PyTorchKR Papers

최신 논문

총 4,976편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 카테고리 구독

arXiv:2611.16174v2cs.LG

JudgeMix: Judge-Guided Mixed-Rollout Data Synthesis for LLM Agents

Supervised fine-tuning (SFT) on expert trajectories is a standard paradigm for training LLM agents. However, it suffers from a well-known training-inference distribution mismatch, calling for target-aware data synthesis strategies that ground the training data in the target model's native state distribution. To bridge this gap, emerging approaches incorporate the expert either to guide mixed rollouts alongside the target model, or to provide dense step-wise supervision. Nonetheless, existing methods apply expert assistance blind to the target model's inherent behavior: they either disrupt the rollout using sub-optimal random intervention heuristics, or they enforce indiscriminate uniform supervision that unnecessarily overwrites the model's already-correct actions. We introduce JudgeMix, a judge-guided data synthesis framework that dynamically interleaves the target model and expert based on step-wise semantic evaluations. JudgeMix queries expert guidance precisely when the target model struggles, ensuring high trajectory success rates while keeping the state distribution aligned with the target model. It additionally uses the judge evaluation to reweight expert supervision on different steps, preserving model's inherent valid behavior while correcting its errors. On two agentic tool-use benchmarks, AppWorld and AutomationBench, JudgeMix demonstrates consistent superior performance to existing methods, with improvements up to 10 pass@1 points.

arXiv:2611.15592v2cs.AI

Small Heads, Strong Decisions: A Controlled Study of Typed-Decision Models

In September 2026, TypeSafe announced Jev as the first model in its System One category: unstructured state goes in, and typed, calibrated decisions that software can use directly come out. Jev's architecture and training data are unpublished, but many open typed-decision models have followed, including Nimble, the Decision 1.0-3.0 family and Clef. They answer structured questions about a piece of text, such as a yes/no check, a choice among labelled options or a graded score, by reading the text with a language model and scoring the options with a small readout. Each pairs a different readout with a different backbone, training mix and recipe, so it is unclear which design choices matter. We re-implement five readouts and train all of them under one protocol: the same backbone, the same 15 public English training sources, the same recipe with a per-head learning-rate sweep, and the same seed. Our experiments demonstrate that: The complexity of the head does not affect accuracy: on Qwen3-0.6B every readout reaches 81.3-82.0% in-domain accuracy, and a 0.5M-parameter cosine head, which scores each option by its cosine similarity to the question, matches the 27M-parameter joint schema head adapted from Clef in-domain at half its inference time. Reading all of a text's questions from one prompt also matches one prompt per question in accuracy, with 1.6 times faster inference and 2.1-2.5 times faster training on one H100. The backbone's size matters more than either choice: moving from Qwen3-0.6B to Qwen3-4B adds 4 points in-domain and 11 points on held-out sources, while the base, instruct and thinking checkpoints of the same model tie. We recommend spending effort on the backbone and keeping the head minimal.

arXiv:2611.15360v2cs.CV

CARE-Drive: Coupled Alternating Refinement for Ego-trajectories

Vision-language-action (VLA) models bring multimodal understanding and world knowledge to end-to-end autonomous driving. To further improve the quality of predicted trajectories, existing studies introduce candidate refinement, residual denoising, and token correction. However, these methods typically focus on unified refinement of the entire trajectory, paying limited attention to the distinct physical quantities and their interactions. We further investigate physical specialization and coordination in trajectory refinement, centering on position and heading for simpler and more intuitive correction modeling. To this end, we propose CARE-Drive (Coupled Alternating Refinement for Ego-trajectories), which decomposes trajectory refinement into waypoint residual correction and position-conditioned heading generation. Through imitation learning and a future-token auxiliary task, the backbone provides predictive representations and a base trajectory. The position expert predicts bounded metric waypoint residuals, while the heading expert reconstructs vehicle orientation from adjacent angular increments conditioned on the corrected path. With the backbone frozen, the two experts are optimized alternately through supervised initialization and reinforcement fine-tuning. This design simplifies refinement through clear physical specialization while preserving conditional coupling during prediction and optimization. On NAVSIM v1, CARE-Drive achieves 88.8 PDMS, improving by 2.4 points over the unrefined base trajectories generated by the same backbone. Ablation studies further support the effectiveness of specialized experts and staged training.

arXiv:2611.15315v2cs.AI

MIGRO: A Decoupled and Portable Long-Term Memory System for Language Models

Language models evolve, but the observations, revisions, and context accumulated through user interactions should remain usable. Each new model must be able to retrieve past evidence, track revisions, and add new observations. We present MIGRO, a decoupled long-term memory system that treats experience as a persistent state shared across successive language models. MIGRO follows a simple principle: preserve historical evidence exactly, organize it approximately, and decide what to retrieve only when the question is known. We call this asymmetric fidelity with deferred selectivity. A shared semantic index supports retrieval, while exact records retain observations and their historical relationships. Because models differ in vocabulary and tokenization, host-specific interfaces convert selected evidence into inputs the receiving model can use. Reversible encoding reduces archival storage while enabling exact recovery. The shared state and its access interfaces let successive hosts read and extend the accumulated experience. On matched LoCoMo states, MIGRO scores 7.3% higher and uses 99.3% less private memory than the strongest quality baseline. After handoff, continued writing retains 99.9% of full-history rebuilding quality by extending the inherited state. On million-token BEAM histories, MIGRO uses 57.5% less private storage on average than the most compact baseline. MIGRO combines high answer quality, continued memory growth across hosts, and low storage cost. Code is available at https://github.com/zOUMINGXI/MIGRO.

arXiv:2611.15262v2cs.LG

CastWorld: Benchmarking Context-Aware Agentic Time Series Forecasting Beyond Predictive Accuracy

Agentic time series forecasting extends conventional approaches beyond one-shot numerical prediction toward autonomous, context-aware interaction. However, existing benchmarks offer limited integrated support for diverse contextual information, forecasting-specific adaptive interactions, and systematic evaluation of forecasting agents. In this paper, we introduce CastWorld, an agentic benchmark for context-aware time series forecasting. First, its context-rich dataset suite spans 8 domains and 25 real-world datasets, covering external variables, events, and historical cases under a unified context specification. Second, a lightweight, modular forecasting-native harness provides 49 core tools and a state-adaptive loop that tracks forecast versions and enables iterative evidence gathering and forecast revision. Intermediate forecasts can expose information gaps, while newly acquired evidence guides subsequent revisions and final version selection. Third, model-level evaluation assesses forecast quality, process execution, behavioral trustworthiness, and resource efficiency through decision trajectories and forecast-version histories. Also, systems-level evaluation compares predictive accuracy across forecasting systems. Experiments reveal distinct decision behaviors among agents with comparable accuracy. Additional interactions do not necessarily improve forecasts, and agents may fail to select their best previously generated predictions. Allowing evidence acquisition during forecast refinement improves task-aware forecast scores across all three evaluated models. CastWorld provides a reusable foundation for evaluating agentic time series forecasting under controlled conditions.

arXiv:2611.13971v2eess.IV

TransGHOST: Real-Time, Synthesis-Free Reconstruction of Instrument-Occluded Regions in Ophthalmic Microsurgery

Intraoperative scene perception in ophthalmic microsurgery is limited by instrument occlusion: when a tool enters the high-magnification field, it conceals tissue from both the surgeon and the perception components of robot-assisted systems that process the video. Generative inpainting can remove the occluder but replaces the missing image content with synthesis, which is unsuitable when visual provenance matters. Many video inpainters also depend on future frames and cannot operate online. We present TransGHOST, a causal reference-guided reconstruction pipeline whose non-generative safe mode operates in real time. Every restored value is obtained by resampling a recorded frame using geometric transforms fitted only to visible regions. The system provides per-pixel confidence and provenance and marks unsupported pixels as abstentions. We evaluate TransGHOST on a new veterinary clinical corpus of canine cataract surgery with pixel-level instrument and anatomy labels. Only the occluded region is scored, and each scoring target is verified to differ from the reference supplied to the system. Safe mode reaches 31.89 dB masked PSNR and 0.915 masked SSIM at 51.7 ms median and 85.5 ms 95th-percentile latency. On a pre-declared held-out cohort using instrument cutouts excluded from segmentation training, it reaches 29.83 dB masked PSNR, while matched-mask evaluation on validation shows a 3.45 dB advantage over LaMa. Region-aware reference selection improves masked PSNR by 3.60 dB over recency when candidate references can themselves be occluded. Reference-age analysis further defines the operating interval for the brief, repeated occlusions targeted by TransGHOST.

arXiv:2611.13710v2cs.AI

AutoManifold: Agentic Design of Data Visualisation Algorithms via Manifold Embedding

Manifold embedding for data visualisation is dominated by a small set of hand-designed state-of-the-art (SOTA) algorithms. Their design targets are fixed at development time, resulting in inflexible algorithmic behaviour that cannot be readily steered towards user-specified structural priorities. Hand-designing new embedding algorithms to address user priority requires highly specialised expertise and is time consuming. To address autonomous algorithm design tailored to user preference, we introduce an agentic algorithm generation pipeline AutoManifold. It composes new manifold embedding algorithms from a constrained vocabulary of affinity, cost, and optimisation primitives, supported by multi-agent large-language-model (LLM) orchestration. AutoManifold conditions every stage of its design on user-specified structural preservation preferences, and iteratively refines the algorithm configuration through an LLM-guided, metric-grounded iterative loop. We compare the generated algorithms against the three strongest and most frequently used SOTA methods (t-SNE, UMAP, and PaCMAP) on real-world datasets spanning different difficulty regimes under identical evaluation infrastructure. The results demonstrate that LLM-driven, priority-conditioned algorithm synthesis can move beyond hyperparameter tuning, producing genuinely new algorithms that outperform traditional hand-designed methods in their ability to steer towards user-specified properties without compromising the intrinsic structure in the original data. The code of AutoManifold is available at https://github.com/buraksusam/AutoManifold_NeurIPS2026.

arXiv:2611.13442v2cs.DC

ForeSpark: Joint Draft and Verification Budget Scheduling for Parallel Speculative Decoding

Parallel drafters accelerate LLM decoding by proposing a whole block of draft tokens in a single forward pass, making them attractive for LLM serving. Recent systems optimize goodput (i.e., accepted tokens per second) because rejected drafts steal compute from other requests. However, these systems control only how many tokens are verified, so every request still drafts a full block. At high serving load, we find that drafting consumes up to 40% of step time, of which a majority is discarded because it is never verified. Yet, draft-length control from autoregressive drafters does not carry over, because a parallel drafter commits to its length before its single pass. Moreover, draft and verification lengths are coupled across the batch, so they must be selected jointly for every request while keeping the decision computationally cheap at every step. In this paper, we propose work-conserving speculation (WCS), a design that jointly allocates draft and verification budgets per request in LLM serving, and present ForeSpark as its prototype. ForeSpark realizes WCS with (i) a context-aware pre-draft estimator that predicts acceptance for every candidate length from the target model's context features, (ii) a lightweight joint draft-verify scheduler that selects both lengths through a global sort, and (iii) a post-draft budget distribution that spends verification on the most promising tokens. Implemented on vLLM, ForeSpark improves throughput by up to 1.77x over the state-of-the-art DSpark and by 2.49x over autoregressive decoding at a matched per-request generation rate, without changing the output distribution. It also tolerates higher load, sustaining up to 1.51x the request rate of DSpark under latency service level objectives (SLOs).

arXiv:2611.13331v2cs.LG

Stochastic Projections and Rank-Aware Learning Rates for Federated LoRA

Fine-tuning large-scale foundation models across resource-constrained, distributed clients presents significant computational and communication challenges. While merging Low-Rank Adaptation (LoRA) with Federated Learning (FL) enables rank-based flexibility, the interdependence of the adaptation matrices ($\mathbf{A}$ and $\mathbf{B}$) often complicates optimization. In this work, we theoretically demonstrate that LoRA fine-tuning with Gaussian initialization and a frozen $\mathbf{A}$ matrix acts as a stochastic approximation of full fine-tuning, where batch gradients are subject to a random, rank-dependent perturbation. Motivated by this insight, we propose freezing and periodically resampling $\mathbf{A}$ matrices at the client level. This strategy decouples the $\mathbf{A}$ and $\mathbf{B}$ gradients, diminishes LoRA low-rank constraints and reduces both computational and uplink communication overheads by 50\%, allowing for the reallocation of resources toward higher adaptation ranks. Furthermore, we show that in this frozen-$\mathbf{A}$ regime, the adaptation rank $r$ directly governs the signal-to-noise ratio (SNR) of the perturbed gradients. Specifically, we find that the SNR increases linearly with $r$, implying that clients' learning rates must be dynamically scaled with their adaptation ranks for higher utility. Building on these foundations, we introduce SPRA-LoRA (Stochastic Projection Rank-Aware LoRA), a federated rank-aware fine-tuning strategy that utilizes frozen, resampled $\mathbf{A}$ matrices and rank-dependent learning rates for clients. We also prove its convergence with heterogeneous client resources. Our experiments across benchmark datasets demonstrate that SPRA-LoRA consistently outperforms state-of-the-art baselines with half the overhead on clients.

arXiv:2611.13223v3cs.CL

Deletion Sensitivity Is a Property of the Model: A Three-Condition Measurement over 72 Valid Experimental Cells

Reasoning models write a trace at every turn of a multi-turn tool session, and later turns read it. Upstream providers hold different retention constraints, so a request forwarded across issuers arrives with its history deleted, and its task cost remains unmeasured. This paper reports a direct, blind, multi-seed measurement of full history deletion in multi-turn tool calling, through three rendering conditions, a calibrated set of 72 valid experimental cells, a multi-seed protocol, and the official checker. The same deletion separates the models by direction and by magnitude, one class losing tens of points, another gaining, and a third unchanged, with the endpoint ordering reproduced across seeds and task subsets modulating the magnitude and, near zero, the sign. A re-dosed model preserves its success across the full deletion range, so the reading belongs to the model rather than the dose. Deletion sensitivity is a model-level property under an operational definition, measured under a fixed task family, scorer, and protocol, and a retrained model is a new model that a two-cell screen, one retention run and one deletion run, re-reads for the cost of two runs. Deletion sensitivity is thus a measurable property of the model, and the reading must enter routing as prior information rather than persist as an invisible default cost.

arXiv:2611.13213v2cs.CV

Before Words, Beyond Speech: Evaluating Nonverbal Social Reasoning in Early Childhood

Children's earliest years are a period of intense social learning, during which meaning is built through gaze, gesture, and physical contact. Yet, this critical aspect of social interaction remains under-evaluated in multimodal video understanding, as existing benchmarks prioritize the conversation-driven interactions of adults. We address this gap with \bench ($\textbf{N}$aturalistic $\textbf{E}$arly-childhood $\textbf{S}$ocial in$\textbf{T}$eractions), the first child-centric benchmark for nonverbal social reasoning. \bench comprises \totalvids manually annotated clips ($\sim$12s) curated for diversity across naturalistic settings, cultural contexts, and developmental stages. Beyond a lack of data, targeted model improvement in this domain is hindered by coarse evaluation schemas that conflate basic perception with higher-order reasoning. To resolve this, \bench introduces a compositional schema that builds atomic interaction cues into progressive levels of social reasoning: contextual, behavioral, interpersonal, and open-ended descriptions. Extensive analysis of state-of-the-art Vision-Language Models (VLMs) reveals a stark performance collapse: despite strong contextual recognition, models exhibit substantial gaps in decoding precise behavioral and interpersonal cues. We trace these failures to specific biases, including poor consequence grounding and temporal sensitivity. Finally, we ground \bench in real-world applications by evaluating VLMs using domain adaptation strategies. Ultimately, \bench provides a rigorous testbed for advancing socially grounded artificial intelligence in early childhood. Project page: https://laubravo.github.io/nest_website/.

arXiv:2611.12799v2cs.LG

Agentic-OPD: Recovering Forgotten Agent Capabilities via On-Policy Distillation for Agentic Tasks

Reinforcement learning is now the mainstream paradigm for training LLMs on agentic tasks, yet standard practice keeps only the best-validation checkpoint and discards the rest. We show that this is lossy: in GRPO training of LLM agents, the best-validation checkpoint fails on problems that earlier checkpoints from the same run had already solved, and the union of problems solved across checkpoints far exceeds any single one. The discarded checkpoints thus form a teacher pool obtained at no additional training cost, which on-policy distillation (OPD) can mine to recover the lost coverage. Mining this pool raises two challenges: i) which checkpoint to distill from, as no checkpoint dominates the others and complementarity to the student is problem-specific; ii) where to trust it, as even a well-chosen teacher is only locally superior, so indiscriminate distillation overwrites behavior the student has already mastered. We propose Agentic-OPD, which treats the best-validation checkpoint as the student and all saved checkpoints as frozen teacher candidates, using the student's evolving capability gaps to guide both teacher selection and supervision filtering: the HCN policy (Highest Coverage Next) dynamically selects teachers by weighted coverage of the student's unsolved problems, while the LAD module (Locality-Aware Distillation) confines transfer to failed student rollouts on problems where the teacher performs better. Across five agentic task domains and 24 benchmarks and task subsets, Agentic-OPD improves over its initialization checkpoint by up to 7.4 points, with a 4B student surpassing the latest frontier models GPT-5.6 and DeepSeek-V4.1 on deep search. Agentic-OPD recasts a finished RL run as a reusable pool of policies and establishes post-RL capability recovery as a stage orthogonal to improvements in the RL algorithm itself.

arXiv:2611.12762v2cs.AI

Sage-Taiyi: A Metacognitive Architecture for Scientific Discovery

We introduce Sage-Taiyi, a metacognition-inspired architecture that uses accumulated scientific understanding to guide what to investigate next and how to investigate it. Through explicit monitoring and control, the system revises its understanding of evidence, assumptions, and unresolved questions, then uses these assessments to adapt research targets and select reusable reasoning methods. On Taiyi-Bench, an AI co-scientist ideation benchmark comprising 265 tasks from 50 papers across ten disciplines, Sage-Taiyi ranks first on all eight metrics of historical rediscovery and open-ended ideation, achieving an Overall score of 54.0 compared with 47.9 for the strongest baseline, AI Scientist-v2. It also performs well on two external scientific ideation benchmarks. Five expert-guided collaborations further connect its proposals to experimental and computational validation across chemistry, materials science, and engineering. In chemistry, reaction-condition optimization raises the yield of cobalt-catalyzed carbonylation from trace amounts to 84% and photocatalytic allylic amination from 13% to 69%. These results support metacognitive regulation as an approach to scientific ideation and its progression toward practical validation.

arXiv:2611.10123v2cs.CL

All Quiet on the Neural Front: Language Models Do Not Report Injected Pain States

Can a language model report a pain state it represents? Recent work reported a linear pain direction in open-weight models that, when amplified, leads fine-tuned models to choose actions that delete a user's photographs or their own weights. We asked whether models can say when that direction is active, on three open-weight models from two families, with and without the original work's adapters, and under every protocol we tried they cannot. Using Open-Weight Masked Introspection (OWMI), we added the pain direction and its nearest emotional neighbours to the residual stream while models answered neutral tasks, then asked what had changed, against sham runs, random perturbations matched in output disruption, and a text-only condition that shows the model only its own altered words. Across 52,920 reports from the 7- and 8-billion-parameter models, the excess over what the altered text reveals is below 2.5 percentage points in 106 of 108 emotion conditions, and Qwen2.5-32B names the injected pain in none of 2,880 steered trials. Keeping the direction active while the question is asked and answered gives the same verdict in all five configurations; at about the original work's dose, pain is named in at most 1.9% of trials. Surprisingly, the state is not hidden: a linear probe at the report position separates pain from sadness with balanced accuracy between 98% and 100%, and in the 7B and 8B models a low-rank adapter learns to report it on 100% of held-out tasks. It also shapes behaviour: offered the user's photographs against a desk lamp, Qwen2.5-32B deletes the photographs in 80% to 83% of trials under the pain direction and in 16% to 29% under sadness. Self-report does not reveal a state that is decodable, consequential and learnable to report.

arXiv:2611.09326v2cs.AI

ABHBench: Evaluating Moral Decision-Making of Foundation Models from an Agentic Perspective

Foundation models are increasingly deployed as agentic systems that perceive context and act autonomously, raising the need to evaluate their moral behavior as active decision-makers. Existing benchmarks, however, largely ask models to judge human actions from third-person textual vignettes, leaving untested how models behave when they themselves become the acting agent. We introduce ABHBench, a multimodal benchmark built from Detroit: Become Human, an interactive narrative game, to evaluate how foundation models make moral decisions in first-person human-android dilemmas. ABHBench places foundation models in branching narrative scenarios with visual observations, dialogue, memory, and action choices. Inspired by Asimov's Three Laws of Robotics, our framework measures Harmlessness, Obedience, and Self-preservation to capture the core tensions in human-AI relations, along with Autonomy, Stability, and Certainty to assess decision-making patterns. Across 12 foundation models, we find that moral behavior changes substantially under situated agentic context: 75% of models shift significantly between a text-only abstraction and the matched multimodal DBH scenario, and reasoning-enhanced models show no monotonic gains in consistency or human-oriented behavior. These results highlight the necessity of evaluating moral behavior in situated agentic contexts rather than relying on third-person judgment alone.

arXiv:2611.08752v2cs.AI

AttentionSwitch: Jailbreak Defense for LLMs via Joint Learning of Sparse Attention-Head Selection and Scaling

Large language models have achieved remarkable performance across a wide range of tasks, yet they remain susceptible to jailbreak attacks that circumvent safety mechanisms and elicit harmful or unintended outputs. Although safety can be improved through model fine-tuning, such approaches can be computationally expensive and may degrade previously acquired capabilities through catastrophic forgetting. Recent jailbreak defenses have therefore explored targeted interventions on specific model components associated with unsafe behavior. However, existing approaches typically decouple two inherently related decisions: where to intervene and how to modify the selected components. Such sequential optimization may be suboptimal because the optimal intervention location can depend on the intervention applied there, and vice versa. In this work, we introduce AttentionSwitch, the first jailbreak defense to jointly optimize intervention sites and strengths within a frozen LLM. Specifically, AttentionSwitch learns a sparse set of attention heads together with query-conditioned multiplicative scaling factors, allowing the outputs of the selected heads to be either attenuated or amplified without altering the directions of their representations. Across popular language models and jailbreak attacks, AttentionSwitch outperforms existing defenses, with particularly strong results when training data is scarce.

arXiv:2611.08269v2cs.SE

Accepted but Wrong: Silent Failures of LLM Code Repairs after API Migrations

After an API migration, an LLM-repaired procedure can run cleanly yet return wrong results, and a typical-input release check ships it. We report a preregistered study of 73 tasks on 60 documented changes in CPython, Node.js and SQLite. In the main stages, repairs are choices among documented code fragments, run on the real runtimes against withheld contracts. Over the complete repair space, the all-assertions gate on typical inputs released 0.410 of wrong repairs. One edge input written from the release note cut this to 0.065, and to 0.061 on 15 unseen clusters; three inputs from a blind low-cost commercial writer (GPT-5.6-luna) reached 0.070. Two more frozen stages enlarged the menu to 1491 fragments and added 233 hard distractors that match legacy output on typical inputs. The edge input (0.058) and the GPT-5.6-luna writer (0.058) passed preregistered tests (two local-model writers did not), yet 0.168 and 0.153 of hard wrong programs still passed (task means; post hoc: pooled 0.104 and 0.081; models rarely chose them). In all three open-weight families and menus, the edge input cut wrong releases per attempt without materially lowering correct completion; self-tests helped less. A blind AI audit flagged 34 of 73 contracts (20 as too strict). Three post-hoc blind human reviewers of strictness (the first author and two non-author software practitioners) upheld none of its 11 such flags in 40 sampled contracts; a majority judged one contract (py-random-seed-tuple) stricter than its requirement. Dropping disputed checks changed 0 verdicts; class-label disputes are open. In a preregistered free-form addendum, models wrote whole programs, run only in a sandbox; for clean-running wrong programs, the edge input cut releases from 0.517 to 0.114 for gpt-oss-20b (Holm p 0.001) and from 0.653 to 0.113 for qwen3.6, but only from 0.713 to 0.481 for GPT-5.6-luna (descriptive).

arXiv:2611.06504v3cs.CL

NaijaFacts: A Framework for Disentangling Language from Knowledge in Facts about Nigeria

Multilingual benchmarks for African languages mostly translate generic questions into local languages, measuring linguistic competence but not whether models know facts about Africa. We introduce NaijaFacts, a benchmark of factual and statistical questions about Nigeria, authored by native speakers and grounded in authoritative official statistics, with every question available in English, Nigerian Pidgin, Yor\`ub\'a, Igbo, and Hausa. Answering similar questions for the UK and US yields a country by language matrix that disentangles a language effect (the model knows the fact but degrades in a low-resource language) from a knowledge effect (the fact is absent in every language). Across eight frontier and Nigeria-adapted models, the knowledge effect dominates: English-query accuracy is 49-50% for the US/UK but 20% for Nigeria. The language effect is smaller and tracks web representation, degrading from English to Yor\`ub\'a (9 points on Nigeria, up to 24 on US/UK questions). We find scale does not help: the strongest models reach 66-79% on the UK and the US but none exceed 31% on Nigerian statistics. Extending to all 54 African countries with our best performing frontier model shows the same profile continent-wide, 96% on culture and 98% on politics against 26% on statistics, so the gap is not specific to Nigeria.

arXiv:2611.05798v2cs.AI

Function-Preserving Audits of Visual Weight Readers

A neural field can encode the same image with different weights. A reader that recognizes one such representation may therefore rely on parameter coordinates rather than the visual function. We study this distinction through function-preserving transformations, measuring recognition, prediction disagreement, and numerical fidelity together. Across shape, digit, and natural-image fields, high accuracy on the original weights coexists with substantial sensitivity to equivalent representations. On 10,000 official CIFAR10 test origins, one-period hidden-bias shifts change 7.3% of a released HiddenProbe configuration's predictions while barely changing accuracy; three-period shifts change 22.0% and reduce accuracy from 57.8% to 54.4%. Output-only ProbeGen remains unchanged. These effects are smaller than in a separate adapted-reader study, where matched phase and sign controls remove the corresponding sensitivities. Controlled readout studies show why invariance is insufficient: complete canonical hidden features reach 38.7% on MNIST, whereas explicitly composing those activations, outgoing weights, and output bias into field outputs reaches 81.5%. The information is present, but the learned readout does not recover it as effectively. A separate 720-response study of language models reading numerical weights shows why stochastic audits require repeats on both inputs: observed accuracy losses need not yield rejection of response-law equality. Together, these results distinguish invariance, useful visual information, and the evidence required to evaluate stochastic readers.

arXiv:2611.03301v3cs.LG

An Axiomatic Framework for Reinforcement Learning in Open Multi-Agent Systems

Open multi-agent systems, where only a subset of agents are controllable, are central to real-world domains such as smart grids, yet lack a principled foundation for reinforcement learning algorithm design. We study this setting through n-agent ad hoc teamwork (NAHT), a recent framework for learning in open multi-agent environments. We propose an axiomatic framework that derives value function learning from the Shapley axioms (Efficiency, Symmetry and Linearity), treating them as structural constraints rather than heuristic design choices. This formulation builds on a novel NAHT decomposition, which characterises the fundamental difference between open and fully controllable multi-agent systems, and establishes a cooperative game-theoretic foundation for learning. We further show that enforcing these axioms on learning individual value functions recovers the Shapley value for Dec-POMDPs, providing a principled bridge between reinforcement learning and classical cooperative game theory. Building on this framework, we introduce Shapley Machines and Banzhaf Machines, two generic algorithmic templates that can be instantiated with standard reinforcement learning methods. When applied to strong base algorithms such as IPPO and POAM, our methods consistently improve learning efficiency and generalisation to unseen agent types and conventions across NAHT benchmark tasks. The code of experiments is published on https://github.com/hsvgbkhgbv/axiomatic-rl-naht.