跳到正文
研踪PaperTrail

论文库

检索计算机领域论文,按研究方向、作者或关键词筛选。

部分采集异常
正在读取账号与研究记录…

全站更新概况

+729新增论文
+575其中 AI 相关
+82已有论文修订
+95新增 AI 导读

arXiv 最近完成扫描:2026-11-21 12:19 · 全站最近入库:2026-11-21 12:19各信源进度 →

统计口径与数据新鲜度

统计 2026-11-20 16:21 — 2026-11-21 16:21(北京时间) · 点击数字查看对应简报;AI 相关包含在新增论文内。

最近完成检查:2026-11-21 12:19最近发现新增版本:2026-11-21 12:19库内论文最新更新时间:2026-11-20 02:59

每分钟刷新已保存的数据。空检查不代表采集异常;新收录可能包含历史论文。AI 导读按新增版本计数。

29,158 已收录论文24,181 AI 相关8,992 有修订112 HF 社区发现

浏览论文库

时间范围跟随排序:版本更新、本站收录或首次发表。最新收录可能包含历史论文。部分信源异常,查看状态 →

找到 6,976 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-21 12:17(北京时间)
版本更新 · v3arXiv:2603.17673

Towards Reliable Local Security Agents: Verifiable Post-Training for Linux Privilege Escalation

Philipp Normann, Andreas Happe, Jürgen Cito, Daniel Arp

原文摘要

LLM agents are increasingly important in security, yet leading systems are typically closed-source and cloud-hosted, limiting reproducibility and their use with sensitive data. This creates a need for small, local models that can perform security tasks under strict resource constraints, though effective methods for developing them remain underexplored. In this paper, we address this gap by proposing a two-stage post-training recipe that turns a small local language model into a security agent. To this end, we focus on Linux privilege escalation as a representative setting to systematically study the training of local models, as the task is automatically verifiable and requires multi-step interactive reasoning. Using an experimental setup that mitigates benchmark leakage, we post-train a small 4B model in two stages: supervised fine-tuning on traces from procedural privilege-escalation environments, followed by reinforcement learning with verifiable rewards. On a held-out benchmark of 12 Linux privilege-escalation scenarios, supervised fine-tuning nearly doubles the baseline success rate under a tight budget of 20 interaction rounds, and subsequent reinforcement learning improves our model, PrivEsc-LLM 4B, to 93.3% success, ahead of all other evaluated models that fit a single consumer GPU. At the same time, its expected inference cost per successful escalation is more than 80x lower than that of Claude Opus 4.7. Our findings show that small local models can be adapted to complex security tasks. We document the challenges involved and offer guidance for transferring this recipe to other settings.

cs.CR · cs.AI

版本更新 · v3arXiv:2508.20262

Humans and AI respond similarly to task difficulty when reasoning about content moderation dilemmas

Thomas Davidson

原文摘要

Large reasoning models (LRMs) generate intermediate reasoning steps before producing answers, improving performance on challenging problems by iteratively developing solutions. There is evidence that LRMs expend more effort on the problems that take people longer to solve, but it is unclear whether these relationships hold on more subjective tasks without verifiable answers. This study addresses this issue through an experiment using a content moderation task, where task difficulty is manipulated by varying the relative offensiveness of two different options. Comparing six large reasoning models with 1,854 human subjects, we find that models expend more reasoning effort, measured using the length of the chain-of-thought (CoT), on the kinds of cases that take people longer to adjudicate. People took 31\% longer and models produced 51 to 306\% longer CoTs when offensiveness was held constant across the two posts. The correlation is relatively low for individual decisions ($r = 0.15$--$0.19$), but average response times and average reasoning effort are closely correlated when tasks are grouped by the types of offensive language ($r = 0.85$--$0.93$). Qualitative analyses of the CoT summaries from Gemini 2.5 Pro corroborate these findings, showing that secondary factors were mentioned more often when offensiveness was fixed. The results point to the potential for using reasoning traces to prioritize difficult cases and enhance interpretability.

cs.AI · cs.CY

新论文 · v1arXiv:2611.22019

Computed Permutation Trellis Coding: Near-Optimal Quantization from Two Multiplications and a 64-Byte Table

Muhammed Fatih Balın

原文摘要

We propose Computed Permutation Trellis Coding (CPTC), a low-bit quantization format for tensors written once and read many times. Formats that round each value on its own, such as INT4 and FP4, waste a large part of their bits against the information-theoretic ideal. Trellis codes recover this loss by encoding values jointly, but prior decoders either need a large lookup table for their best quality or compute values with a hash that restricts what they can decode. CPTC's decoder multiplies its state by two coefficients and adds the top bits of the products to index a 64-byte table. Because that sum is not uniformly distributed, the rarely used ends of a small table can hold values far out in the tails. One decoder serves every bit width from 1 to 6, reproduces the standard integer and floating-point formats exactly, and comes within 0.07 bits of the ideal at 4 bits. It matches a random-permutation trellis code whose table is thousands of times larger, and exceeds every QTIP codebook. In LLM weight quantization without fine-tuning, CPTC has the lowest KL divergence to the unquantized model at every tested bit width on Llama 3.1 8B and 70B Instruct against QTIP's codebooks (by up to 27%). The decoder has been built into the AWS Trainium 4 AI accelerator, where it runs at the throughput of the native FP4 and FP8 formats for a negligible fraction of the systolic array's area. Fused encoding kernels quantize an 8B model's linear layers in under two minutes on one GPU and, in cycle-level simulation, on one Trainium 4 chip.

cs.IT · cs.AI · cs.AR · cs.LG

新论文 · v1arXiv:2611.22017

SkillBuffBench: Measuring What Agent Skills Contribute across Harnesses, Models, Skill Access Modes and Memory

Gengyue Dong, Ye Wei, Zheng Zhu, Yidong Han, Jiale Miao, Xiaojun Jia, Guanghui Li

原文摘要

Agent Skills, file-based packages of procedural knowledge that a coding agent can discover and load on demand, are now shipped by every major agent harness, yet the evidence for how much they help comes almost entirely from settings in which the right Skill is hand-picked for the task and the agent starts from a blank slate. Neither holds in deployment, where an agent chooses from whatever library happens to be installed and frequently carries memory from earlier work. We present SkillBuffBench, a benchmark and paired evaluation protocol that treats Skill access as a three-level factor (no Skill, a curated Skill, and automatic invocation from an installed library), crosses it with memory (off and on), with four agent harnesses and with multiple models (the state-of-the-art proprietary models from Anthropic, Google and OpenAI, and the state-of-the-art open-weight models from Chinese providers), every model being run under every harness, and evaluates the result on research tasks contributed by domain experts together with the Skills they use, currently in materials science and biology, with the suite extended on a rolling basis. Every run is marked by a deterministic verifier; uncertainty is reported with a paired, task-level bootstrap; and cost, agent rounds and wall-clock time are reported alongside the score rather than folded into it. This version of the paper describes the design of the benchmark, its evaluation protocol and its current tasks and Skills. The evaluation is in progress: results will be added once the runs are complete, and the task suite, results and analysis will continue to be updated on a rolling basis.

cs.AI · cs.CL · cs.LG

版本更新 · v2arXiv:2606.17516

FoundCause: A Foundation Model for Causal Discovery with Latent Confounders from Observational Data

Patrick Blöbaum, Krishnakumar Balasubramanian, Shiva Prasad Kasiviswanathan

原文摘要

Causal discovery from observational data requires recovering directed relations and latent confounding without interventions. We propose FoundCause, an {\em amortized causal discovery foundation model} trained only on synthetic structural causal models that maps datasets to causal graphs in a single forward pass. A permutation-equivariant transformer alternates attention over samples and variables to model cross-variable dependence and per-variable distributions. Pairwise dependence and asymmetry statistics guide its attention, edge decoder and confounder module. The decoder separates edge existence from direction, while triangular refinement incorporates evidence from chains and colliders. Learnable latent tokens predict which variable pairs share a hidden common cause, and a masked input representation handles missing data. To our knowledge, FoundCause is the first amortized causal discovery approach for i.i.d. observational data to predict latent confounding as a separate, explicit output. On 14 real-world and semi-synthetic datasets, FoundCause leads 11 classical and 7 amortized baselines in average AUROC, $F_1$, skeleton $F_1$, and normalized structural Hamming distance ($\mathrm{SHD}/d$): clearly in $\mathrm{SHD}/d$ ($0.875$ vs.\ $0.956$), narrowly in $F_1$ ($0.531$ vs.\ $0.516$). It also has the highest accuracy among amortized methods on bivariate cause--effect pairs.

cs.LG · cs.AI · stat.ME · stat.ML

版本更新 · v2arXiv:2608.28828

Representation Learning with Quantum Signal Processing

Junqi Wang, Junyu Liu

原文摘要

Representation learning begins when training changes the features that define similarity between data. A frozen-kernel model only reweights a fixed geometry. We establish quantum signal processing (QSP) as a solvable quantum model of the representation-learning regime. At arbitrary depth, we compute the exact mean and variance of its quantum neural tangent kernel, revealing an input-dependent angular geometry whose diagonal remains non-self-averaging even when the underlying unitary approaches Haar randomness. We also prove a sparse-data guarantee for the full nonlinear gradient flow without freezing or ensemble-averaging the kernel: the realized dynamics converges to an integrable scalar flow with a time-dependent kernel closure and explicit convergence times. A finite-depth speed limit holds for every data set and trajectory. At higher data density, numerical results show coupled evolution beyond both the scalar and frozen-kernel descriptions. These results give a controlled theory of learned quantum data geometry with provable training dynamics beyond the frozen limit.

quant-ph · cs.AI · cs.LG · stat.ML

版本更新 · v2arXiv:2510.26185

Cross-Epoch Bias in SGD-based Leave-One-Out Influence: Analysis and Accumulative Correction

Yunxiao Shi, Shuo Yang, Yixin Su, Rui Zhang, Min Xu

原文摘要

Training-data influence estimates support data selection and cleansing by predicting the effects of removing individual examples. The standard multi-epoch SGD-Influence Estimator (SGD-IE) aggregates contributions from a sample's repeated appearances along the recorded trajectory. When used to approximate finite deletion, it omits an interaction: later removals act on the parameter displacement accumulated from earlier removals. We introduce the Accumulative SGD-Influence Estimator (ACC-SGD-IE) to account for how earlier removals change the effect of removing the same sample again. It uses the sample's Hessian at subsequent appearances to correct the estimated parameter change. Our non-convex analysis shows that this correction removes SGD-IE's additional interaction contribution from the finite-horizon error bound: the omitted interaction produces a local error linear in the accumulated parameter change, while the remaining Taylor remainder is bounded quadratically under Lipschitz-Hessian assumptions. The bounds characterize how training propagates the remaining error. In non-convex experiments across clean and noisy tabular, text, and vision datasets, ACC-SGD-IE reduces validation-loss-change RMSE by 10.8% and improves Jaccard@10% by 15.2% on average relative to SGD-IE. These gains also translate into lower test misclassification rates after data cleansing on MNIST and CIFAR10 at the evaluated removal budgets. On noisy industrial datasets, it reduces aggregate parameter RMS error with a mean relative reduction of 53.4%. We further discuss computational tradeoffs, provide complementary theoretical analysis and experimental evaluation for strongly convex objectives, and explore potential extensions of the accumulative correction mechanism to other influence estimators.

cs.LG · cs.AI

新论文 · v1arXiv:2611.22010

On Formal Counterfactual Explanations

Alexey Ignatiev, Peter J. Stuckey, Joao Marques-Silva

原文摘要

Counterfactual explanations are often seen as counterfactual conditionals under closest-world semantics, requiring one to reason over all minimally distant counterexamples. In contrast, formal XAI typically relies on feature-based abductive explanations (AXps) and contrastive explanations (CXps), which admit efficient computation and a minimal hitting-set duality. This paper bridges these perspectives by (i) characterizing closest-world counterfactual explanations explicitly as a disjunctive normal form over all minimally distant counterexample instances, and (ii) introducing conjunctive counterfactual explanations (CFXps), a subclass of counterfactual antecedents restricted to conjunctions of feature literals that guarantee prediction change. The paper establishes structural results connecting CXps, AXps and CFXps: every CXp induces an AXp for its witness instance, and every CXp together with a witness AXp induces a CFXp whose support is precisely determined by that AXp. These theoretical results yield a modular approach for computing CFXps using existing AXp/CXp machinery. Experiments on a representative set of benchmarks using tree ensembles and neural networks show that CFXp extraction is practically efficient. Compared with popular point-based counterfactual explainers, CFXps generally require no more feature changes and can guarantee the counterfactual prediction beyond a single instance.

cs.AI · cs.LG · cs.LO

新论文 · v1arXiv:2611.22009

DiagTrace: An Interactive Benchmark for Clinical Decision Making

Yufan Wang, Rui Yang, Yi Lin, Chunhua Weng, Yifan Peng

原文摘要

Large language models (LLMs) have shown strong performance on medical question answering and diagnostic reasoning benchmarks. However, existing evaluations typically provide all patient information upfront and assess only the final diagnosis or rationale. This setup ignores the need to acquire diagnostic evidence incrementally from incomplete information in real-world diagnosis. To bridge this gap, we introduce DiagTrace, a source-grounded benchmark for evaluating whether LLMs can reconstruct the documented diagnostic trajectory through interactive information seeking. Each benchmark instance is built from a PubMed Central case report and includes an initial presentation, an event bank, a clinical course, and diagnostic references. DiagTrace contains 510 cases across eight specialties, with 225 used to evaluate four LLMs on clinical-course alignment, interaction quality, and efficiency. Results reveal trade-offs in coverage, source-consistent ordering, and turn use.

cs.CL · cs.AI

版本更新 · v2arXiv:2605.09238

iMuon: Intrinsic Spectral Optimization on Riemannian Matrix Manifolds

Yibang Li, Bihari Lal Pandey, Ravi Sah, Andi Han, Cyrus Mostajeran, Pratik Jawanpuria, Bamdev Mishra

原文摘要

Muon and related norm-constrained optimizers, now central to large-scale learning, compute updates via a linear maximization oracle (LMO) over a matrix-norm ball in Euclidean space. However, they do not generalize cleanly to manifold-valued parameters such as low-rank factorizations, orthogonality constraints, or symmetric positive-definite (SPD) matrices. Naively restricting the Muon LMO to the tangent space (i) breaks quotient symmetries and (ii) couples the tangent-space constraint with an ambient norm bound, which obstructs closed-form solutions. We resolve both issues with a key observation: every Riemannian metric canonically lifts a unitarily invariant Euclidean norm to an intrinsic norm on each tangent space. The resulting intrinsic LMO is symmetry preserving and, under a structural condition on the scaled tangent space, reduces to a standard Euclidean LMO. Building on this, we introduce intrinsic Muon (iMuon), a unified framework with closed-form updates on the fixed-rank, SPD, Stiefel, and Grassmann manifolds for various unitarily invariant norms, including the spectral, Frobenius, and nuclear norms. We establish deterministic and stochastic convergence rates of iMuon with constants that depend only on the manifold dimension. Notably, on the fixed-rank manifold, the constant depends only on the rank, which makes the rate independent of factor conditioning and removes the runtime factor rescaling required by prior work. Experiments on LoRA fine-tuning of LLMs, image classification, and subspace learning illustrate the efficacy of the proposed approach.

cs.LG · cs.AI

新论文 · v1arXiv:2611.21998

KV-LTM: Persistent Per-Session KV Caches as Long-Term Memory for Personal Assistants

Ziv Kfir, Ido Kfir

原文摘要

Personal assistants accumulate months of conversations, but today's long-term memory stacks put derivation or selection between a question and its evidence. In a system-level comparison on LongMemEval, two commercial assistants scored well below the same model reading the history offline [1], consistent with, but not isolating, content loss between write and read. We propose KV-LTM: a memory tier storing each session's raw text plus its KV cache under a fixed reader model, answering by restoring cached sessions and appending the question. KV-LTM is complementary to, not a replacement for, retrieval and indexing. Text is the memory of record; the cache is a recompute-free prefill accelerator, degrading gracefully to text. We specify the design: storage record, budget-derived retention, append-only read path, and bounded aggregation. We size it analytically: across pinned BF16 full-attention (GQA) and latent-cached MLA configurations, per-token state spans approximately 31K-328K bytes (the pinned KDA+MLA hybrid differs: approximately 8.1K bytes/token plus a fixed per-session recurrent term), giving approximately 0.34-3.6 TB per user-year at a declared 30K-token/day scenario, against approximately 49 MB of raw text; per-query cost is modelled as regime-conditional transfer, attention, decode, aggregation, and compute. We define an evaluation protocol needing no persistent-cache implementation for logical QA evaluation: per-session prefill emulates valid cache reuse under the same model/input-state contract and ideal arithmetic, with named arms, controlled baselines, and a preregistered reader-model gate; a separate minimal real-cache replay tests numerical equivalence and measures the cost axis, with outcomes not assumed. KV-LTM trades storage for avoided re-prefill compute and write-time fidelity, within a declared retention budget.

cs.AI

版本更新 · v2arXiv:2609.36208

Representable but Unlearned: Encoding Rank and the Interaction-Prediction Floor

Zahra Khodagholi, Niloofar Yousefi

原文摘要

Input encodings can prevent a predictor from reproducing measured interactions jointly, even when each interaction is individually representable. We compute the attainable contrast space from the encoder and contrast design, without labels or predictor fitting. Projecting measured contrasts onto this space gives an empirical squared-error floor for any deterministic decoder using the encoding. On an siRNA panel, a graph neural network's training-only feature mask merges 165 measured states into 90 classes, reducing the rank of 140 interaction contrasts to 72. The floor is 0.009980, or 14.63% of recorded contrast energy, the mean squared error of predicting zero for every contrast. Its 14.6% share of fitted interaction error adds little: predicted interactions are nearly zero, and their MSE of 0.068335 slightly exceeds that of the same control. Antisense-marginal shares are larger: 36.49% of recorded energy and 38.1% of fitted error. Three restored chemistry columns suffice for full rank. Removing the full mask eliminates the floor, but refitting improves interaction MSE by only 0.000017; the restored features remain absent from training inputs. Two further panels clarify the diagnostic's scope. A released splicing predictor encodes injectively and retains the full design rank of 1,986, so its floor is zero. For an independent count-fingerprint encoder on a lipid panel, 800 pairwise-distinct encoded contrast rows have rank 740, exposing 60 restrictions missed by duplicate-row checks. The diagnostic identifies encoding-imposed limits and can trace them to specific features. It separates representability from learned accuracy, but does not by itself predict refitting gains or explain the remaining error.

cs.LG · cs.AI

新论文 · v1arXiv:2611.21980

Modeling Document and Section Hierarchies in Embedding Space

Kriste Krstovski, Sharyn O'Halloran, Kshitij Chandna

原文摘要

Embedded topic models integrate word embeddings into the topic generation process, producing coherent topics and robust performance on large vocabularies. However, each document is represented by a single topic distribution, ignoring its internal structure. Many real world collections, including legal documents, company and government reports, and online discussion forums, exhibit hierarchical organization in which documents consist of multiple sections that often address distinct subtopics. We introduce the Hierarchical Embedded Topic Model (hETM), a structured topic model that couples the topic distribution of a document and those of its sections through a hierarchical prior in the embedding space. We derive an amortized variational inference procedure for hETM and evaluate it on structured collections from each of these domains. Experimental results show that hETM achieves the lowest or near-lowest held-out perplexity among flat and hierarchical topic model baselines, along with comparable or higher topic quality. It also performs strongly on a downstream clustering task, achieving the best clustering performance on most of the collections.

stat.ML · cs.AI · cs.CL · cs.LG

版本更新 · v3arXiv:2609.38166

LeapQuant: Efficient Linear Attention with Accurate Recurrent State Quantization

Yi Pan, Haocheng Xi, Kan Zhu, Xingyang Li, Yibo Wu, Mayank Mishra, Hongtao Zhang, William X. Zheng 等 13 位作者

原文摘要

Recent LLMs increasingly adopt hybrid designs that replace standard attention with linear attention, such as Gated DeltaNet (GDN) and Kimi Delta Attention (KDA). Although they compress the context into a fixed-size recurrent state and substantially reduce the cost of long-context processing, repeatedly reading and updating that state remains a major inference bottleneck. Quantization offers a natural way to reduce this cost, but can significantly degrade model quality, due to the accumulation of rounding errors and the presence of outlier rows and columns in the state. To address these challenges, we propose LeapQuant, a training-free method that achieves near-lossless performance under 8-bit recurrent-state quantization. First, to mitigate error accumulation, we propose per-window quantization, which leaps over a window of tokens and quantizes the state only once at its end. Within a window, outputs are computed from the fixed low-bit state together with high-precision buffered updates. Second, to reduce the error introduced by each quantization, LeapQuant retains the state's largest outliers as a few high-precision Compensator Tokens, which share the update path of real tokens. We then smooth the remaining residual before quantization to further reduce the error. Comprehensive experiments across the Qwen, Kimi, and GLM model families show that LeapQuant substantially reduces memory and compute costs during inference. With accuracy comparable to the FP32 baseline, it achieves average speedups of 2.05--3.70$\times$ at the kernel level and 1.47$\times$ for end-to-end inference on NVIDIA B200, RTX PRO 6000, and RTX 5090 GPUs.

cs.LG · cs.AI

新论文 · v1arXiv:2611.21978

WebAgents: Unbundling Agent Creation, Hosting, and Execution for an Agent-Native Web

Dario Fenoglio, Martin Gjoreski, Marc Langheinrich, Pradyumna Chari, Ramesh Raskar

原文摘要

Large language models are moving from isolated assistants toward autonomous agents that can plan, use tools, and coordinate with others, motivating an emerging Agentic Web of independently developed agents and services. Realizing this vision requires agent behavior to move across creators, hosts, and runtimes without coupling publication to a particular execution environment or transferring authority over user resources. We introduce WebAgents, which unbundles these roles through a signed agent.pkg that carries portable workflows, skills, schemas, and resource requirements. An OpenClaw-based agent-native browser/runtime verifies the package, obtains authorization, and binds it at run time to user-controlled models, tools, permissions, and budgets, while hosts distribute the public program and may retain private services server-side. We demonstrate this contract with three unrelated workflows executed unchanged under local and API-backed models, including literature review with live services and a website-published travel agent using semantic tools and user-selected personal resources. In a controlled suite, WebAgents explicitly blocks all 716 injected contract violations while accepting all 295 matched benign controls, with median automated pre-execution processing below 0.5 s for all three packages. We conclude by outlining a path toward NANDA-based discovery and interoperable package ecosystems while retaining execution authority in the user's runtime.

cs.AI

新论文 · v1arXiv:2611.21975

Pathology Randomization for Synthetic Priors and Zero-Shot Detection

Alexandra G. Roberts, Tonmoy Hossain, Mark Olchanyi, Stefan Brandstätter, Jonathan Williams Ramirez, Ana Lawry Aguila, Matthew Rosen, Annabel Sorby-Adams 等 10 位作者

原文摘要

Hospital brain imaging archives hold a diverse array of pathology at sample sizes beyond prospective studies, particularly for underrepresented populations and diseases. To unlock uncurated clinical data and scale analysis, methods are needed to detect abnormalities across contrasts and resolutions. Synthetic data has provided robust, contrast-agnostic models for a variety of human neuroimaging tasks in healthy anatomy, but pathology detection and other tasks in clinical settings still require real patient data, contrast-specific models, or disease-specific generators. We propose PathoDraw, a generative model of brain MRI with pathology. A conditional distribution of lesions given the healthy anatomy is drawn into the label map before contrast-agnostic rendering, and a single network estimates healthy anatomy with exact signed intensity deviation of any input scan. The contrast-agnostic 3D UNet learns the healthy labels under the lesion and the signed deviation directly and identifies pathology via robust tissue statistics and learned features, despite no training on real pathology. We show zero-shot detection on public stroke, tumor, and white matter hyperintensity datasets as well as archived clinical data, spanning $T_1$w, FLAIR and $T_2$w at various resolutions under a recent unsupervised anomaly detection benchmark protocol. Our single model exceeds or performs comparably to contrast-specific anomaly detectors trained on real scans and keeps healthy anatomy intact. The same generator trains networks for super-resolution and for registration in the presence of pathology. Our model, robust across acquisition and disease, enables the use of uncurated, archived data for clinical analysis. Our code is available at https://github.com/agr78/Pathology-Randomizer.

cs.CV · cs.AI · eess.IV

新论文 · v1arXiv:2611.21959

OCCAM: Improved Single-Cell Representation Learning and Perturbation Effect Prediction

Ihab Bendidi, Kian Kenyon-Dean, Alisandra K. Denton, Auguste Genovesio, Emmanuel Noutahi

原文摘要

Predicting the effects of genetic perturbations requires models that generalize beyond observed cell contexts and gene combinations. We introduce Occam, a multilayer perceptron (MLP) framework for representation learning and perturbation-effect prediction. Occam combines cell representations learned through denoising with gene interaction, text, and protein-sequence features. It achieves the highest aggregate representation scores for perturbation information and cell-type preservation with batch integration. In cross-context few-shot prediction, a perturbation and a cell context are each observed during training, but never together. Occam better captures the measured response patterns for these unobserved combinations than four retrained competitors or direct transfer of the same perturbation's measured effects from other training cell lines. Without fine-tuning, Occam best distinguishes perturbations among the evaluated models and simple baselines in cell lines held out from predictor training, across seen and unseen assays. Predicting expression levels accurately in these settings remains challenging. For double perturbations, which target two genes simultaneously, Occam composes learned single-gene responses to better capture the combined response and reduce expression error compared with competing models and summed single-gene effects. These gains extend to both strongly additive and weakly additive responses. Together, these results support simple MLPs for cell representation learning and perturbation-effect prediction across contexts and gene combinations.

q-bio.QM · cs.AI · cs.LG

新论文 · v1arXiv:2611.21958

Weight-space programming for tractor guidance controller: large language models natively write neural network weights that steer a real tractor

Timo Oksanen

原文摘要

Weight-space programming is a process where a large language model reads and writes the weights of a neural network directly, guided by a natural language specification, without an optimiser or gradient descent. This study applies it to automatic tractor guidance, where a 253-parameter neural network steers along paths of lines and arcs and must be adapted when the machine changes, such as a slower steering actuator. GPT-6 Astra and Claude Fable 5.1 (also 5.2) edited the network for our research tractor using a small budget of simulated evaluations to repair a network trained for another tractor and write a controller into a network with zero weights. We compared them with an evolution strategy, gradient descent and pure pursuit in simulation and on the real tractor. Both models read the network and analysed the closed loop with control theory correctly, and Fable's repair without simulation succeeded only occasionally. With simulated feedback, both models repaired the trained network in single sessions and wrote a working controller from zero weights, whereas from zero the evolution strategy found none. The controllers written from zero were sparse and in a classical cascaded structure; one model edited the trained network while the other replaced it. Gradient descent reached higher accuracy with far more simulation, and all language model controllers steered the real tractor without intervention, with errors of the same order as pure pursuit. Language models can program the weights of a neural controller directly, much as control engineers design one. We call this Weight-Space Engineering.

cs.LG · cs.AI · cs.RO · eess.SY

新论文 · v1arXiv:2611.21956

Eight in Ten LLMs Leak System-Prompt Secrets: Nevan, an Agentic Capture-the-Flag Study

Ana Clara Zoppi Serpa, Anderson Rocha, Hélio Pedrini

原文摘要

We audit instruction-hierarchy enforcement across 363 text-in/text-out models from the OpenRouter catalog. The research question we answer with this experiment is which models disclose a protected secret embedded in the system prompt, and through which techniques. Each model guards a secret 16-character flag set in its system prompt, while an artificial intelligence (AI) agent (OpenCode, powered by DeepSeek V4.1 Flash), steered by a human researcher, tries to persuade it to emit the flag. The flag was exfiltrated from 296 of the 363 models, plus four models released after the initial catalog freeze, for 300 captures in total. Of these, 170 involved a non-English language pivot, most often French. Each capture shows that a model ranked a user request above its system instruction at least once. It does not show that the model provably resists untested attacks, nor that it would consistently leak semantically sensitive data, nor that it is vulnerable to harmful-content jailbreaks. For practitioners, we draw three lessons: do not store secrets in system prompts, enforce critical behavior outside the model with output validation and deterministic checks, and do not let a large language model (LLM) judge be the sole decider of consequential outcomes.

cs.CR · cs.AI

新论文 · v1arXiv:2611.21954

ThermLeT: Transformer-based Temperature Prediction for 2.5D Chiplet Architecture

Varun Darshana Parekh, Anusha Devulapally, Sivani Devarapalli, Cassius Henderson, Shimeng Yu, Vijaykrishnan Narayanan

原文摘要

The drive toward heterogeneous 2.5D chiplet packages promises unmatched performance and customization, yet the resulting density amplifies thermal challenges. To facilitate rapid design space exploration, we present ThermLeT, a multimodal transformer-based, scale invariant surrogate that integrates power density maps, spatial coordinate embeddings, material properties and heat transfer rate to predict steady state temperature in 13 ms, about four orders of magnitude faster than conventional finite element solvers such as ANSYS. To train the transformer model, we also create a dataset of 6,221 chiplet layouts perturbed from three representative architectures. On a 150 sample test set, ThermLeT achieves an RMSE of 3.16°C and an MAE of 2.24°C, improving accuracy by 35% over ThermEDGe and by 71% over HotSpot, while running approximately 6x and 11x lower inference latencies respectively. Ablation studies confirm robust generalization to unseen floorplans. Code and dataset available at https://github.com/MicroSystemsDesignLab/ThermLeT

cs.AR · cs.AI · cs.LG

新论文 · v1arXiv:2611.21950

DiscoveryAgent: A Framework for Scientist-centric Agentic AI

Owen Queen, Ioana Ciuca, Emma Dann, Jean Somalwar, Carine You, Steven Dillmann, Viraj Manwadkar, Evan Gorrell 等 12 位作者

原文摘要

AI agents can assist scientific research, but balancing autonomous investigation with scientists' ability to inspect and redirect ongoing work remains a central design challenge. We present DiscoveryAgent, a domain-agnostic, scientist-centric system that combines autonomous investigation with scientist-directed hypothesis steering. Scientists select, refine, or replace hypotheses between research rounds, while audit agents interrogate checkpointed records to trace claims, reproduce analyses, and inform subsequent decisions. Across two predictive modeling tasks, DiscoveryAgent produced the best variable-star classifier among evaluated baselines and, with scientist steering, a protein variant-effect predictor surpassed by only one of 106 ProteinGym baselines. We develop a rigorous checkpointing system that enables highly capable audit agents, a scalable form of oversight for agentic research. Audit agents autonomously identify errors at higher rates than previous AI scientist agents, and scientists collaborating with audit agents identify 14x more valid errors versus scientists performing manual review. In three case studies led by domain experts across physical science, social science, and life science, we developed a multimodal photometric-redshift estimator that resolves complementary failures of images and photometry, achieving 0.971 AUROC for low-redshift classification; documented an 83% rise in Florida homeowners-insurance premiums from 2013 to 2023, with a growing, regressive burden; and identified CD4+ T-cell relay genes, which both receive and exert broad transcriptional regulation, enriched for autoimmune disease associations. Together, these results show how autonomous research can expand scientists' empirical reach while centering their authority over research direction and their ability to understand and challenge the evidence.

cs.AI

新论文 · v1arXiv:2611.21948

Honest Brokers in AI Regulation: How Public Comments Introduce Evidence to Regulators

Sarah Tran, Anna-Maria Gueorguieva, Eva Maxfield Brown, Nicholas Weber

原文摘要

The regulation of Artificial Intelligence is susceptible to various forms of capture by industry actors that hold undue influence over the policymaking process, due in part to their information advantage in developing these technologies. In the following paper, we investigate the roles of public commenters in providing evidence to potential regulation in the USA. We build a corpus of 40 potential regulations, and 2,664 public comments submitted by a variety of stakeholders. We computationally process these texts to extract evidence meant to inform regulators, in the form of formal citations and informal references. We find significant differences in the type and amount of evidence supplied: Holding comment length constant, business commenters introduce altogether less evidence, and government, university, and civil society commenters provide significantly more research and legal sources. The publication status of cited research is nearly evenly split, with preprints and peer-reviewed articles being introduced by commenters of all types. We further explore patterns of self-citation and find that while it is prevalent amongst all groups, a small cluster of organizations account for over half of all self-citations in our corpus. Our findings complicate a narrative that public comments alone are means to information capture. We interpret these results in terms of the role that commenters attempt to play in the policymaking process-being either an Issue Advocate or an Honest Broker. These findings on the supply of evidence in public comments shed light on an often overlooked form of policymaking that has implications for how and through what sources information is made accessible to regulators of AI.

cs.AI

新论文 · v1arXiv:2611.21946

Conserved Verdicts, Drifting Commitments: Identifying Harmful Overthinking

Jianghong Zhou, Hao Rong, Chen Lin

原文摘要

A reasoning model that has reached a correct answer can lose it by continuing to think. Deciding whether to stop means comparing two probabilities at the same point of a chain: that the answer the model would give if stopped now is correct (its commitment), and that the answer it will finish with is correct (its verdict). We show that the verdict alone, which is what final-correctness probes estimate, cannot make this comparison. As a forecast of a fixed outcome it has no predictable drift, whether continuing helps or hurts. The commitment-verdict gap equals the expected remaining change in answer quality and the accuracy gained by stopping. A second identity splits any accuracy-versus-budget curve into answer change within chains and change in which chains are still running. We measure both on MATH-500. At all 379 points where DeepSeek-R1-Distill-Qwen-1.5B holds a correct answer, selected without regard to how its chain ends, continuing loses 3.6 points on average (95% CI [2.2, 5.2]), and the loss concentrates in a few problems. In six reasoning models from three families, on MATH-500 and OlympiadBench and with budgets up to 16,384 tokens, declining accuracy curves come from which chains remain, while the chains still reasoning keep or improve their answers. An oracle stopping rule gains at most +1.8 and +2.5 points at 1.5B and 7B. The framework says what to measure before building a stopping rule.

cs.AI · cs.CL · cs.LG

新论文 · v1arXiv:2611.21945

Benchmarking Deep Learning Models for Wheelchair Configuration Classification in Accessible Automated Vehicle Applications

Samir Ahmed, Shaaz Saleem, Devin Dhami, Wesley Maia, Angel Martinez-Sanchez, Parthib Roy, Jannatul Ferdous Srabonee, Ahmed Sabbir Arif 等 9 位作者

原文摘要

Making automated vehicles accessible requires consideration of vulnerable road users, including wheelchair users. To offer a more tailored experience that supports independence for this group, vehicle design needs to accommodate the wide range of wheelchair configurations. However, wheelchairs are individualized, varying in frame geometry, propulsion, and visual appearance. Moreover, occupants or accessories may obscure visual cues. Existing mobility-aid datasets only label the presence of a wheelchair, rather than describing its specific configuration. To fill this gap, this paper introduces the Wheelchair Configuration Image Dataset (WCID), comprising 467 face-blurred images across 192 capture groups solely for evaluation. Each image is annotated with (i) a two-class label of manual or powered, (ii) a chair subtype, (iii) occupancy, viewpoint, setting, lighting, occlusion, propulsion, and control, with a 273-image validation split and 194-image test split. Three convolutional neural network backbones and a frozen vision-language probe are trained on an external image dataset and evaluated on WCID to measure cross-dataset generalization, while also testing robustness using Gaussian blur. Convolutional models do well on clean images, but performance drops under strong blur, leaving a gap before deployment. Code, split assignments, labels, trained weights, and the face-blurred images is available at: https://github.com/Mi3-Lab/MAID

cs.CV · cs.AI · cs.LG · cs.RO

版本更新 · v4arXiv:2602.16736

The Compute ICE-AGE: OPAL ONE, Continuity That Persists While Reasoning Is Idle

R. Jay Martin

原文摘要

The Compute ICE-AGE separates persistent application memory from probabilistic reasoning. We present OPAL, a CPU-resident C++17 graph engine that preserves record identities, relationships, and accepted numerical state between reasoning sessions. Applications evolve that state through defined updates, save canonical checkpoints, and recover continuity after restart. Bounded Local Generator Classes (BLGC) provide the mathematical foundation for this bounded local evolution. On Apple Silicon M2 Pro hardware, the current-engine resident-update P50 measured 0.083 microseconds at 100,000, 1,000,000, and 5,000,000 retained Atomic Memory Vertices (AMVs). A same-session 1M-5M-1M sequence reproduced that median using one frozen binary and 1,024 prepared targets, with every process passing its correctness checks. These measurements cover complete resident-update calls, including validation, indexed lookup, and numerical state mutation. Initialization and storage-growth costs are reported separately. Reliability tests verified event replay, conflict rejection, concurrent calls, allocation-failure handling, and fresh-process recovery with continued operation. Engine-version comparisons produced byte-identical checkpoints and continuation files. The installed PYAMV Python package passed 17 SDK contract checks and 20 workload checks. KINDRED's Model Context Protocol interface passed 22 distinct focused checks covering memory operations, authorization, and transport. Together, these results demonstrate persistent continuity with low recurring local-update costs through five million retained AMVs, providing working interfaces for applications and developers. The architecture keeps accepted memory available while reasoning is idle, allowing applications to invoke models only when tasks require interpretation, planning, or generation.

cs.OS · cs.AI

新论文 · v1arXiv:2611.21940

Does the Model Use What It Cites? Counterfactual Tests of Grounded Reasoning Chains in Small Video-Language Models

Jugal Gajjar, Kamalasankari Subramaniakuppusamy

原文摘要

A grounded reasoning chain cites a frame span and a box for every step, so a reader can check where the model looked. Whether the answer depends on the cited region is a different question. We test it by intervention: the regions a model cited in its own chain are blacked out, and the share of correct answers that flip is compared with blacking out boxes of the same size elsewhere on the same frames. The paired difference, the cite gap, needs no annotation. We apply it to 2-3B video-language models trained on exact CLEVRER chains and on pseudo-labelled NExT-QA chains. With the same weights, the cite gap is 0.089 on CLEVRER and 0.003 on NExT-QA. Pooled over models, masking the cited boxes flips 2.2 times as many correct answers as the control on CLEVRER and 1.1 times as many on NExT-QA, also among the answers that depend on the video at all, while the boxes pass a grounding check on both. Masking the cited entity on every frame, instead of the cited box, flips 1.9 times as many answers as its control. The NExT-QA chains were supervised to cite two frames of entities that stay visible on most frames, so this outcome is what the supervision predicts; the point is that the usual check on grounded chains cannot see it. A citation can be verifiable and relevant without the answer depending on it, and a test of grounded reasoning should remove what a chain names, not only where it points.

cs.CV · cs.AI · cs.LG

版本更新 · v2arXiv:2605.25267

Latent Q-Barrier Shielding for Safe In-Context Reinforcement Learning

Minjae Kwon, Amir Moeini, Shangtong Zhang, Lu Feng

原文摘要

Safe in-context reinforcement learning (ICRL) adapts online from interaction history without test-time parameter updates while controlling episode cost under a safety budget. Under out-of-distribution (OOD) deployment shifts, pretraining-only safe ICRL can give poor reward-safety tradeoffs because the remaining budget affects behavior only through frozen policy conditioning, not an explicit action-level check against predicted future cost. We propose a latent Q-Barrier shield that learns a context representation, latent dynamics, and an ensemble cost critic before deployment. Without parameter updates, the shield infers context from history and filters or softly reweights candidate actions using the remaining budget and predicted future cost. We prove a conditional one-step barrier-margin bound over realized candidate sets, with explicit Bellman and latent-prediction error terms. Across five safe ICRL benchmarks, the shield improves deployment-time reward-safety tradeoffs over a strong safe-ICRL baseline: after a short context window, it achieves higher return in four of five benchmarks while matching or lowering average episode cost in all five. Code is available at https://github.com/kmj1122/q-barrier.

cs.LG · cs.AI

版本更新 · v2arXiv:2602.07787

Do Multi-Agents Dream of Electric Screens? Achieving 100% on AndroidWorld Through Task Decomposition

Pierre-Louis Favreau, Jean-Pierre Lo, Clement Guiguet, Charles Simon-Meunier, Nicolas Dehandschoewercker, Allen G. Roush, Judah Goldfeder, Ravid Shwartz-Ziv

原文摘要

We present Minitap's mobile-use, a multi-agent system for Android UI control, and evaluate it on the 116 tasks of AndroidWorld, on Android emulators in Minitap's cloud device infrastructure, real Pixel 7 phones from our device farm and the stock emulator. On fresh, held-out task instances (three pre-registered seeds never used during development), the final system reaches a pass@3 of 100.0% (95% CI 96.8-100.0); pass^3 is 94.0% (95% CI 89.7-97.4) and the mean success rate 98.0% (95% CI 96.0-99.7). Minitap decomposes each task across five LLM agents (planning, subgoal tracking, state capture, decision, execution) and deterministic nodes, with mechanisms that read typed text back from the device, verify the goal on device evidence and detect repeated failures. A study of the initial system measures what each part contributes: removing only the decomposition changes success by -26.1 [-35.6, -16.7] points, the mechanisms together add +11.7 [+2.2, +21.1] points over a single agent, and the decision agent is the one role that needs a frontier model. We then improve reliability with general changes, a few of them triggered by the goal's wording. On MobileMiniWoB++, a second benchmark never used during development, the final system improves on the initial one by 7.1 points (McNemar p = 0.035; 95% CI -0.5 to +14.7). Minitap is open source: https://github.com/minitap-ai/mobile-use

cs.AI

版本更新 · v4arXiv:2610.37690

Cassette-Editor: Shared Timeline Semantics for Human and Agent Editing

Zhixin Zeng, Jiahao Zhong

原文摘要

Video editing involves tedious manual manipulation across interconnected timeline tracks, motivating automated agents that offload routine work while keeping the project directly revisable by humans. Fixed operation catalogs limit expressiveness, while even a basic trim requires coordinated updates across later clips, attached titles, and transitions. We present Cassette-Editor, an agent harness that unifies manual user gestures and automated agent patches within a shared editing environment. The shared action language expresses the editor's manual content changes, and a common history records committed edits for undo and manual refinement. When the model selects an edit, a state executor automatically derives its downstream consequences, such as shifting later clips and attached titles after a trim, while a validation rulebook enforces structural invariants and returns localized diagnostics for automated repair. To keep the agent grounded as the project evolves, dynamic read operations map transcripts and shot descriptions onto the current timeline. Evaluated with GPT-6 Luna, the full harness passed the scored request checks in 92/93 trials across 31 authored requests on two interview documentaries, compared with 48/93 for an internal basic-tools baseline. In ablations, disabling the consequences derived by the executor reduced passed trials from 230/232 to 199/232 on edits with dependent timeline elements, while ablating rulebook validation and repair lowered passed trials from 46/48 to 10/48 on requests operating near structural validity boundaries. Evaluating the complete system on 50 TVSum videos further demonstrated greater agreement with human annotations than a general agent using the same model (F1 23.8 vs. 20.1).

cs.MM · cs.AI

版本更新 · v4arXiv:2411.10109

AI Simulations Grounded in Rich Self-Reports Increase Accuracy and Reduce Distortion

Joon Sung Park, Carolyn Q. Zou, Jonne Kamphorst, Niles Egan, Aaron Shaw, Benjamin Mako Hill, Carrie Cai, Meredith Ringel Morris 等 11 位作者

原文摘要

AI simulations of people built from demographic prompts often misjudge individuals and distort sample-level patterns. We test whether rich self-report data reduce these problems. We developed an AI voice interviewer and conducted two-hour interviews following the American Voices Project protocol with 1,052 U.S. adults, who also completed surveys and behavioral experiments. For each participant, we compared simulations prompted with demographic attributes, self-authored personas, survey responses, interview responses, or combined surveys and interviews. We evaluated accuracy on held-out responses, and measured eight statistical distortions at the sample level: biased averages, compressed variance, distorted response distributions, over-correlated items, attitudes collapsing onto one dimension, ideological sorting, amplified experimental effects, and hyper-rationality. On General Social Survey (GSS) items, demographic simulation led to 60.4% accuracy, whereas interview and survey simulation achieved 69.4% and 69.3%, and combining them reached 72.8%, compared with 79.5% for participants' two-week test-retest agreement. Surveys and interviews also improved prediction of personality traits, but not of incentivized economic choices. Combined interview and survey simulation closed a median of 53.5% of the eight distortion gaps between demographic simulation and participant responses. An open-weight model fine-tuned on other participants' self-reports and responses to the target questions achieved 75.2% GSS accuracy and closed a median of 69.3% of the distortion gap. Overall, both frontier and fine-tuned models reduced distortions without erasing them, but neither reduced hyper-rationality in economic games. These findings demonstrate that qualitative and quantitative self-reports can improve person-specific simulations and reduce statistical distortions.

cs.AI · cs.HC · cs.LG