aggregate metrics that can hide temporary performance degradation during adaptation. This work introduces a simple CRL benchmark that varies observation representations while preserving environment dynamics and rewards, supporting proof-of-concept demonstrations, inexpensive experimentation, and easily reproducible sanity checks. Using this benchmark, we demonstrate the stability gap in CRL: a substantial drop in prior-task performance after the introduction of new tasks, followed by partial or complete recovery. This pattern occurs despite the agent having complete access to past tasks.]]></description>
      <pubDate>Thu, 19 Nov 2026 15:26:41 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.26252</guid>
    </item>
    <item>
      <title>When Should Deployed BCI Decoders Adapt? A Resource-Priced Ladder for Frozen EEG Trunks</title>
      <link>https://tldr.takara.ai/p/2611.25315</link>
      <description><![CDATA[When should a deployed MI decoder spend adaptation resources-not only how to adapt under shift? We study a thin control layer outside a frozen encoder: a resource ladder (factory -> Bayesian head -> affine map, with full fine-tune as an escape hatch) governed by regime-aware gating under a utility that prices accuracy, fit time, and label cost. The head is Bayesian in the conjugate sense: a Normal-Inverse-Wishart prior over the class conditional Gaussians in embedding space, updated in closed form (no gradients) as labelled trials arrive. Full estimators, constants, and the reanalysis code are in the companion paper and the ancillary bundle of this submission. Thesis: adaptation should be gated and capped, not always-on and not blindly fine-tuned. Evidence: (i) utility-max gating over-adapts on clean data that already carries session drift; a clean-referenced gate restores low false-entry rates (e.g. dropping the adaptation rate from 0.56 to 0.11 on field-like sessions, with the cascade staying on the factory decoder for ~87% of clean cells). (ii) Under severe synthetic stress, a lightweight LDA head matches or exceeds the frozen trunk's accuracy across five encoder architectures (n=155 subject x trunk cells, pooled across shared subjects), at ~23x lower fit time than full warm-start fine-tuning. Always-on baselines favor capping the default below full fine-tune. (iii) In two no-perturbation BNCI session checks the factory model transferred without intervention (no reliable head gain), so the control layer's value under natural drift is cost-efficiency and clean-safety rather than demonstrated accuracy gains. (iv) Adaptation severity does not follow a universal ordinal ladder across datasets, highlighting the need for dynamic gating over fixed adaptation policies.]]></description>
      <pubDate>Thu, 19 Nov 2026 00:34:41 +0000</pubDate>
      <guid isPermaLink="true">https://tldr.takara.ai/p/2611.25315</guid>
    </item>
  </channel>
</rss>