Artificial Intelligence
Exploration-Preserving Policy Optimization
Hangzhan jin, Mohammad Hamdaqa and Doina Precup
Reinforcement learning with verifiable rewards improves
reasoning, while the allocation of learning signal shapes which solutions remain accessible under repeated sampling. Group-relative objectives assign equal advantages to equally rewarded responses, making aggregate credit proportional to sampled mode frequency. We introduce Exploration-Preserving Policy Optimization (ExPPO), a lightweight advantage-shaping rule that redistributes credit using
prompt-relative, length-normalized response surprisal and
prompt pass rate. ExPPO combines bounded shaping with shared normalization to preserve verifier polarity and approximately maintain each
prompt group's total absolute sequence-advantage mass. Our analysis characterizes response-level credit allocation alongside sampled mode updates, deriving local conditions for gains in
entropy and correct-mode discovery. Experiments show improved in-domain and out-of-domain
reasoning coverage, higher aggregate response accuracy, and strong coverage at large sampling budgets. A controlled multi-answer evaluation further demonstrates increased correct-mode yield and gains in diversity among verified-correct responses. Code is available at https://github.com/jinhangzhan/ExPPO
Read the full abstract ▾
Artificial Intelligence
SkillScriptBench: Benchmarking Self-Evolution of Executable Agent Skill Packages Beyond Markdown
Yuxuan Liu, Haoran Li, Yuhao Zhang et al. (+10 more)
Executable
Agent Skills combine natural-
language instructions and scripts into reusable packages for LLM
agents, and revising them requires fixing errors without breaking correct behavior. Existing
benchmarks do not systematically distinguish documentation repair, script repair, and preservation when evaluating skill self-
evolution. We introduce SkillScriptBench, a 350-task
benchmark designed to evaluate these capabilities separately. From a survey of over 35,000 GitHub-hosted Skill roots, we select 100 packages and construct 150 repair tasks. Each task pairs a package containing injected script faults with a maintenance request and executable checks of the required behavior. A complementary controlled track contains 200 tasks from 50 packages, each evaluated under the same maintenance request in four states: clean, documentation faults, script faults, and faults in both. Across four LLMs, methods that edit both documentation and scripts can repair script faults but do not consistently outperform Markdown-only revision on documentation repair or preservation. We therefore introduce AST-Guided Skill Revision, which uses abstract syntax trees and calling relationships to link maintenance requirements to relevant code locations. It restricts script edits to these locations and updates the documentation to match the revised scripts. Averaged across models, this revision stage yields absolute gains in repair success of 21.9% for Raw Package and 27.7% for CoEvoSkills on faulty packages. Absolute gains in the proportion of tasks solved in all three runs reach 20.8% and 31.5%, respectively, indicating more consistent repair success across repeated runs.
Read the full abstract ▾
All authors Yuxuan Liu, Haoran Li, Yuhao Zhang, Jiahe Guo, Hongyu Luo, Wenbin Hu, Huihao Jing, Kawai Chung, Junle Chen, Changxuan Fan, Qing Zong, Lingyun Xie, Yangqiu Song
Artificial Intelligence · Cross-listed from cs.LG
Selective Backpropagation for Efficient Few-Shot Class-Incremental Learning
Eeham Khan, Abdulmoumen Al-Atrash and Ali Ayub
Few-Shot Class-Incremental Learning (FSCIL) requires models to continuously learn new classes from limited samples while retaining prior knowledge, under strict constraints on compute and
memory. Existing approaches lie along a difficult trade-off: simple fine-tuning is computationally efficient but suffers from catastrophic forgetting, replay-based methods mitigate forgetting at the cost of substantial compute and
memory, and exemplar-free methods often reduce forgetting by freezing most of the backbone, improving efficiency at the expense of adaptability. We propose Selective Backpropagation (SBP), a deterministic parameter budgeting framework that bridges this gap. SBP restricts gradient updates to a pre-allocated subset of network parameters, freezing past knowledge and preserving unbiased capacity for future learning, enabling rapid adaptation without costly mask optimization. We show that SBP achieves strong performance across standard FSCIL
benchmarks while requiring training time close to that of naive fine-tuning and substantially lower training time than prior SOTA methods. Crucially, our experiments expose a limitation of standard FSCIL evaluation: performance on short, distribution-consistent
benchmarks does not necessarily predict behavior under distribution shift or over substantially longer learning horizons. We therefore evaluate FSCIL methods in cross-domain settings and over an 80-session ImageNet-1K stream. SBP remains strong across these regimes while maintaining low training cost, providing a favorable stability-plasticity-efficiency trade-off. Our code is available at https://github.com/PaInt-Lab/sbp-main-public.
Read the full abstract ▾
Artificial Intelligence
Behavioral History Outperforms Descriptions of the Person for LLM Synthetic Personas
Khashayar Pourtaheri and Ahmad Zareei
Large
language models (LLMs) are increasingly used as synthetic personas representing survey respondents. Their validity as substitutes for particular respondents depends on whether they reproduce individuals' decisions. We examine what information helps synthetic respondents predict each individual's later choices, using five conditions that add progressively richer information: no personal information, demographics, personality traits, cognitive scores, and finally the respondent's earlier survey choices as behavioral history. We use a two-wave panel of 845 US adults who completed measures of 14 behavioral biases (spanning risk, time preferences, overconfidence, and
reasoning), so each respondent's earlier answers provide a human test-retest
benchmark; in the behavioral-history condition, all items that score the target bias are withheld. At the population level, the average number of biases per respondent in every condition is close to the human average (7.1-8.1 biases, against 7.1 for humans). This aggregate similarity masks differences in variance: persona descriptions recover only 53-67% of human between-person variation, whereas adding behavioral history restores it to approximately the human level. At the individual level, description-based personas achieve only 7-12% of the informedness observed in human test-retest responses, while adding behavioral history raises this to 28%. The condition including behavioral history has the highest estimated informedness in all 17 demographic groups, whereas description-based conditions provide little or no information for some groups. Synthetic responses also exhibit stronger education- and income-related differences than human responses. For LLM synthetic personas, a respondent's past answers add more to individual-level prediction than a description of who they are.
Read the full abstract ▾
Artificial Intelligence · Cross-listed from cs.LO
Solving VeriContest with a Lean-Backed Rust Verifier
Traian Serbanuta, Jun Xu, Andrei Stefanescu et al. (+1 more)
VeriContest is a
benchmark of 1007 competitive-programming problems in Rust, each with a Verus specification, a judge-accepted solution, and a Verus proof. Its authors report that proof generation is the bottleneck for frontier models: given the specification and the code, the best model produces an accepted Verus proof for 13.95% of the problems on the first attempt. We report on solving the same proof-generation task with Rust-Prover, a verifier for Rust backed by Lean 4. The Verus specification and the Rust code are restated and translated into Lean, each specification becomes a
theorem, and
agents prove the
theorems with Lean's kernel as the final check. All 1325
theorems of all 1007 problems were proved. 1259 of them were proved in one run of under 32 hours on Claude Opus 5.5, at a median of 3.2 minutes and $1.17 per proof, and 70% of them on the first iteration. The restated specifications were checked against the
benchmark's test suites, and reviewed where no suite applies. None was wrong or weakened. The translated Lean programs were run on 21,413 of the
benchmark's test cases and produced the same output as the Rust programs on every one. Across four Claude and four GPT models at five
reasoning-effort settings, every current frontier model proves nearly all of a ten-
theorem sample at every setting, and more effort raises the cost without raising the number of proofs. The cheapest Claude setting, Sonnet 5.5 at low effort, proves all of the 50 hardest
theorems. We also rerun the
benchmark's own Verus protocol with Claude Opus 5.5 on the 50 problems with the longest reference proofs. Opus 5.5 alone fails to prove one of them.
Read the full abstract ▾
All authors Traian Serbanuta, Jun Xu, Andrei Stefanescu, Cosmin Radoi
Artificial Intelligence
Teaching Agents to Code Reliably
Muhammad Ahmed Mohsin, Myeongsoo Kim, Kangrui Ruan et al. (+3 more)
Autonomous coding
agents solve repository issues by reading code, running commands, editing files, and submitting patches. Extra inference-time compute yields gains only when it produces a useful repair and supplies reliable evidence for choosing one. Three behaviors decide both, and we argue they are teachable rather than byproducts of scale, so a policy can carry them instead of a scaffold. Location diversity remains narrow, since attempts return to the same site and extra samples add no coverage. Edit diversity is left unexploited, since methodologies that differ resolve complementary issues no single run reaches. Verification misleads, since a test the
agent writes for its own patch accepts many incorrect ones. Directing search by execution feedback and scoring each patch against its own reverted tree resolves 52.8% of SWE-bench Verified using 48.1% of the
agent-steps an eight-sample baseline spends. Training moves these behaviors into the policy. On the 270 issues held out from SFT and RL training, weighted supervised fine-tuning raises pass@1 from 31.9% to 35.2% and pass@8 from 46.7% to 51.1%. A reinforcement objective then trains the verifier against gold-labeled repairs and incorrect variants, crediting the assertions that detect them. It raises pass@1 to 43.0% and pass@8 to 60.7%, lifts verifier precision from 26.8% to 41.7%, and more than halves false acceptance. Resolution improves on two of three out-of-distribution suites and verifier precision on all three, and the gains hold at 7B, 14B, and 30B against published coder baselines.
Read the full abstract ▾
All authors Muhammad Ahmed Mohsin, Myeongsoo Kim, Kangrui Ruan, Shweta Garg, Varun Kumar, Murali Krishna Ramanathan
Artificial Intelligence · Cross-listed from cs.AR
An Executable Benchmark for LLM-Based HLS Repair:Design Complexity and Repair Underconstraint
Maisha Mastora and Dean Sullivan
Automated repair of High-Level Synthesis (HLS) designs using large
language models (LLMs) is an emerging but underexplored problem. While LLM-based repair shows strong results on register-transfer level (RTL) Verilog, the only prior systematic study of HLS logic repair reports just 10.5% correction accuracy for GPT-4, with no analysis of why repair fails or what drives difficulty. We present the first comprehensive evaluation of LLM-based HLS repair across four models (GPT-4o, GPT-4o-mini, GPT-5.4, and Claude Opus 4.6) on 125
benchmark instances spanning eight logic bug types across three open-source HLS suites (CHStone, MachSuite, Polybench). We construct the first executable APR-style HLS repair
benchmark with suite-specific functional oracles, enabling pass@k evaluation rather than the string-match approximations used in prior work. Design context and scale, rather than bug type alone, dominate repair difficulty: repair rates range from 6-45% on complex cryptographic kernels (CHStone) to 81-93% on compact algorithmic kernels (MachSuite) despite identical bug type distributions. We introduce solution multiplicity, the fraction of distinct patches generated across repeated repair attempts, as an empirical measure of repair underconstraint, and show it strongly predicts repair failure across all four models (Spearman rho = -0.393, p = 6.61 x 10^-20). Frontier models reduce underconstrained instances from 41-46 (GPT-4o, GPT-4o-mini) to just 3 (Claude Opus 4.6), with pass@1 improving from 42-53% to 73-76%. SHFT bugs remain consistently hard across all models, and semantic bug classes such as buffer indexing become reliably repairable only at frontier scale. These findings show that future APR
benchmarks must include complex, executable, and weakly identifiable designs that remain challenging after frontier LLM repair.
Read the full abstract ▾
Artificial Intelligence
ROAR: Unifying Runs across Heterogeneous AI-Driven Research Systems
Leo Y. Lin, Vishakha Ramani, Z. Berkay Celik et al. (+2 more)
Each run of an AI-driven research system (ADRS) is an expensive search over a vast solution space, and dependable evaluation requires many runs, making run data both costly to produce and valuable to retain for large-scale analysis. Yet this data remains fragmented: teams operate in isolation, ADRS frameworks emit results in different formats, and no shared infrastructure exists to aggregate or compare runs across problems and systems. We present ROAR, a solution for systematically unifying and analyzing heterogeneous ADRS outputs. ROAR addresses two challenges: reconciling heterogeneous ADRS outputs and enabling analytics across runs with different objectives and scoring functions. We achieve this through a relational schema and parsing layer that normalize heterogeneous ADRS outputs while preserving data lineage and temporal structure, and accommodating new systems without requiring schema modifications. From building a corpus of more than 900 runs from multiple ADRS, we show how pooled data can reveal properties of problem landscapes that are difficult to observe. Consistent with prior work, runs with identical configurations may converge to different scores. We find that many runs realize most gains early, and that the effectiveness of different strategies for incorporating prior solutions into the search process varies across problems. We further show that the pooled corpus is actionable and not merely analytical by using ROAR to configure ADRS runs. Together, these results illustrate how pooled ADRS data can expose problem-dependent structure in search behavior that is difficult to detect from any single system, team, or
benchmark. Such cross-cutting insights are difficult to obtain while runs remain siloed; ROAR is the first infrastructure designed to unify them.
Read the full abstract ▾
All authors Leo Y. Lin, Vishakha Ramani, Z. Berkay Celik, Paul Castro, Marquita Ellis
Artificial Intelligence
LatentQuant: Preserving the Policy-Facing Latent Contract under NVFP4 VAE Quantization
Ziye Deng, Lufang Chen, Shuyu Feng et al. (+7 more)
Recent world action models (WAMs) reuse pretrained
video VAEs whose encoder latents directly condition downstream action policies. Quantization must therefore preserve not only reconstruction fidelity but also the policy-facing latent contract expected by the frozen policy. Direct NVFP4 leaves W4A4 quantization error uncompensated, whereas joint quantization-aware training (QAT) can recover reconstruction by moving this representation. On Wan2.1, joint QAT nearly matches FP32 VBench-7 (0.7403 versus 0.7409), yet LIBERO success collapses from 95.5% to 10.5%. Controlled decoder-only experiments show that activation quantize-dequantize operations alter the reconstruction signal and decoder Jacobian, redirecting the gradient returned to the encoder and inducing persistent latent drift. Based on this mechanism, we introduce LatentQuant, a two-stage NVFP4 QAT framework that first aligns the quantized encoder with its high-precision counterpart, then freezes it while adapting the decoder. Across Wan2.1 and Wan2.2, LatentQuant preserves near-baseline control and high reconstruction quality, achieving 95.75% success on LIBERO and 68.8% on RoboTwin. On NVIDIA B300 GPUs, NVFP4 execution achieves 1.17x-1.26x end-to-end VAE speedups over BF16 cuDNN.
Read the full abstract ▾
All authors Ziye Deng, Lufang Chen, Shuyu Feng, Zhenwei Duan, Zicong Ye, Yu Sun, Xiaofan Li, Ruyi Gan, Hao Wang, Hao Zhang
Artificial Intelligence
Retrieval-Augmented Large Language Model Decision-Making for Autonomous Driving Guided by Chinese Philosophical Wisdom
Xiaojun Bi, Xiaoyuan Ma, Yiwen Sun et al. (+5 more)
Autonomous driving decision systems must balance safety, efficiency, and social norms in complex traffic interactions. Philosophical and ethical considerations have received limited
attention in existing autonomous driving decision-making approaches based on numerical optimization, sequence prediction, and large
language models (LLMs). We propose Chinese Philosophical Wisdom-Guided Driving (CPW-Drive), a closed-loop
retrieval-augmented generation (RAG) framework that incorporates value guidance derived from Chinese philosophy into autonomous driving decision-making. Using Chinese Confucian thought as its knowledge source, CPW-Drive consolidates LLM-extracted keywords from relevant classical texts into driving-relevant value principles through manual screening and validation. It then contextualizes these principles through scenario-specific cases to form retrievable and reusable value guidance. We further propose Physics-aware Spatial Similarity
Retrieval (PSSR), which compares vehicle layouts and velocity-extrapolated states to retrieve physically relevant historical cases. On Highway-env's multilane highway-driving task, CPW-Drive achieves success rates of 93.0%, 86.0%, and 72.0% across three traffic configurations. These results outperform the strongest baseline by 8.0, 22.5, and 25.0 percentage points, respectively. Across all configurations, CPW-Drive achieves the highest collision-free step count and maintains a low lane-change frequency. The results suggest that structured value guidance can improve simulated closed-loop safety and stability while introducing efficiency and latency trade-offs.
Read the full abstract ▾
All authors Xiaojun Bi, Xiaoyuan Ma, Yiwen Sun, Tianren Huang, Chaoran Liu, Bokai Huang, Hao Yang, Baichuan Mo
Artificial Intelligence · Cross-listed from cs.CL
A Step Towards Forgetting: Optimiser History and the Loss of Answer Mass
Valeria Ruscio, Seth Nabarro and Keiran Thompson
During fine-tuning, a
language model can assign less probability to previously learned answers even when the current gradient acts to preserve that probability. With momentum, each update also carries gradients computed at earlier model states, and these stored contributions can push the model in the opposite direction. We investigate how this optimiser
memory contributes to forgetting by separating old-task loss into confusion among its answers and leakage of probability outside the answer set. Across three
language-model families, answer mass consistently declines while discrimination among old answers usually improves: the model becomes less likely to produce answers that it can still distinguish correctly. Decomposing Adam updates reveals opposing contributions to this loss of answer mass. Over training, accumulated history favours leakage, while the current gradient opposes it. Resolving history by age shows that the harmful contributions come mainly from older gradients of the new task, whereas recent gradients tend to protect the old answers. Changes in history's effect are dominated by its orientation relative to the old-task gradient. Interventions that reset momentum while matching the initial update norm establish that stored history affects retention, with state-dependent immediate effects and lower final old-task loss over longer Adam continuations, mainly through recovered answer mass. Finally, integration along finite updates shows that most sampled large loss increases are captured by local projections, while
curvature along history amplifies some events. Together, these findings reveal how an optimiser's
memory can erode learned behaviour even as its current gradient acts to preserve it.
Read the full abstract ▾
Artificial Intelligence
MLLMs Fail to Refuse when Using Tools Agentically
Rikiya Takehi, Ryo Hachiuma, Shaona Ghosh et al. (+3 more)
Agentic
multimodal large
language models (MLLMs) have recently pushed the frontier of visual
reasoning by calling tools such as zooming and tagging. Despite the recent strong success of agentic MLLMs, this work uncovers a critical safety failure in the tool-use paradigm: agentic tool-using MLLMs become less capable of refusing harmful requests. Our experiments confirm that, across three popular safety
benchmarks, all the top open- and closed-weight MLLMs we test exhibit significantly lower safety in tool-using settings than in non-tool settings, with a relative refusal failure rate increase of up to 68.7%. Based on analysis of 100,000+ responses, including extended experiments, we also propose two possible reasons for this safety degradation.
Read the full abstract ▾
All authors Rikiya Takehi, Ryo Hachiuma, Shaona Ghosh, Dan Zhao, Yu-Chiang Frank Wang, Yusuke Hirota
Artificial Intelligence
SGAnalog: An End-to-End Circuit Benchmark from Open-Source Silicon Tapeouts
Yueting Li and Weihang Ding
Existing analog integrated
circuit design
benchmarks make two questions hard to answer: whether a model has learned transferable
circuit skills rather than recalled familiar examples, and whether its output works under defined process and test conditions. We introduce a
benchmark built from human-designed, open-source
circuits associated with Tiny Tapeout manufacturing shuttles. The collection contains 273 topologically distinct top-level designs. Every source is retrieved at the revision recorded for its shuttle submission and processed in a fixed containerized environment. The pipeline exports each eligible schematic image and its SPICE netlist from the same source file, giving transcription an exact structural reference. Commit dates support model-specific training-cutoff analysis, while author testbenches provide the simulation context for sizing. The
benchmark evaluates schematic-to-netlist transcription and device sizing. Across seven models on a fixed set of 66 transcription tasks, the strongest model reaches 56.1% exact graph isomorphism, and six of seven models drop sharply from the small to the medium tier. For one frontier model, removing author-chosen labels reduces exact matches while preserving aggregate structural F1, suggesting that labels can aid connectivity tracing. On the 17 sizing tasks, the leading model converges on all 17 proposals and reaches 91.2 out of 100 against the human reference, while the two newest Claude models refuse 4 and 11 of the same
prompts they transcribe without objection; a proposal without sizes scores zero. The two tasks produce different model rankings, exposing distinct visual and design capabilities and, in one family, a policy rather than capability limit.
Read the full abstract ▾
Artificial Intelligence
Proxy Confidence: Auditing Black-Box LLM Agents with a Surrogate's Log-Probabilities
Yikai Zhao, Saurabh Pandey and Pradeep Kumar Misra
A deployed LLM
agent emits tool calls, queries, and code that can be silently wrong – by the time the error surfaces, the action has run. Frontier chat APIs hide the model's
token probabilities; the
agent's stated confidence barely beats chance on the mistakes that matter; and resampling does not help, since frontier models are highly repetitive, reproducing the same call across samples. We recover the missing signal from a low-cost open-weight surrogate run in parallel. It reads the same context, schema, and proposed action as the
agent, then scores the call from its own log-probabilities through a family of complementary readouts: teacher forcing and request-PMI weigh the likelihood of each argument value, a discriminative verdict judges the call as a whole, and tool-choice competition tests the function against its siblings. One principle says which to trust: a generative likelihood localizes wrong argument values, while the verdict catches holistically wrong calls. When the error type is unknown, an ensemble is the low-regret default. The readout is training-free, needs no access to the
agent's internals, and costs one prefill pass alongside the tool call. On difficult coding tasks it reaches AUROC 0.825 where the actor's stated confidence is near chance (0.598), and the generative readouts beat it by +0.07 to +0.28 across three further actors. Against self-consistency it gains +0.14 to +0.19 on near-deterministic actors, at 1/K the cost. The signal drives two deployment modes: a real-time gate escalating the least-trustworthy calls for review (+0.05 to +0.30 accepted-action accuracy at 50% coverage), and confidence feedback, returning the tool result with the score so the
agent adapts its next step – lifting task success on live-execution
benchmarks (+0.119 and +0.137, p <= 1e-4) and beating a random-value control where step errors are silent (+0.078, p = 0.003).
Read the full abstract ▾
Artificial Intelligence
REACT: Physically and Chemically Consistent Reconstruction of Marine Active Tracers
Wenbin Dai, Hao Zheng, Shiyu Liang et al. (+6 more)
Reconstructing global sea surface pH from sparse observations is critical for monitoring ocean acidification and understanding marine carbon cycling. Traditional assimilation and inverse models are physically grounded but costly for large-scale reconstruction. Recent black-box and physics-guided AI models improve efficiency, but are mainly designed for passive tracers, where the reconstructed variable is also the transported inventory. In contrast, pH is an active carbonate tracer: it is the prediction target, while dissolved inorganic carbon (DIC) is the conserved carbon inventory. This mismatch can produce low pH error while violating carbonate closure and source-free carbon conservation. To address this, we introduce REACT, a carbon-first reconstruction framework that decouples transport, active correction, and chemical decoding. REACT transports a latent carbonate state with a conservative advection–
diffusion solver, captures non-conservative carbon-cycle variations with a source module, decodes the corrected state into pH, and constrains the output through carbonate
equilibrium. This design keeps pH as the target while enforcing consistency on the underlying carbon state. On simulation data, REACT reduces pH NRMSE by (14.7%) and chemical consistency error by (24.0%) over the best baseline. Cross-temporal-scale evaluations show robustness against error accumulation from coarse to fine temporal scales, and ablation studies validate the effectiveness of each component.
Read the full abstract ▾
All authors Wenbin Dai, Hao Zheng, Shiyu Liang, Chaofan Sun, Xueying Zhang, Hanbo Huang, Xuan Gong, Yiran Zhang, Enhui Liao
Artificial Intelligence · Cross-listed from cs.CV
Streaming Multi-Track Timeline Control for 3D Human Motion Generation
Yangsong Zhang, Anujith Muraleedharan, Rikhat Akizhanov et al. (+3 more)
Text-driven human motion generation has advanced substantially, yet most methods assume instructions are available before synthesis. Interactive applications require responding to new instructions while continuing ongoing actions, such as answering a phone while walking. Existing approaches address streaming generation or simultaneous composition without explicitly combining streaming instruction arrival with independently timed, overlapping actions. We introduce streaming multi-track timeline control and propose TimelineControl to incorporate new instructions alongside ongoing actions. Interval-aware conditioning preserves instruction timing, while causal part-structured representations and part-aware denoising coordinate concurrent actions across body regions. We also construct TimelineMotion, a
dataset with overlapping instruction intervals and body-part annotations. Experiments on TimelineMotion and MTT demonstrate improved semantic
alignment and temporal adherence over evaluated streaming baselines, including models retrained on the same data. Ablations and human evaluations validate our design, complemented by spatial conditioning and humanoid execution demonstrations. Our code, data and models will become publicly available.
Read the full abstract ▾
All authors Yangsong Zhang, Anujith Muraleedharan, Rikhat Akizhanov, Gül Varol, Fabio Pizzati, Ivan Laptev
Artificial Intelligence
Training Numerical Intelligence via Auto-Diagnosis and Skill Discovery
Peter Chen and Wotao Yin
AI
agents are becoming increasingly capable of generating scientific code, but generating code is not the same as improving the
algorithms behind it. For numerical solvers, execution feedback can expose poor performance, but rarely reveals its underlying cause and how to address it. We introduce Auto-Diagnosis and Skill Discovery (ADSD), a framework that links numerical diagnosis to reusable solver self-improvement. ADSD follows a diagnosis-first paradigm that first explains why a solver performs poorly, then uses this diagnosis to guide the discovery of appropriate numerical methods. The resulting knowledge is packaged into reusable solver skills, turning solver improvement from trial-and-error editing into a structured process of diagnosis, discovery, and implementation. Across four challenging numerical domains–power flow equation, AC optimal power flow control, stiff ordinary differential equations, and heterogeneous
diffusion PDEs–ADSD consistently improves solver accuracy, robustness, and efficiency. On GOC-500 power flow, for example, ADSD reduces mean solver error by nearly $71\times$, with improvements further transferring to unseen grid
topologies and operating regimes.
Read the full abstract ▾
Artificial Intelligence · Cross-listed from cs.RO
GOTT: Object-centric Dexterous Manipulation with a Reusable Cross-Embodiment Primitive
Yulin Liu, Lai Wei, Yen-Jen Wang et al. (+4 more)
Foundation models and large-scale human data provide rich sources of manipulation intent, but translating this intent into multi-fingered
robot behavior remains difficult. Dexterous hands still lack a reusable low-level primitive that reliably establishes contact across tasks and embodiments. We propose GOTT, a reach-acquire-move framework built around a single cross-embodiment contact-acquisition primitive. Given a
robot-agnostic object trajectory and a reach specification, GOTT first brings the hand near a task-relevant contact region. The shared closed-loop primitive then establishes stable contact from this approximate initialization, and a pose-conditioned controller tracks the desired object motion. Reach specifications may come from future-aware planning, external models, or human demonstrations, while the primitive and tracking backend remain unchanged. Simulation and real-world experiments show that GOTT is able to establish robust contact across diverse objects, arm-hand platforms, and seen and unseen hand morphologies. It also consistently improves end-to-end task success over open-loop grasp execution.
Read the full abstract ▾
All authors Yulin Liu, Lai Wei, Yen-Jen Wang, Akash Sharma, Pieter Abbeel, Henrik I. Christensen, Haozhi Qi
Artificial Intelligence · Cross-listed from cs.LG
Retrieval-Centric Deep Learning in Growing Nonparametric Neural Networks
Maximilian Schlegel, Rajai Nasser, Seijin Kobayashi et al. (+6 more)
We investigate a general-purpose layer for deep learning that, instead of compressing arbitrary-size training data into fixed-size weight
matrices, stores a new pair of key-value representations for every data point during training, and retrieves and recombines these representations through an
attention mechanism at inference time - resulting in a growing neural net (NN). While Irie et al. (arXiv:2202.05798) have put forward this perspective from the classic duality expressing any linear layer in a deep NN trained by gradient descent as linear
attention (LA) over the training data points, replacing LA by more powerful
attention functions, as they suggest, turns out to be non-trivial: we show that naively applying learning rules from the LA case to advanced kernels does not lead to principled optimization. Here we fill this gap and develop functional gradient-based learning rules for kernelized
attention layers, based on radial basis function (RBF) and softmax-like kernels - establishing the principled "
retrieval-centric deep learning" (RCDL) paradigm. Empirically, we demonstrate the promising performance and learning-efficiency of RCDL on image classification and synthetic teacher-student learning tasks. Moreover, we show that replacing LA in the dual form of NNs by advanced LA variants, namely MesaNet/DeltaNet, yields a formal connection to recently proposed optimizers for conventional fixed-size NNs, offering a novel perspective on deep learning optimization.
Read the full abstract ▾
All authors Maximilian Schlegel, Rajai Nasser, Seijin Kobayashi, Yanick Schimpf, Oliver Sieberling, Robert Obryk, Kazuki Irie, João Sacramento, Johannes von Oswald
Artificial Intelligence · Cross-listed from cs.CR
Beware EviLLM: Enabling Vulnerability Injection via Large Language Models
Zeezoo Ryu, Simon Chung, Muhammad Faraz Karim et al. (+4 more)
Advances in large
language models (LLMs) have enabled AI-driven code generation from natural
language specifications, introducing new attack surfaces for injecting vulnerabilities into software. Prior work has studied this problem only in benign settings where vulnerabilities are introduced inadvertently, or under unconventional threat models where the LLM itself is malicious (backdooring) or the user is the attacker (jailbreaking). In this paper, we study a more realistic threat model: a third-party adversary, with capabilities comparable to existing cybercriminals, compromises the AI code generation pipeline to deliberately introduce vulnerabilities. We call this the EviLLM attack. We have implemented two instances of EviLLM, each of which only requires the underlying LLM to be accessed through a compromised account or browser, and can inject vulnerabilities from 13 CWE classes. As we show in our feasibility study, both attack vectors are already used to implement many existing cyberattacks. Our user study shows that 7 out of 8 and 10 out of 13 participants did not notice the vulnerabilities injected by the two instances of EviLLM, and 13 out of 21 participants "rarely" or "never" considered the risk of an attack like EviLLM.
Read the full abstract ▾
All authors Zeezoo Ryu, Simon Chung, Muhammad Faraz Karim, Anna Raymaker, Karan Singh Jodha, Yash Chaturvedi, Sukarno Mertoguno
Artificial Intelligence · Cross-listed from cs.CL
SYNLAT: Syntax-Aligned Text-Latent Compression for Chain-of-Thought Reasoning
Yifeng Zhao, Hongjun Yu, Shibo Wang et al. (+5 more)
Long chain-of-thought (CoT) traces impose substantial output-
token costs. Under constrained budgets, compression must preserve answer-critical information, making boundary placement central.
Token-level and fixed-length boundaries can fragment coherent spans such as phrases, formulas, and local derivations, whereas step-level boundaries can bind content requiring different compression actions. We introduce SynLat, a text-latent CoT framework that aligns compression boundaries with syntactic structure through non-overlapping Syntax-Aligned Units (SAUs). An answer-conditioned Teacher constructs progressive KEEP/LATENT targets for a single compression-conditioned Student, which generates mixed
reasoning from only the question and requested compression level at inference. Across two Qwen3 Student scales, Standard-CoT and Long-CoT groups, and three compression levels, SynLat matches or exceeds the strongest evaluated baseline in all 12 task-group aggregates and strictly leads in 11 under the reported achieved-CR selection protocol. Overall gains reach 3.6/2.6 points at MEDIUM and 7.0/5.5 points at HIGH for Qwen3-8B/14B, with larger advantages under stronger compression, particularly on Long-CoT groups.
Read the full abstract ▾
All authors Yifeng Zhao, Hongjun Yu, Shibo Wang, Yunjiao Zhou, Zixiao Zhu, Zhipeng Ning, Kezhi Mao, Junlang Qian
Artificial Intelligence · Cross-listed from cs.NE
Distribution Matching Evolutionary Algorithms for Rare Event Sampling
Yonatan Gideoni and Yarin Gal
A novel discovery is one which is both useful and surprising: a generative model's output is a useful discovery if it has a low probability of being generated (it's surprising) and a high reward (it's useful). Global optimization can directly increase the probability of sampling high rewards but typically requires updating model weights. Such gradient based optimization is expensive and bars using capable closed-source models. Instead, modern search methods for discovery sacrifice the global target, and use evolutionary
algorithms with local reward maximizing objectives, permitting the search to focus only on high probability samples. In this paper, we interpret various evolutionary
algorithms as approximate Markov Chain Monte Carlo, an optimization-free method to sample from complex distributions. This interpretation allows developing Distribution Matching Evolutionary
Algorithms (DME), a class of search methods which sample from a global target distribution without updating weights. Empirically, DME has a higher sample efficiency than existing methods on problems requiring many samples to find a solution.
Read the full abstract ▾
Artificial Intelligence · Cross-listed from cs.CL
OncoNoteBERT: A Foundation Representation Model for Natural Language Processing of Real-World Outpatient Oncology Notes
Wuraola Oyewusi, Eliana Vasquez Osorio, Goran Nenadic et al. (+1 more)
Real-world outpatient oncology notes contain specialised terminology, tumour staging expressions, treatment names, toxicity descriptions, and institution-specific de-identification markers that may not be represented efficiently by general biomedical or adjacent clinical
language models. We developed and evaluated oncology-specific BERT-style encoders using a governed UK outpatient oncology corpus comprising 290,026 notes from 21,564 patients treated for lung and head-and-neck cancer. We compared RadBERT and PathologyBERT with two local strategies: OncoNote-RadBERT, produced by continued masked
language model pretraining, and OncoNoteBERT, trained from scratch with an oncology WordPiece tokenizer. Models were evaluated using masked
language modelling loss and perplexity on the validation set, tokenizer fragmentation metrics, clinical term tokenisation, masked-
token probes, and exploratory representation analysis. Both external encoders fit the oncology corpus poorly in zero-shot evaluation (perplexity 113.04 for RadBERT; 2035.03 for PathologyBERT), while continued pretraining produced the strongest fit (2.10 for OncoNote-RadBERT). OncoNoteBERT achieved perplexity 2.83 but produced the most efficient tokenisation, with lower subword fertility and shorter normalised sequence length. It also returned a clinically acceptable prediction for 12 of 13 masked-
token probes, compared with 7 of 13 for OncoNote-RadBERT. This divergence between corpus-level fit and masked-
token performance was partly attributable to tokenizer fragmentation rather than learned semantics alone. Both locally developed models represented the institutional placeholder as a single learnable
token. These findings show that continued adaptation and bespoke tokenisation provide complementary benefits, and that representation-layer design matters before adjacent-domain encoders are applied to oncology NLP.
Read the full abstract ▾
All authors Wuraola Oyewusi, Eliana Vasquez Osorio, Goran Nenadic, Gareth Price
Artificial Intelligence · Cross-listed from cs.CL
The Score Is Not the Structure: Brain Alignment and Cross-Lingual Transfer
Saman Rahbar
Similarity scores are often offered as evidence that a model shares structure with the
brain or across
languages. We ask what such a score reads when that structure is removed, or when the instrument measuring it does not work, in two settings. Across seventeen
languages, a grammaticality probe transfers worse between more distant
languages (r = -0.66). But the probe's own accuracy falls along the same axis and is at chance in four
languages, four of the five most distant. Dropping them halves the explained variance, and because it also narrows the range of distances, the design cannot say how much of the gradient is the instrument. Counting the 272
language pairs as independent gives p = 0.0006 for a steering effect that is null when the seventeen
languages are the unit (p = 0.155). In
brain alignment, a training objective raises a
language model's similarity to fMRI responses from 0.10 to 0.34 (reliability ceiling 0.54), yet targets with the
brain correspondence destroyed still reach 0.31. A shuffled target scores about k/n against the real one, for rank k and n sentences, a baseline that can be computed before any model is trained. Both interventions work, yet we detect no distance-graded steering effect and no syntactic benefit from
alignment. Before reading a correspondence score, measure how much of it survives without the correspondence.
Read the full abstract ▾
Artificial Intelligence · Cross-listed from cs.CR
From Requirements to Attack Trees: Grounded LLM Agents for Design-Time Security Review
Akash Iyer, Taha Demirkan, Keerthi Koneru et al. (+3 more)
Design-level security weaknesses can arise from requirements, trust assumptions, missing controls, and data flows before implementation begins. Existing security practices often identify these issues after code is written. We present a multi-
agent LLM framework for design-time security analysis from product requirement documents and architecture diagrams. The proposed framework parses architecture diagrams into graph representations, generates misuse and failure cases, constructs attack trees, checks governance and compliance gaps, recommends mitigations, assigns enterprise security-domain tags, and produces a candidate revised architecture recommendation for expert review. The framework does not retrieve from Common Weakness Enumeration (CWE) databases at inference time. Instead, it analyzes system behavior, trust boundaries, component interactions, and data-flow assumptions. Misuse cases act as intermediate representations that link findings to system components and attack paths, while a validation and refinement loop filters unsupported findings and improves grounding, traceability, and actionability. We evaluate the framework on a Microsoft reference-labeled threat-modeling example, labeled synthetic PRD–architecture pairs, and two open-ended systems: Berty and Gas Town. The reference-labeled case supports threat-recovery and actionability analysis, while the open-ended cases evaluate validity, noise, traceability, actionability, redundancy, and attack-tree quality. Results show that architecture-informed, misuse-driven
reasoning improves review quality compared with single-shot and ablation baselines. Keywords: LLM Multi-
Agent Systems, Design-Time Security, Threat Modeling, Vulnerability Discovery, Architecture Diagrams, Security Analysis, Misuse Case Derivation, Attack Trees, Iterative
Reasoning, Security Governance.
Read the full abstract ▾
All authors Akash Iyer, Taha Demirkan, Keerthi Koneru, Aaryan Siddharthan, Sheethal Kumar, Ramesh Radhakrishnan
Artificial Intelligence · Cross-listed from cs.CV
Less Decoder is More Encoder: Geometric Representation Learning from Novel View Synthesis
Keerthi Kaashyap, Dennis Anthony, Akshay Krishnan et al. (+5 more)
This paper examines the role of Novel View Synthesis (NVS) in
geometric representation learning. In principle, NVS should reason about 3D scene structure, thereby enabling transferable multi-view
geometric representations. Yet, existing encoder-based NVS methods yield poor representations. This is not because of a lack of supervisory signal, but rather due to inconspicuous architectural choices: spatially expressive decoders that dilute representational capabilities of the scene encoder, and low-level pixel-space targets that hinder feature learning. We present SNAP, a self-supervised encoder-decoder
transformer that addresses both through a pose-conditioned local decoder and a latent-space reconstruction objective. SNAP is task agnostic, and we show that it is competitive with special-purpose geometry-supervised methods. SNAP also performs competitively against self-supervised representations across five tasks: visual localization, pose estimation, point correspondence, depth estimation, and
robot manipulation. Remarkably, SNAP's patch features exhibit emergent viewpoint invariance that approaches heavily supervised models despite lower compute and data budgets. Under camera shifts where standard 2D representations collapse, SNAP degrades more gracefully, revealing that restricting decoder expressivity actively prevents the suppression of transferable
geometric structure. https://snap-nvs.github.io
Read the full abstract ▾
All authors Keerthi Kaashyap, Dennis Anthony, Akshay Krishnan, Nhi Ngoc Nguyen, Jeremy Collins, James Hays, Shreyas Kousik, Animesh Garg
Artificial Intelligence · Cross-listed from cs.CV
4DCodeBench: Benchmarking Agents on Inverse Graphics of Dynamic Scenes
Ruihong Shen, Žiga Kovačič, Peter Kulits et al. (+6 more)
We introduce 4DCodeBench, a
benchmark for 4D inverse graphics through code generation, in which
agents reconstruct dynamic scenes from
video as executable graphics programs. To accomplish this,
agents must translate visual observations into compact representations of scene structure and dynamics, by implementing abstractions such as physical simulations to reproduce complex behavior. To evaluate this capability, we curate a set of real-world
videos and construct synthetic scenes spanning diverse physical phenomena, including deformation, fluid flow, and fracture. We perform extensive benchmarking of frontier models, finding that strong static reconstruction capabilities do not yet translate into reliable reconstruction of complex dynamics. 4DCodeBench provides a testbed for tracking progress toward
agents that can interpret the dynamics of the world through code. Our
benchmark is available at https://github.com/4DCodeBench/4DCodeBench
Read the full abstract ▾
All authors Ruihong Shen, Žiga Kovačič, Peter Kulits, Xingrui Wang, Zizhang Li, Joshua B. Tenenbaum, Alan Yuille, Jieneng Chen, Jiajun Wu
Artificial Intelligence · Cross-listed from cs.LG
What Should World Models Forget? Stratified Retention for Continual Adaptation
Nishit Anand, Ramani Duraiswami and Dinesh Manocha
Continual learning treats degradation on previously seen data as evidence of failure, a convention inherited from settings with a stationary prediction target, where a correct label remains correct indefinitely. World models do not satisfy this condition. Their prediction target is the environment, which changes, so knowledge that was accurate when acquired may later become false, and discarding it is required behavior rather than a defect. Non-stationary ground truth is well studied in the concept drift literature and in the temporal factuality of
language models, but has not been formulated for world models, which are distinctive in that they also encode knowledge that must never be revised. We argue that continual world models require retention stratified by invariance timescale, separating invariants such as physics and object permanence, which must never be revised, from instance-level facts that should be revised as soon as the environment changes. Standard forgetting metrics cannot distinguish a world model that has correctly revised outdated knowledge from one that has suffered catastrophic forgetting, and consequently rank a frozen model highest, while existing physical-
reasoning benchmarks evaluate only frozen checkpoints. We propose differential retention, which reports invariant regression testing across the adaptation stream jointly with revision latency, without aggregation.
Read the full abstract ▾
Artificial Intelligence · Cross-listed from cs.RO
EyeRobot 2.0: Active Gaze for Precise Manipulation without Wrist Cameras
Kush Hari, Justin Kerr, Nidhya Shivakumar et al. (+7 more)
Inspired by human
vision, we introduce a framework using active gaze to enable fine-grained bimanual manipulation with only a single stereo camera. EyeRobot 2.0 physically attends to a 3D fixation point in the scene by swiveling two eye viewpoints to center their gaze on it. The resulting images are processed foveally by allocating more visual
tokens to the image centers, focusing computation on task-relevant features. Such Active Visual Fixation (AVF) requires carefully coordinated gaze during task execution, which we accomplish hierarchically by first training a low-level gaze servoing policy conditioned on a goal object, then training a target selector which emits fixation goals based on task progress. Both modules are trained with RL on real-world data: the first is trained with a dense
geometric reward and the second co-trains with the BC gripper policy which allows it to discover fixation sequences that can resemble a human's fixation sequence while performing the task. EyeRobot 2.0 further takes advantage of fixation by canonicalizing gripper information into a fixation-relative SE(3) frame, which compacts the size of the action distribution to learn. We collect teleoperation data for 7 real-world and 6 simulated tasks, and conduct over 1000 physical and 1800 simulated
robot trials comparing EyeRobot 2.0 against passive stereo and ego + wrist camera policies trained on the same data. Removing wrist cameras is costly for standard policies: with only passive stereo, real-world success drops from 52% to 27%. EyeRobot 2.0 closes this gap with only stereo, outperforming passive stereo by 40% in real and 20% in sim. It matches ego + wrist policies when their wrist views are clear (69% vs. 64%), and more than doubles their success when grasped objects occlude the wrist cameras (48% vs. 22%)
Read the full abstract ▾
All authors Kush Hari, Justin Kerr, Nidhya Shivakumar, Samarth Mahapatra, Carmelo Sferrazza, Jiahui Lei, Jitendra Malik, C. Karen Liu, Ken Goldberg, Angjoo Kanazawa
Artificial Intelligence
Transcriptome-informed multi-modal AI for predicting neoadjuvant therapy response from breast cancer biopsies
Jungkyu Park, Dhruva Biswas, Joseph Cappadona et al. (+22 more)
Scarcity of labeled data limits development of deep learning biomarkers in oncology. We develop a two-stage AI model predicting pathological complete response (pCR) to neoadjuvant therapy in breast cancer. The first stage learns the transcriptome from histopathology using 8,742 patients across 32 cancer types, corroborated by pathologist review and spatial agreement with measured expression. This simplifies the second stage to predicting pCR from inferred expression and clinical variables. Developed using 1,080 patients (five cohorts) and evaluated in 1,412 patients (nine cohorts), the model achieves a pooled AUROC of 0.79 (95% CI, 0.73-0.85), discriminating responders within molecular subtypes. It outperforms histopathological biomarkers, remaining stable across intratumoral sampling and with minimal biopsy tissue. Ablations show transcriptome-wide inference improves discrimination over clinical variables alone or one-stage pathology models, and robustness by avoiding genomic assays' gene selection constraints. These results indicate that biologically informed compression may generalize to data-sparse applications in precision oncology.
Read the full abstract ▾
All authors Jungkyu Park, Dhruva Biswas, Joseph Cappadona, Cerise Tang, Ken G. Zeng, Bartosz Machura, Chuwen Liu, Paolo Tarantino, Coral Omene, Francisco J. Esteva, Rohit Bhargava, Marcin Braun, Kamila Paździerz, Jakub Czerwiński, Hanna Romańska-Knight, Albert Grinshpun, Bareket Daniel, Michele Buchinger, Frederick Howard, Piotr Wysocki, Brie Chun, Freya Schnabel, Rich Caruana, Jan Witowski, Krzysztof J. Geras