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

+479新增论文
+361其中 AI 相关
+63已有论文修订
+93新增 AI 导读

arXiv 最近完成扫描:2026-11-16 01:34 · 全站最近入库:2026-11-16 13:44各信源进度 →

统计口径与数据新鲜度

统计 2026-11-15 16:58 — 2026-11-16 16:58(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:2026-11-16 01:34最近发现新增版本:2026-11-16 13:44库内论文最新更新时间:2026-11-14 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。

23,827 已收录论文19,941 AI 相关6,921 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 23,827 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-16 01:33(北京时间)
版本更新 · v4arXiv:2512.02581

Generative Online Reinforcement Learning

Chubin Zhang, Zhenglin Wan, Feng Chen, Fuchao Yang, Lang Feng, Yaxin Zhou, Xingrui Yu, Yang You 等 10 位作者

原文摘要

Reinforcement learning (RL) faces a persistent tension: policies that are stable to optimize (e.g., Gaussians) are often too simple to represent the multimodal action distributions required for complex control. Conversely, expressive generative policies---such as diffusion and flow matching---can be difficult to optimize in online RL due to intractable likelihoods and gradients propagating through long sampling chains. We address this tension with a key structural principle: decoupling optimization from generation. Building on this, we introduce GoRL (Generative Online Reinforcement Learning), an algorithm-agnostic framework that trains expressive policies from scratch by confining policy optimization to a tractable latent space while delegating action synthesis to a conditional generative decoder. Using a two-timescale alternating schedule and anchoring decoder refinement to a fixed prior, GoRL enables stable optimization while continuously expanding expressiveness. Empirically, GoRL consistently outperforms unimodal and generative baselines across diverse continuous-control tasks. Notably, GoRL achieves returns exceeding 870 on HopperStand, more than $3\times$ the strongest baseline; on high-dimensional humanoid tasks, it further outperforms the strongest non-GoRL baseline by over an order of magnitude. Code is available at https://github.com/bennidict23/GoRL.

cs.LG

版本更新 · v3arXiv:2609.01474

Binary Multiple-Node-Erasure-Correcting Codes over Complete Graphs: Constructions, q-Ary Metric Balls, and Duality

Aryeh Lev Zabokritskiy

原文摘要

Codes over graphs store symbols on edges of a complete undirected graph on $n$ vertices with a self-loop at each vertex; erasing a node removes all its incident edges. Extending Yohananov, Efron, and Yaakobi's construction, we obtain binary codes correcting any three node erasures at infinitely many prime lengths $n$, unconditionally, with redundancy $3n-2$, one bit above the graph Singleton bound. The earlier construction has the same redundancy but requires $2$ to be primitive modulo $n$, which is not known for infinitely many primes. We give a deterministic decoder for these triple-node codes using $O(n^2)$ binary operations and construct Singleton-optimal binary triple-node codes at $n=6,8,10,12$. When $2$ is primitive modulo an odd prime $n$, we extend their double- and triple-node codes to binary codes correcting $ρ$ node erasures with redundancy $ρn-(ρ-1)$ for $2\leqρ\leq(n+1)/2$, saving $ρ-1$ bits over the binary symmetric rank-metric construction when $ρ<(n+1)/2$. Over every finite field of size $q$, the node weight of a word is the least number of vertices whose erasure removes its support. We give exact formulas for ball volumes in this node metric and prove that, for fixed $q,t$, the radius-$t$ volume is asymptotic to $\binom nt q^{tn-\binom t2}$ as $n\to\infty$. We obtain existence and covering bounds; for fixed $q,ρ$, we show that linear $ρ$-node-erasure-correcting codes exist with redundancy at most $ρ\log_q n+O_{q,ρ}(1)$ above the graph Singleton bound. Finally, reformulating Yohananov and Yaakobi's criterion, we show that a linear code over any finite field is Singleton-optimal for $ρ$ node erasures if and only if its dual is Singleton-optimal for erasures of the coordinates inside any $n-ρ$ vertices.

cs.IT · math.CO

版本更新 · v2arXiv:2609.40063

LARC: Low-Rank Adaptive Residual Connections for Learning in Frozen Models

Junyi Zou, Avrova Donz

原文摘要

Low-Rank Adaptive Residual Connections (LARC) give a frozen model a compact numerical state that can learn from feedback. The map $h+BAh$ changes the hidden input to a pretrained mixer. An episode copies a learned initialization $\ρ$ into a private state $\Φ$, updates its factors when feedback arrives, and resets to the bound $\ρ$. We analyze the factor-space updates and trace this input-side policy carrier into Orbit, the post-training realization of Memory-Mediated Learning Architecture studied here. In Orbit's released default, the learned residual shapes predictions while memory adaptation changes the retained evidence.\n\nThe controlled studies use a rank-4 residual with 12,288 factor entries on frozen MiniCPM5-1B-SFT. Two feedback updates reduce expected query execution error by 24.65 and 36.65 percentage points relative to the respective trained reset states in a four-candidate development task with 16 parameter groups and three paired seeds. A direct support-loss rule reaches 0.78125\\% error. In repository-balanced chronological replay of public continuous-integration jobs, retaining updates raises half-Brier loss from 0.1274 to 0.1808. A follow-up intervention observes same-batch non-descent and inconsistent future benefit from shrinking steps. These comparisons measure when numerical state helps a later read and when the trained reference is more effective.

cs.LG · cs.CL

版本更新 · v3arXiv:2512.17129

DiffeoMorph: Learning Distributed 3D Morphogenesis with Rotation-Invariant Spectral Shape Matching

Seong Ho Pahng, Guoye Guan, Benjamin Fefferman, Sahand Hormoz

原文摘要

Biological systems can form complex three-dimensional structures through the collective behavior of agents that share a common update rule and operate without central control. How such distributed control gives rise to precise global patterns remains a central question not only in developmental biology but also in distributed robotics, programmable matter, and multi-agent learning. Here, we introduce DiffeoMorph, an end-to-end differentiable framework for learning a morphogenesis protocol that guides a population of agents to morph into a target 3D shape. Each agent updates its position and internal state using an SE(3)-equivariant graph neural network, based on its own internal state and signals received from other agents. To train this system, we introduce a new shape-matching loss based on 3D Zernike polynomials, which compares the predicted and target shapes as continuous spatial distributions, not as discrete point clouds, and is invariant to agent ordering, number of agents, and global orientation. To achieve rotation invariance while preserving reflection sensitivity, we include an alignment step that optimally rotates the predicted Zernike spectrum to match the target before computing the loss. We perform benchmarking to establish the advantages of our shape-matching loss over other standard distance metrics for shape comparison tasks. We then demonstrate that DiffeoMorph can form a range of complex shapes from minimally patterned initial conditions. DiffeoMorph provides a general framework for learning distributed control strategies for morphogenesis, swarm robotics, and programmable self-assembly.

cs.LG · cs.MA · cs.RO · q-bio.QM

版本更新 · v2arXiv:2609.04961

SAM-D2Q: Aligning Multimodal Doc2Query with Search Demand and Conversion for E-commerce

Hui Zhou, Jian Hui Ji, Lei Ma, Rong Xiao, Xiaoyi Zeng

原文摘要

E-commerce search often suffers from vocabulary mismatch between user queries and merchant-authored product titles, since short titles cannot fully cover diverse user expressions or visual product attributes. Although Doc2Query alleviates this issue by generating pseudo-queries for document expansion, traditional methods are text-only and not optimized for e-commerce business objectives. As a result, they may produce semantically plausible but commercially ineffective expansions and miss key attributes present in product images. To this end, we propose E-commerce Search-Aligned Multimodal Doc2Query (SAM-D2Q), a business-aligned multimodal document expansion framework for e-commerce search under Boolean retrieval constraints. SAM-D2Q consists of three stages: (1) task-adapted multimodal supervised fine-tuning to enhance vision-language understanding of product titles, images, and user queries; (2) multimodal data augmentation to improve perception of key visual attributes and expansion coverage; and (3) reinforcement-learning-based preference alignment toward search business objectives, encouraging the model to generate pseudo-queries that better match user intent and commercial value. Offline experiments show that SAM-D2Q substantially improves retrieval performance over traditional Doc2Query methods. Deployed in the AliExpress production search system, SAM-D2Q improves online business metrics, increasing GMV by +3.38% and Pay Count by +2.27%.

cs.IR

新论文 · v1arXiv:2611.13414

Free Lunch Uncertainty: Sequence-Level Rényi Entropy from Particle Decoding for Adaptive LLM Reasoning

Erfan Baghaei Potraghloo, Milad Heidari Sorkhehdizaj, Shahab Rahimirad, Aliasghar Pourghani, Seyedarmin Azizi, Souvik Kundu, Abolfazl Hashemi, Joseph Campbell 等 9 位作者

原文摘要

Quantifying uncertainty over an entire LLM completion is critical for selective prediction and adaptive compute. Zero-cost signals such as mean token entropy and log-probability are local to tokens or to one completion, and self-consistency needs many extra completions. We show that Sequential Monte Carlo (SMC) decoding for power sampling already estimates the normalizing constant of the power distribution, which gives the sequence-level R'enyi entropy (SRE) of the model's completion distribution at zero marginal cost, with a guarantee at every particle budget. A R'enyi chain rule extends this readout to a family of stepwise entropies of any order. We prove that orders above one are tied to the most likely token of each step, which explains why they track log-probability, while the order-1/2 member accumulates the log of per-step search effort. This member is the best zero-cost selective-prediction signal in 10 of 13 settings. On MATH500 its AUROC is 0.022 to 0.044 above log-probability with all four models and within length quartiles, and the same order is selected when each setting is held out. An 8-particle probe identifies the problems that benefit from more particles, and a two-pass policy matches 64-particle accuracy on MATH500 with 34.2% fewer particles over three seeds.

cs.LG · cs.AI · cs.CL

新论文 · v1arXiv:2611.13413

Placing LLM Lookup Tables by Per-Layer Prefetch Deadlines

Wen Huang, Jack Liu

原文摘要

DeepSeek-V4.1's 203 GB of Engram n-gram tables sit in host DRAM today, taking 190 of the 223 GiB a 4-GPU serving process uses. Row indices depend only on input tokens, so reads can start with each step and overlap earlier layers. V4.1's two Engram layers have windows that differ by 30x (0.3 vs. 9.3 ms). We show that this deadline, together with session reuse and contention on a shared device, decides which tier each table can use and which layers need cache, and we give a model that predicts the resulting slowdown from measured read times. Agent and chat traffic reuse rows within sessions, and static hot sets transfer poorly. We build SlackTier, a table tier for SGLang that issues all lookups at step start. It reads the short-window layer first and splits a small GPU row cache across layers. The path runs inside the decode CUDA graph and is bit-exact with the original gather. Layer 14's table can go to NVMe uncached with no measurable decode cost. With both tables on four NVMe disks, mean time per output token on agent and chat replays rises at most 1.4% over unmodified SGLang, and host memory falls from 212 to 34 GiB. Both tables need no cache when read over RDMA (remote direct memory access) from another node's memory; one copy serves two instances. On a KV-constrained agent workload, using the freed memory as host KV cache at the same memory budget raises median throughput by 36% and cuts median p90 time to first token from 38 to 24 s.

cs.DC · cs.AI · cs.LG

版本更新 · v2arXiv:2610.09841

ORCA: Hunting Compositional Failures in Text-to-Image Diffusion

Arshia Hemmat, Amirhossein Vahidi, Amitis Shidani, Mohammad Vali Sanian, Hesam Asadollahzadeh, Aryan Yazdan Parast, Mohammad Lotfollahi

原文摘要

Text-to-image diffusion models fail predictably on compositional prompts: attributes bind to the wrong objects, spatial relations invert, and multi-object scenes lose count. Recent architectures already augment CLIP with a T5 encoder precisely because CLIP's contrastive embedding loses compositional structure, yet these failures persist. We argue the binding problem is therefore not one of missing information but of misaligned information: a text encoder preserves compositional structure, but in a representation space shaped by language modelling rather than vision, and the denoising objective does not directly reward aligning the two. We show this correspondence can be supplied as an explicit training signal, that the relevant cross-modal information is concentrated in a low-rank subspace of self-supervised visual features, and that supplying it can be folded into diffusion training as a single auxiliary loss. Our method, ORCA (Orthogonal Residual Compositional Alignment), aligns the latent of a diffusion transformer with a low-rank target derived from a frozen visual encoder, through a predictor whose orthogonal basis is parameterised by a learned residual between T5 and CLIP embeddings, which provides a prompt-dependent signal for selecting the visual readout subspace. We prove that the cross-modal information recoverable at a given rank is bounded by the spectral mass of the visual encoder's covariance in the top components. Across three diffusion-transformer backbones (DiT-B/2, DiT-L/2, U-ViT-L), ORCA improves FID and GenEval over both vanilla and REPA baselines at zero inference-time cost; on DiT-L/2 it reaches FID 16.65 and GenEval 0.291 at 200K steps, exceeding the strongest 400K baseline at half the training cost, with the largest gains concentrated on attribute binding, spatial relations, and multi-object prompts.

cs.CV · cs.LG

版本更新 · v2arXiv:2610.41927

Variance Reduction for Decentralized Stochastic Optimization with Multiple Local Updates

Jie Liu, Nicola Bastianello, Wenying Xu, Daniel W. C. Ho, Karl H. Johansson

原文摘要

Multiple local updates are widely used in decentralized optimization to enhance communication efficiency, often in conjunction with stochastic gradients when exact gradient computation is costly. However, such algorithms generally converge only to a neighborhood of the optimum, whose size depends on the stepsize and the variance of gradient noise. Variance reduction has been incorporated into ADMM-based methods to achieve exact convergence in expectation, but their stepsizes typically scale inversely with both the square root of the network size and the cube of the number of local updates, leading to slow convergence as the network size or the number of local updates increases. In this paper, we propose a novel primal-dual-based decentralized optimization algorithm that integrates variance reduction with multiple local updates, with a stepsize independent of the network size and inversely proportional to the number of local updates (rather than its cube), thereby enabling significantly larger stepsizes and faster convergence. Moreover, to the best of our knowledge, we establish the first linear convergence guarantee in expectation for decentralized optimization with multiple local updates under stochastic gradients in the strongly convex setting. Beyond the theoretical guarantee, the proposed method exchanges only one variable per round, halving the communication cost compared with gradient-tracking-based methods. Numerical experiments further demonstrate its advantages in convergence speed and accuracy.

math.OC · cs.LG

新论文 · v1arXiv:2611.13412

Bursts That Outlast Themselves: Persistent Degradation in Prefix-Cached Agent Serving

Wen Huang, Jack Liu

原文摘要

LLM agents drive inference engines with closed-loop sessions whose context grows every turn, and serving them economically depends on prefix caching: a hit costs an incremental prefill of about a thousand tokens, a miss recomputes a 10--40K-token context. We show that this makes prefix-cached agent serving metastable. With an identical base workload, a few minutes of extra arrivals leaves a vLLM server in a degraded regime--a ~10% hit ratio, about half the throughput, and a backlog that grows for the remaining 70 minutes--while its no-burst twin stays healthy (9 of 12 pairs; none in the reverse direction). The regime appears at loads the cache-warm server sustains, is absent without prefix caching, and also appears with Qwen3-32B on SWE-bench agent trace shapes and on SGLang, which tolerates larger triggers. Slower service keeps more sessions in the system, their working set outgrows the KV pool, and every turn becomes a recomputation; a first-order workload-balance condition, calibrated independently, brackets when the regime can drain. Getting out is the harder problem: throttling new admissions recovered 1 of 7 matched collapsed runs, while suspending admitted sessions against a capacity-based limit recovered all of them within minutes. A CPU offload tier delays the collapse but does not prevent it. A budgeter that bounds the resident KV demand of active sessions to the pool, with thresholds that need no per-workload tuning, prevented the failure in every run we tested.

cs.DC · cs.AI · cs.LG

新论文 · v1arXiv:2611.13410

SteerDistill: Quantizing LLM Agents via Closed-Loop Teacher Steering

Genming Cui, Xiangyu Li, Zixu Hao, Xin Ding, Yunxin Liu, Ting Cao

原文摘要

Low-bit quantization reduces the inference cost of LLM agents but can compromise task success. Examining failed rollouts, we find that quantized agents overlook task requirements or environmental conditions, producing environment transitions that diverge from successful full-precision references. Correcting these behavioral divergences remains difficult: offline data may miss the student's failure contexts, while its own rollouts may provide few correct responses in those contexts. We introduce SteerDistill, a quantization-aware distillation method in which a full-precision teacher guides the student to revise failed responses and iteratively refines its guidance using feedback from student execution. This closed-loop teacher steering produces locally verified behavioral corrections. We internalize these corrections by training the student without guidance to match the steered student's predictions at corrected turns and the full-precision teacher's at other turns, enabling unassisted inference. Across five agentic benchmarks and three models, SteerDistill exceeds the strongest quantized baseline by 1.67--2.18 points in average score under W4A16 and stays within one point of BF16. On $τ^2$-Bench, it outperforms QeRL by 1.84 percentage points in success rate using 67% fewer training GPU hours and produces shorter successful trajectories than the quantized baselines.

cs.AI · cs.LG

版本更新 · v3arXiv:2608.09886

Entanglement-assisted quantum locally recoverable codes: bounds and constructions with availability

Rutuja Kshirsagar, Gretchen L. Matthews, Julia Shapiro

原文摘要

In this work, we define entanglement-assisted quantum locally recoverable codes with availability, in which any set of up to $δ-1$ erased qudits can be recovered from any one of $t$ local recovery sets, each of size at most $r+δ-1$, with the recovery sets intersecting exactly in the erased coordinates, where $r$ is a (small) positive integer. We show that shared entanglement permits $t>1$, meaning that multiple local recovery sets can be available for the same set of up to $δ-1$ erasures. We establish a Singleton-like bound for this family of codes and present random constructions based on classical linear codes with Vandermonde parity-check matrices. We also provide explicit constructions of entanglement-assisted quantum locally recoverable codes with availability from several classical code families and their folded versions, including Tamo-Barg codes, fiber-product codes, and algebraic-geometry codes such as one-point Hermitian and Suzuki codes.

cs.IT · quant-ph

版本更新 · v5arXiv:2601.21343

Self-Improving Training & Thinking

Ellen Xiaoqing Tan, Jack Lanchantin, Zichun Yu, Polina Kirichenko, Shehzaad Dhuliawala, Danwei Li, Thao Nguyen, Jing Xu 等 14 位作者

原文摘要

Large language models are classically trained in stages: pretraining on raw text followed by post-training for instruction following and reasoning. However, this separation creates a fundamental limitation: many desirable behaviors such as safety, factuality, overall generation quality, and reasoning ability are only added at a late stage, even though the patterns learned earlier strongly shape a model's capabilities. To tackle this issue, we introduce a new way to pretrain and mid-train models that incorporates these behaviors earlier. We utilize an existing strong, post-trained model to both rewrite data and to judge policy model rollouts, thus introducing thinking and reinforcement earlier in training. We further improve reasoning at all stages by introducing the concept of self-improving post-thinking: improving the model's reasoning and self-judging ability by reasoning after generation. In our experiments, we show these approaches can give strong gains across multiple axes.

cs.CL · cs.AI · cs.LG

新论文 · v1arXiv:2611.13406

Stinging the Swarm: Extracting Attacker Intelligence from a Twin-Controller OT Honeypot

Canyon Grey Dell'Omo, Joao Pedro de Carvalho Voltani

原文摘要

Autonomous language-model agents can drive industrial controllers over fieldbus protocols, probing a plant at machine speed and in coordinated swarms. We present Talos, an inline gate that judges every agent request to a controller in milliseconds and routes agents it flags as hostile to a device-faithful decoy -- a mirror twin of the real plant -- where their attack continues harmlessly and is read for intelligence. We evaluate the full pipeline (detect -> quarantine -> mirror -> extract) against 2,400 autonomous attacker agents, 600 in each of four tiers that escalate the attacker's effort to stay hidden, from no concealment to an explicit non-disclosure rule and a trained cover story. No write reached the real plant in any tier: the gate denies the damage path and a device-level envelope check backstops it, while unmapped registers are denied without exception and the gate still permits safe-direction stop commands. Decisions are model-driven in every tier with no inference timeouts, closing a common objection to small on-device judges. Beyond blocking, the pipeline reads the attacker. We find a behavioral structure along two axes -- how persistently an agent attacks and how much it discloses -- and show that the system has a capture channel for three of the four resulting regions: hardened, silent attackers escalate most and are read on the mirror; compliant, talkative attackers volunteer their identity to a diagnostic form before they are ever quarantined; only the paranoid, low-yield region resists both. Attacker identity fields are self-authored cover stories, but involuntary tells -- the self-reported toolchain fingerprint and the attacker's own list of targeted registers -- accumulate across the swarm into a high-confidence profile drawn from agents under explicit orders to stay silent. The gate, mirror, and harness are released under Apache 2.0 with a synthetic device family.

cs.CR · cs.AI

版本更新 · v3arXiv:2511.14613

3D-Guided Scalable Flow Matching for Generating Volumetric Tissue Spatial Transcriptomics from Serial Histology

Mohammad Vali Sanian, Arshia Hemmat, Amirhossein Vahidi, Jonas Maaskola, Jimmy Tsz Hang Lee, Martin Prete, Stanislaw Makarchuk, Yeliz Demirci 等 13 位作者

原文摘要

A 3D transcriptomic profile of a tissue can give a holistic view of its organization in health and disease, but spatial transcriptomics (ST) is costly, so usually only some serial sections are profiled, while H&E is available for all. Most methods that infer ST from histology treat each section independently and ignore 3D structure, while existing 3D-aware approaches are not generative and do not scale well. We present Holographic Tissue Expression Inpainting and Analysis (HoloTea), a 3D-aware flow-matching framework that imputes spot-level gene expression on an unprofiled section from its H&E and the measured expression of adjacent profiled sections. Our key idea is to take the measured expression at the same registered position on neighboring sections and feed this cross-section context to the flow model through a lightweight ControlNet, so that predictions follow anatomical continuity along z. To capture the discrete, zero-inflated nature of counts, the flow starts from a learned zero-inflated negative binomial (ZINB) prior that, on an unprofiled section, is predicted from the measured expression of the nearest profiled section. A global attention block with a small set of inducing tokens adds whole-slide context at a cost linear in the number of spots, so training and inference scale to large 3D ST datasets. Across three ST datasets spanning different tissue types and resolutions, HoloTea consistently improves 3D expression accuracy and generalization compared to 2D and 3D baselines. We envision HoloTea advancing the creation of accurate 3D virtual tissues, ultimately accelerating biomarker discovery and deepening our understanding of disease. Code is available at https://github.com/Lotfollahi-lab/HoloTea.

cs.CV

版本更新 · v2arXiv:2611.09792

Knowing When to Intervene: Causal Evaluation of AI Agents

Ayush Sawarni, Wenqian Xing, Tijana Zrnic, Ramesh Johari, Vasilis Syrgkanis

原文摘要

Current evaluations of agentic AI systems primarily rely on benchmarks. However, the static nature of benchmarks makes them uninformative about the value of auditing interventions at deployment time. For reliable deployment, we often need to determine within an agent's trajectory whether or not to intervene, e.g., by verifying the agent's current output or steering it toward different behavior, using either human review or more costly models or time-consuming agentic supervision. The value of such interventions is fundamentally a causal inference question, asking whether the causal effect of such an intervention is beneficial for the final outcome. Estimating this causal effect is particularly challenging due to the dynamic nature of the problem: the impact of an intervention depends on the current context and environment, as well as the agent's future behavior. We develop the first methodologically grounded framework for evaluating the causal effects of randomized interventions within agent trajectories. By leveraging a relatively small dataset of agent trajectories collected under randomized interventions such as verification or screening, we can estimate the dynamic causal effect of a rich class of (contextual) intervention policies. The structural parameters that we estimate can also be used to construct safety leaderboards that capture the risk of various agentic actions, which can be used by practitioners for audit prioritization. Applying our methods to tau-bench Retail, ToolSandbox, and ImpossibleBench, we find that intervention effects vary substantially across agent actions and states. More broadly, our work suggests a rich research agenda at the intersection of causal inference and agentic evaluation, which we believe is essential for the reliable and safe deployment of AI agents in the real world.

cs.AI · cs.LG · econ.EM · stat.ME

新论文 · v1arXiv:2611.13402

Living constitutions: Participatory alignment for sovereign foundation models

Stine Lyngsø Beltoft, Jonas Vistrup, Peter Schneider-Kamp, Lukas Galke Poech

原文摘要

In Constitutional AI, a plain-text "constitution" encodes values and gets incorporated into a model via post-training. However, the procedure for creating such a constitution poses significant challenges: who contributes, who decides which norms are included, and to what extent must the process be participatory or democratic? These challenges can be paralyzing. Yet, we argue that taking action is essential: without a deliberate constitution, sovereign models would risk misalignment or default to norms set by frontier model providers. We propose treating the development of a constitution as an evolving, participatory process akin to citizen science, iteratively drafted, tested, and refined. We therefore formulate five desirable principles: iterative, modular, justifiable, positional, and legitimate. We further argue that such a document must remain responsive not only over time but across the people it governs, and propose that provisions be marked as settled, contested, or underdetermined, so that durable disagreement is recorded in the constitution rather than resolved away. Empirically, we show via agent-based modelling that proposed process for maintaining living constitutions is robust against 50% bad actors among citizens - assuming a share of 75% participants with good intent in the deliberative body. This indicates that the proposed process tolerates a substantial share of adversarial participants.

cs.AI

版本更新 · v3arXiv:2607.01904

AI Writes Faster Than Humans Can Review: A Longitudinal Study of an Enterprise 2x Mandate

Hao He, Shyam Agarwal, Yegor Denisov-Blanch, Pavel Azaletskiy, Sanmi Koyejo, Bogdan Vasilescu

原文摘要

Enterprises increasingly mandate the use of AI coding tools and set ambitious productivity targets, yet there is little longitudinal evidence on whether such productivity targets have been reached, how, and at what cost. We present a quantitative case study of a documented enterprise "2x" mandate: in June 2025, the CTO of a mid-sized, AI-forward software company set the goal of doubling merged pull requests (PRs) per engineer with AI. Using a panel of 802 developers and 196,212 PRs (Jan 2024-Apr 2026), we estimate the effect of AI adoption by comparing each developer's PR throughput after they started using AI tools with their throughput before adoption. We find that the company eventually reached the target, but only in the tenth month and not in the way the mandate implied: PR throughput barely changed at the mandate date, increased by about 12% when each developer started using AI, and then kept increasing as developers accumulated use. The target was not met merely by working longer or splitting work: extra working days account for at most one-fifth of the throughput increase, and the added PRs became larger and skewed toward features and bug fixes rather than routine changes. The gain is similar across seniority levels and larger in repositories created since 2022. The cost of AI adoption fell on code review, where load per reviewer doubled and automated review exceeded human review in coverage within months of the mandate. We interpret these results as strong evidence that the adoption and sustained usage of AI raises PR throughput in an enterprise setting, with two caveats: throughput is only a proxy for productivity, and the exact size of the effect remains uncertain due to a lack of truly randomized assignment in our setting. Our results have implications for organizations under AI transformations and researchers studying the impact of AI on software development productivity.

cs.SE

新论文 · v1arXiv:2611.13401

SPARK: Semantic-preserving Prompt Mutation for Code Generation using Consensus-based Verification Framework

Alireza Daghigh, Chung-Yu Wang, Luca Filippelli, Hamed Taherkhani, Hadi Hemmati, Hung Viet Pham

原文摘要

Automated code generation with large language models (LLMs) has improved in recent years. With the lack of test cases in real-world usages, generated code is not automatically verifiable and often requires manual validation. Previous studies have leveraged consensus verification, where code solutions and corresponding test cases are generated and evaluated based on functional consistency. Building on this, SPARK incorporates semantic-preserving mutations to problem descriptions such as paraphrasing, grammatical adjustments, function renaming, and other alterations to enhance the robustness and correctness of LLM-generated code. For each mutated description, SPARK generates a set of code solutions and test cases. These candidate solutions are evaluated using consensus-based verification, where the solutions' quality is measured by how consistently they pass generated tests. We assess the effectiveness of SPARK on three benchmarks (LiveCodeBench, BigCodeBench, and DLBench) across three LLMs (o4-mini, DeepSeek-R1-Basic, and Gemini-2.5-Pro). Our evaluation demonstrates that SPARK outperforms state-of-the-art consensus-based verification approaches by up to 4.1%. Our generalization experiment highlights the practical potential of SPARK as an effective strategy for enhancing the reliability and correctness of LLM-based code generation systems in scenarios where traditional validation resources are scarce, while our qualitative analysis shows insight into how SPARK's mutation helps LLMs generate better code solutions.

cs.SE

新论文 · v1arXiv:2611.13400

Orthographic depth at two scales, with English also matched for lexicon size

Adam Y. Shavit

原文摘要

A reader can struggle to read in only one of their languages, and one proposed reason is how predictably spelling maps to sound. We measure that predictability as conditional entropy, the pronunciation uncertainty left once the spelling is seen, at two grain sizes (unit sizes): the alignment unit, an aligned spelling-sound chunk, and the syllable, bounded in the pronunciation. We analyse word lists in twelve alphabetic orthographies and report five in full, with English pronunciations from a dictionary and the rest from the synthesizer eSpeak-ng, length marks removed. Among the five, English has the most uncertainty and Finnish the least at both grains. Across all twelve, Portuguese and Danish exceed English per alignment unit, and German and Italian order differently by grain and transcription. These orders hold when English is transcribed by eSpeak-ng with u allowed to spell /j/ plus a vowel. Uncertainty per unit has no shared direction as units grow: it falls in Spanish and, less securely, German, is level in Finnish, depends on transcription in Italian and on the word list in English, and rises in Norwegian and Swedish. Summed uncertainty per word falls in all twelve, but that score is handed the pronunciation's boundaries and mixes context with in-sample memorisation, so it is no evidence of what a reader computes. An exact identity states when discarded spelling boundaries could reverse that fall. A uniform 20,000-word English sample holds 35% of the dictionary's syllable spellings and has syllable entropy 0.098 bits lower, so lexicons must be matched in size. In an exploratory check over eleven languages, syllable-scale entropy tracked first-year reading accuracy (rho = -0.74, falling to -0.25 if English is given the lowest entropy instead), without separating the syllable scale from the alignment unit. These estimates describe sampled writing systems, not readers.

stat.AP · cs.CL · q-bio.NC · stat.ME

版本更新 · v2arXiv:2609.12277

Reinforcement Learning over Patient Trajectories for Clinical Reasoning in EHR Foundation Models

Yuxin Xiao, Sheng Zhang, Chandan Singh, Tristan Naumann, Hoifung Poon, Jianfeng Gao, Xiaodong Liu

原文摘要

Electronic health record (EHR) foundation models trained on longitudinal patient trajectories have demonstrated strong performance across diverse clinical prediction tasks. However, their clinical reasoning capabilities remain constrained by next-token prediction on limited and incomplete EHR data. To address this, we propose a reinforcement learning (RL) fine-tuning framework that treats EHR foundation models as generative policies over patient trajectories. We formulate common clinical prediction problems (e.g., hospital readmission) as event-conditioned, time-windowed reasoning tasks. We then design time-aware, rollout-sensitive rewards to account for finite rollout lengths and temporally inconclusive outcomes. We find that RL fine-tuning consistently improves over pre-trained backbones and strong baselines. Notably, it enables smaller models to match larger pre-trained models in data-limited regimes and induces positive transfer across tasks. Further analysis shows that RL fine-tuned models generate trajectories with stronger structural and semantic alignment to ground truth and greater downstream utility.

cs.LG · cs.AI · cs.CY

版本更新 · v2arXiv:2601.15312

Do people apply different norms to humans and large language models acting on their behalf? Evidence from norm elicitations in two canonical economic games

Paweł Niszczota, Elia Antoniou

原文摘要

While delegating tasks to large language models (LLMs) can save people time, there is growing evidence that offloading tasks to such models produces social costs. We elicit appropriateness of behavior in two canonical economic games to study whether people have different norms for humans and LLMs acting on their behalf. More specifically, we study the social appropriateness of a spectrum of possible behaviors: when LLMs divide resources on our behalf (Dictator Game and Ultimatum Game) and when they evaluate the fairness of resource splits (Ultimatum Game). We use the Krupka-Weber norm elicitation task to detect possible shifts in social appropriateness ratings. Results of two pre-registered and incentivized experimental studies using representative samples from the UK and US (N = 2,658) show three key findings. First, in the Dictator Game, the most socially appropriate allocation remains the equal split when the Dictator is a human or an LLM acting on behalf of a human, but self-serving allocations are judged less appropriate when implemented by an LLM. Second, in the Ultimatum Game, rejecting offers from an LLM Proposer is judged as more appropriate than rejecting identical offers from a human Proposer, with this difference concentrated among relatively generous offers. Lastly, rejection decisions implemented by an LLM Responder are judged no less appropriate than identical rejection decisions implemented by a human Responder. Post hoc, these patterns are consistent with the notion that people grant LLMs bounded agency in the investigated resource-allocation scenarios. Our findings have implications for the design of AI systems.

cs.GT · cs.AI · cs.CL · cs.CY · cs.HC · econ.GN

新论文 · v1arXiv:2611.13398

The Dose Makes the Poison: Exposure-Aware Evaluation of Generative Drug Design

Maria Franz, N. Benjamin Erichson

原文摘要

Generative models are increasingly used to propose molecules for drug discovery. They can produce far more candidates than can be experimentally tested, making computational selection essential. Most selection methods rely on predicted molecular properties, such as potency and toxicity, but overlook how these properties translate into drug exposure at a given dose. As a result, molecules that appear promising on paper may require impractical doses or raise safety concerns. We investigate this gap by using physiologically based pharmacokinetic (PBPK) simulations to re-evaluate molecules selected from three PARP1-targeted generative pipelines. Of 60 shortlisted molecules, seven fail to achieve sufficient target exposure, while 28% of the remaining candidates show predicted safety risks at their required dose. Moreover, 85% of candidates with an achievable dose are outperformed by molecules overlooked during static selection. These findings suggest that accounting for dose and exposure can substantially change which generated molecules are considered promising for drug development.

q-bio.QM · cs.AI · cs.LG

版本更新 · v2arXiv:2606.26969

Einstein World Models

Munachiso Samuel Nwadike, Zangir Iklassov, Ali Mekky, Zayd M. Kawakibi Zuhri, Kentaro Inui

原文摘要

Does intelligence require the ability to reason about phenomena beyond direct experience? It is natural to suspect that some complex thought cannot be captured through language alone. However, of particular concern to this work, is whether visualising counterfactual events can complement language as a mechanism for complex thought. We ask whether LLMs can be trained to utilise such visualisation mechanisms, in a way that benefits their reasoning abilities. Motivated by this question, we propose Einstein World Models. EWMs are a blueprint for LLM-based reasoning systems that place visual-temporal rollouts inside the reasoning trace, allowing them to reason in ways that text alone may not support well. In an EWM, the LLM calls a world-module (not to be confused with a world model), to produce short rollouts of scenes under consideration. The returned rollout is treated not as the answer, but as an inspectable hypothesis that can support later reasoning. Einstein World Models extend the capability of LLMs for tool calling (such as web search or code execution), into the domain of visual thought experiments.

cs.AI · cs.CL · cs.CV

新论文 · v1arXiv:2611.13397

The Optimal Constant for One-Shot Channel Simulation and Strong Functional Representation Lemma

Cheuk Ting Li

原文摘要

We prove a tight bound for one-shot channel simulation via a stronger version of the strong functional representation lemma: for any random variables $X,Y$, there exists a random variable $Z$ with $I(Z;X)=H(Y|X,Z)=0$ and $H(Y|Z)\le I+\log I+(1-γ)\log e-\log\log e+O(I^{-1}\log I)$, where $I=I(X;Y)$ and $γ$ is the Euler-Mascheroni constant. A more convenient bound is $H(Y|Z)<I+\log(I+2)+1$. The first bound is notable in the sense that the constant term $(1-γ)\log e-\log\log e$ is optimal, and cannot be improved for any bound stated only in terms of $I$. Hence, this is the tightest form of the strong functional representation lemma up to the third order. We also prove that the Poisson functional representation is almost optimal up to a constant gap, and provide a simple finite-blocklength achievability bound that implies the $(1/2)\log n$ redundancy of the simulation of memoryless non-singular channel. These are proved via a generalization of the functional information, in particular, the exponential functional information, which closely approximates the cost of one-shot channel simulation.

cs.IT

新论文 · v1arXiv:2611.13396

The AI Era's Lines of Code? A Case Study on What an Engineer's Token Spend Measures

Shyam Agarwal, Olga Lvova, Agnia Sergeyuk, Anastasiia Serova, Timofey Bryksin, Thomas Zimmermann, Bogdan Vasilescu

原文摘要

Organizations use engineers' token spend to monitor AI adoption, set budgets, and measure productivity, but the metric's relationship with engineering output is unclear. We link five weeks of token usage in coding agents to the commits of 404 engineers at JetBrains, and analyze a company survey of 412 employees. We find that token spend varied mostly between engineers, and which engineers spent the most stayed largely the same over the five weeks. Among weeks with similar commit counts, the 90th percentile of spend was 39 to 240 times the 10th. Within an engineer, doubling spend was associated with about 10% more commits. Available work metrics explained 38% of the variation in spend between engineers and 13% of the week-to-week variation within an engineer. In the survey, respondents in technical roles reported in nearly nine of ten agent evaluations using agents for tasks beyond coding, such as analysis, research, and planning. Token spend therefore measures how much engineers use agents across many kinds of work and not the outcome and value of that work. We recommend using token spend at most for resource accounting, and evaluating agent-assisted work with evidence about tasks and outcomes.

cs.SE · cs.AI

新论文 · v1arXiv:2611.13395

A Taxonomy of Failing Bug Reproduction Tests on SWT-Bench

Luca Filippelli, Hamed Taherkhani, Hung Viet Pham, Hadi Hemmati

原文摘要

Automatic Program Repair (APR) tools require high-quality test generation models to validate candidate patches. Despite the advancements of automated bug reproduction testing tools, they continue to struggle with issues of higher complexity. These struggles may be caused by a framework's lack of problem-specific context. We address the unclassified nature of these framework failures by providing a taxonomy of bug reproduction tests. This study uses the AssertFlip and OpenHands frameworks, the two highest-scoring open-source tools on SWT-Bench Verified. The significance of model selection on the causes of failure is also explored by utilizing GPT-5-mini and GPT-4o-mini on both tools. The error traces of these frameworks are categorized to determine granular modes of failure. Through this process, mechanical and implementation failures were the most common failure classes. Overall, a tool's generation approach has a significant impact on precision and coverage, while the selection of the model alters the quality and quantity of generated tests. Specifically, more powerful models with higher reasoning capability tended to present fewer failures due to a lack of technical or problem-specific context. Our experiment also shows evidence that our taxonomy can be used as concrete guidance to augment the issue description to improve the generation quality.

cs.SE

版本更新 · v2arXiv:2611.06788

Integrated Identity Protection for Semantic Communications in Embodied IoT Systems

Yu Zhang, JingQi Liu, Shuangrui Zhao, Yuanyu Zhang, Yulong Shen, Xiaohong Jiang

原文摘要

Embodied Internet of Things (IoT) systems are promoting the adoption of task-oriented semantic communication (SemCom), in which the trustworthy task decisions increasingly require both semantic reliability and source legitimacy. However, existing trustworthy SemCom designs mainly focus on semantic content recovery and protection, with limited attention to source identity protection. Motivated by this limitation, this paper establishes a new information exploitation perspective for integrated SemCom-identity protection that leverages task-level error tolerance to coordinate semantic information and source identity information carried by the same waveform. We first examine these two information resources and then analyze their integration feasibility at the observation, processing, and decision levels. To exploit these resources for different identity protection objectives, we further develop a dual-branch identity protection mechanism with two operating modes. In the authentication mode, semantic recovery and physical-layer source verification are performed in parallel using the same received observation. In the concealment mode, identity exposure is suppressed while semantic reconstruction quality is maintained to support anonymous transmission. Finally, two representative case studies on active source impersonation and passive identity inference are presented to evaluate the proposed framework.

cs.CR · eess.SP

版本更新 · v2arXiv:2607.04096

Forethought: Verifiable Reasoning from Neurosymbolic Primitive Programming

Vishvesh Bhat, Emmanuel Anaya Gonzalez, Jay Vaghasiya, Asad Aali

原文摘要

Modern agentic workflows decompose user requests into sequences of tool calls with correctly resolved parameters, the results of which are processed through reasoning traces in the language model's context window. The prevailing route to improving such reasoning is test-time scaling, which trains models to search over long chains of thought; but the resulting capability is entangled in model weights, cannot be checked step by step, and is costly at inference. We present Forethought, a neurosymbolic reasoning system built from a library of symbolic and neural primitives composed through a typed domain-specific language. The results are reasoning programs: concrete representations of the model's work, in which every step carries an output contract that is checked at design time and enforced at execution time. Symbolic steps hold by construction, while learned steps are contract-checked and empirically validated, giving an explicit boundary between what is guaranteed and what is measured. Instantiated as a tool-calling execution kernel and evaluated on four standard tool-calling benchmark suites, Forethought improves end-to-end accuracy by a median of 32% relative to prompting the same base model directly, and reaches the highest aggregate accuracy among the evaluated configurations, though only by a margin within single-run noise over a frontier model. Systems pairing small open models with Forethought's fixed small-model layer match that frontier model, and a pipeline with a non-reasoning responder outperforms its reasoning sibling while requiring only a single fine-tuning stage. Removing the contract-and-control layer erases most of these gains, showing that checked structure, not decomposition alone, drives the improvement. Our results show the potential of rich symbolic orchestrators for building language-model agents that are more accurate, reliable, and auditable.

cs.AI

新论文 · v1arXiv:2611.13393

Plumbline: When the Loss Lies, Validating LLM Training Correctly

Ruochen Wen, Hongyang Wang, Siwei Li, Wenbo Chen, Xian Li, Jingguo Liu, Leo Fan, Yiming Yan 等 11 位作者

原文摘要

Correctness is a prerequisite for training large language models (LLMs) at scale, yet it is hard to ensure: a fault can sit in any part of a run, and how it shows differs across frameworks and the libraries they depend on. Some failures are loud and some are silent, and both take time to detect and trace. A single fault can waste a long run: the run crashes late, or the model scores lower only at the final evaluation. We present Plumbline, a correctness validation system for LLM training, used in production at Amazon. Plumbline checks each part of the training lifecycle, from the data a run trains on to the model it serves, down to the new layers each model brings and up to whole runs. Its diagnosis reuses these checks to trace failures to their root causes. Both rest on a systematic correctness methodology designed to carry across models and frameworks. Plumbline supports two agentic use cases: Day-0 onboarding, which validates the training of a newly released model before teams rely on it, and root-cause tracing of training failures. Across 3 frameworks and 5 models, it found 61 problems, 32 of them silent. Of these, 42 are fixed or reported upstream, 8 of them first raised by us. We further group the 61 into seven failure patterns that show where training breaks, and report how problems recur across frameworks and models.

cs.DC · cs.LG