跳到正文
研踪PaperTrail

YOUR RESEARCH, IN SIGHT

让下一篇重要论文，不再错过。

覆盖计算机全学科，以 AI 为快捷入口。关注方向、作者和关键词，把发现变成持续的研究积累。

部分采集异常
正在读取账号与研究记录…

全站更新概况

+1673新增论文
+1257其中 AI 相关
+197已有论文修订
+100新增 AI 导读

arXiv 最近完成扫描：2026-11-13 18:26 · 全站最近入库：2026-11-13 14:16各信源进度 →

统计口径与数据新鲜度

统计 2026-11-12 18:40 — 2026-11-13 18:40（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：2026-11-13 18:26最近发现新增版本：2026-11-13 14:16库内论文最新更新时间：2026-11-13 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。

23,348 已收录论文19,580 AI 相关6,726 有修订112 HF 社区发现

浏览论文库

研究日报 / 周报 →按信源浏览 →

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 17,362 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-13 18:24（北京时间）
版本更新 · v3arXiv:2609.19150

Sampling Reveals Style: Unsupervised, Training-Free Discovery of Prompt-Conditional Stylistic Axes in LLM Activations

Ajit Mallavarapu, Ziwei Gu

原文摘要

Large language models (LLMs) encode rich stylistic structure in their hidden activations, but discovering which stylistic dimensions are salient for a given prompt typically requires supervised contrastive data. We present a training-free, prompt-conditional alternative: we repeatedly sample completions of a single prompt at elevated temperature, apply Principal Component Analysis (PCA) to the pooled hidden activations, and label the resulting axes automatically from the pole generations. We validate the discovered axes against 245 human-elicited stylistic annotations in a two-phase study. On our strongest model (Qwen-3.5-4B-Instruct), the top two axes match spontaneously requested human dimensions with 72.8% precision and 43.6% macro-recall, and 75.6% of validity ratings judge the axes' polar generations accurate to their labels, with 90.9% adjacent inter-annotator agreement. Discoverability is strongly model-dependent: both Qwen models and Llama-3.2-3B expose human-salient axes, while DeepSeek-7B-Chat drops to 35.3% precision, its leading components dominated by structural rather than stylistic variance. Simple PCA over a model's own decoding variance is thus an effective, low-cost probe of stylistic structure in LLM representations, one that also exposes sharp cross-model differences in how that structure is organized.

cs.CL · cs.LG

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.08809

Sensing-Based Communication: A Non-Orthogonal Passive Reflection Framework for ISAC

Xin Xiong, Mathew Salas, Sihua Shao, Chen Chen, Mostafa Hassanalian, Rafael Fierro

原文摘要

While integrating uplink without disrupting the primary downlink remains a critical challenge for existing reflective communication systems, this paper proposes a non-orthogonal passive integrated sensing and communication (ISAC) mechanism termed Sensing-based Communication. Originating from and validated within an optical retro-reflective system, this mechanism possesses an underlying wave-agnostic architecture with the potential to be generalized across other physical modalities. It leverages incident downlink signals for parasitic modulation via passive tags. By relying solely on fundamental on-off keying (OOK) reflection, the tag unifies the reflective sensing channel and the uplink communication channel, and simultaneously achieves both sensing and duplex communication. Consequently, our approach introduces uplink capability without sacrificing the downlink data rate or consuming additional spectrum resources. At the receiver, we design a joint decoding architecture combining cross-correlation computation and successive interference cancelation (SIC) to fully exploit the degrees of freedom in broadband signals for multi-user interference suppression. We derive closed-form expressions for the interference and bit error rate (BER) assuming zero-mean Gaussian signals. Using an experimental testbed, we validate the pivotal theoretical finding, the scaling relationship of SINR with signal bandwidth and cross-correlation time--where linear regression yields a strong goodness-of-fit with $R^2 = 0.9688$. Furthermore, evaluations under noise-like orthogonal frequency division multiplexing (OFDM) waveforms exhibit high consistency between theoretical derivations and simulation data ($R^2 = 0.9372$, mean absolute percentage error, $\text{MAPE} = 14\%$), confirming the model's accuracy and waveform generalizability.

cs.IT · eess.SP · eess.SY

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2609.20797

The Strong Secretary Conjecture is True for Linear Matroids

Kristóf Bérczi, Shaddin Dughmi, Vasilis Livanos, José A. Soto, Victor Verdugo

原文摘要

We prove a $1/e$ guarantee for Matroid Secretary on linear matroids: each element of a fixed optimal basis is selected with probability at least $1/e$. This settles the strong secretary conjecture for this class of matroids. Our result holds both when the matroid is known in advance and when a linear representation is given online. In the known-matroid model, the result extends to fully modular extendable matroids. We also obtain $1/e$ guarantees for a more general model on linear matroids, both when the matroid is known and when its representation is revealed online. In XOS Matroid Secretary, agents with XOS valuations over the elements arrive in uniformly random order, and the algorithm must assign each arriving agent a bundle of elements. The bundles must be pairwise disjoint, and their union must be an independent set of the matroid. We obtain an expected total value of at least $1/e$ times the offline optimum. This model contains Group Matroid Secretary, in which the elements are adversarially partitioned into groups that arrive in uniformly random order, and the algorithm selects a subset of each arriving group. For this special case, we select each element of a fixed optimal basis with probability at least $1/e$

cs.DS

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.08804

Generative Search and Recommendation at Shopify

Juhana Kangaspunta, Khalil Hajji, Darshan Fofadiya, Kyumin Park, Michael Garner, Anthony Maltsev, Ryan Synk, Jose Sepulveda 等 11 位作者

原文摘要

Search and recommendation at Shopify require retrieving relevant products from a changing catalog of billions of products under strict latency constraints. We present SID-LLM, which casts both tasks as generating semantic IDs: short sequences of content-derived tokens representing products. We jointly pre-train SID-LLM on catalog and behavioral data, giving search and recommendation a shared representation and training objective. We show that reinforcement learning further improves judged search relevance. Furthermore, distillation from task-specific experts improves recommendation quality over joint supervised fine-tuning without a statistically detected loss in search quality. Finally, for efficient serving, we pair multi-token prediction with a lightweight ranker, improving retrieval quality over our autoregressive baseline while reducing latency. We evaluate SID-LLM both as a standalone retriever and as a component of Shopify's production search and recommendation systems. As a standalone search system, SID-LLM achieves quality comparable to Shopify's existing multi-model search stack while meeting production latency requirements. Integrating SID-LLM into the search and recommendation stacks provides significant gains in both offline evaluations and online A/B tests. Together, these results demonstrate the practical value of semantic-ID generation as a unified approach to product discovery at scale.

cs.IR

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.08802

Scale as a Modality: Multi-Scale Pixel-Space Diffusion Transformers

Ruibo Chen, Kaishen Wang, Tianyi Xiong, Chenxi Liu, Yanshuo Chen, Heng Huang, Jian Du

原文摘要

Latent diffusion models rely on pretrained autoencoders that introduce lossy compression and constrain spatial granularity. Pixel-space diffusion avoids this bottleneck and enables flexible tokenization at multiple spatial scales, yet existing approaches largely adopt two-level coarse-to-fine architectures with limited cross-scale interaction. We introduce MS-pDiT, a multi-scale pixel-space diffusion transformer that treats spatial scale as a modality. MS-pDiT tokenizes the same noisy image at multiple patch sizes, processes each scale in a dedicated transformer stream, and enables bidirectional information exchange through compacted MM-DiT-style joint attention, with attention cost independent of the finest patch size. This design establishes the number of scales as a new architectural scaling dimension. On ImageNet $256\times256$, increasing the number of scales from two to five consistently improves FID from 4.39 to 2.31, 2.26, and 2.19, while adding an intermediate scale substantially outperforms increasing depth under comparable parameter budgets. Further analysis highlights the importance of bidirectional cross-scale interaction and late introduction of fine-grained streams. MS-pDiT achieves FID scores of 1.43 and 1.44 on ImageNet $256\times256$ and $512\times512$, respectively, and naturally extends to text-to-image generation, achieving 0.86 on GenEval and 83.9 on DPG-Bench.

cs.CV

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.10055

HeteroTask: Multi-Variant Task Scheduling and Execution on Distributed Heterogeneous Clusters

Hongzheng Tian, Alok Mishra, Haocheng Xu, Dejan Milojicic, Sitao Huang

原文摘要

Heterogeneous systems that combine CPUs and GPUs across multiple servers offer substantial compute capability but complicate scheduling. For DAG-based applications with multiple task implementations, effective execution requires coordinating variant selection, device placement, and data movement. Offline planning can exploit global dependency information, yet plans derived from isolated kernel profiles may diverge from execution because of runtime overheads. We introduce HeteroTask, a framework integrating kernel-wise profiling, PEFT-based multi-variant planning, host-buffer lifetime modeling, and schedule-driven MPI execution. A Bayesian-optimization loop calibrates empirical overhead corrections for repeated, fixed-DAG workflows on exclusively allocated nodes. We evaluate five benchmark and workflow-derived applications on up to six 8-GPU nodes. On two 8-GPU nodes, HeteroTask reduces application-level completion time by 19.7-72.1% relative to the compared StarPU implementations. For Canny Edge Detection, calibration reduces task-window makespan by 25.0% relative to planning without overhead corrections. Combined with CPU/GPU variant selection, calibration reduces it by 58.9% relative to an uncalibrated GPU-only execution.

cs.DC

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.08800

On the Few-Shot Adaptation of Vision-Language Models: A Few Gradient Signs Are All You Need

Nairouz Mrabah, Ismail Ben Ayed, Mahdi S. Hosseini, Eric Granger

原文摘要

The limited supervision available for the few-shot adaptation of vision-language models makes gradient estimation particularly noisy. We find that the gradient magnitudes are severely miscalibrated, while the signs of a small top-magnitude subset remain informative. This motivates SignSparseSGD and DynSignSparseSGD, two state-free optimizers that use gradient magnitudes only to select where to update and fixed-magnitude signs to decide how to update. The static variant fixes the support after the first minibatch. The dynamic variant refreshes it at every step. Across 11 datasets, three CLIP backbones, and five shot levels, the proposed methods are competitive with or outperform parameter-efficient and optimization-based baselines on average, while exhibiting markedly lower sensitivity to long training schedules. They introduce no task-specific modules, maintain no first- or second-moment optimizer states, and use one shared configuration across datasets, backbones, and shot levels. Our code is included in the supplementary materials and will be released upon publication.

cs.LG · cs.CV

查看论文
arXiv ↗PDF ↗
版本更新 · v3arXiv:2604.23853

ClawTrace: Cost-Aware Tracing for LLM Agent Skill Distillation

Boqin Yuan, Yue Su, Renchu Song, Sen Yang, Jing Qin

原文摘要

Skill-distillation pipelines learn reusable rules from LLM agent trajectories, but they lack a key signal: how much each step costs. Without per-step cost, a pipeline cannot distinguish adding a missing step to fix a bug from removing an expensive step that never affected the outcome. We use the cost-attribution gap to ask whether the rule types inside a distilled skill transfer the same way to new tasks. ClawTrace records cost-attributed agent traces and compiles each session into a TraceCard; CostCraft reads TraceCards and writes three kinds of skill patches: preserve, prune, and repair. We find a pattern aggregate metrics hide. On 30 held-out SpreadsheetBench tasks across two seeds, removing prune patches roughly tripled the quality-regression count without lowering median cost. Across the full 84-task SkillsBench transfer, CostCraft saves no aggregate cost. All three quality regressions trace to the preserve lane, and both quality wins trace to the prune lane: prune patches act as quality guardrails while preserve patches drive regressions. We argue that reusable agent skills should be evaluated at the rule-type level, not as monolithic instruction packages. To support this, we release ClawTrace, the TraceCard schema, and the full set of typed skills.

cs.AI

查看论文
arXiv ↗PDF ↗
版本更新 · v3arXiv:2608.26515

Sharp Minimax Regret for Infinite-Memory Logistic Prediction

Vaneet Aggarwal

原文摘要

We study sequential prediction of binary marks whose log-odds are a linear function of all past inputs. The inputs are independent uniform signs. The coefficient of the input observed $j$ steps earlier is unknown, but its absolute value is at most a known number $r_j$, and $\sum_j r_j\le B$. A learner is judged by its expected cumulative log loss in excess of the loss of the true conditional probabilities. We show that for every such sequence $r$ and every horizon $T$, the minimax value of this regret lies between $c\,Γ_T(r)/\max\{1,B^2\}$ and $\frac12Γ_T(r)$, where $Γ_T(r)=\sum_{j\le T}\log(1+(T-j+1)r_j^2)$ and $c$ is an absolute constant. The term for lag $j$ combines the size bound $r_j$ with the number $T-j+1$ of rounds in which that lag influences a prediction, and no monotonicity or other regularity of $r$ is needed. The upper bound is attained by a Bayesian mixture with independent Gaussian priors. For the lower bound, we apply the van Trees inequality at each prediction step and then bound the expected log-determinant of the random matrix whose rows are the lagged input vectors. We also characterize when $\frac12$ is the exact leading constant: for envelopes with a common bound $B$, possibly changing with $T$, the regret is asymptotic to $\frac12Γ_T(r)$ if and only if $\sum_{j\le T}\min\{1,(T-j+1)r_j^2\}=o(Γ_T(r))$. This gives, for example, $\log^2(A^2T)/(8α)$ for $r_j=Ae^{-αj}$, whereas for $r_j=Aj^{-s}$ with $s>1$ the regret is of order $T^{1/(2s)}$ and its ratio to $Γ_T(r)$ stays below $\frac12$. Finally, for a known horizon, an online Newton predictor with features scaled by $r_j$ attains regret of order $Γ_T(r)$ with polynomial computation per round.

cs.IT · cs.LG

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.08797

MotionBench-XAI: Benchmarking What Shapes Shapley Explanations in Spatiotemporal Machine Learning

Mathilde Papillon, Louisa Cornelis, Nina Miolane

原文摘要

Predicting outcomes from spatiotemporal data, such as assessing Parkinson's severity from human motion, often requires explainability in addition to accuracy. Practitioners need to know what drove the prediction, whether it be spatial (a body joint), temporal (a specific window), or both. Shapley values are attributions computed by treating input features as players in a cooperative game and measuring how the model output changes as coalitions of players are removed. Computing them involves choosing what counts as players, how absent players are imputed, and how the game is evaluated. How these choices affect explanation quality is poorly understood for spatiotemporal data, and real data carries no ground-truth attributions to check against. We present MotionBench-XAI, a benchmark built on i) five synthetic spatiotemporal datasets with exact ground-truth Shapley values, ii) three real datasets across clinical gait, electrocardiogram, and audio, iii) ten attribution methods across three player sets, and iv) three trained architectures. We show that grouping features into players, across time, space, or both, matters more than the choice of imputer. Generative imputers improve explanations only when their conditional models are accurate. On real data, the standard proxy metrics select contradictory best methods. We distill these findings into practical insights and provide an open-source pipeline.

cs.LG

查看论文
arXiv ↗PDF ↗
版本更新 · v4arXiv:2509.22738

Enabling Approximate Joint Sampling in Diffusion LMs

Parikshit Bansal, Sujay Sanghavi

原文摘要

In autoregressive language models, each token is sampled by conditioning on all the past tokens; the overall string has thus been sampled from the correct underlying joint distribution represented by the model. In contrast, masked diffusion language models generate text by unmasking tokens out of order and potentially in parallel. Generating an overall string sampled from the correct underlying joint distribution would (again) require exactly one token unmasking in every full-model forward pass. The more tokens unmasked in parallel, the further away the string is from the true joint; this can be seen in the resulting drop in accuracy (but, increase in speed). In this paper we devise a way to {\em approximately} sample multiple tokens from the joint distribution in a single full-model forward pass; we do so by developing a new lightweight single-layer ``sampler" on top of an existing large diffusion LM. One forward pass of the full model can now be followed by multiple forward passes of only this sampler layer, to yield multiple unmasked tokens. Our sampler is trained to mimic exact joint sampling from the (frozen) full model. We show the effectiveness of our approximate joint sampling for both pretrained-only (Dream-7B-Base, Llada-7B-Base) and instruction-tuned (Dream-7B-Instruct, Dream-7B-Coder) models on language modeling and math \& coding tasks. When four tokens are unmasked for each full-model denoising step, our sampling algorithm achieves a MAUVE score of 0.87 (vs marginal baseline of 0.31) with respect to the true joint distribution.

cs.CL · cs.LG

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.08795

MotionDecipher: LLM-Assisted Clinical Interpretation of Human Motion Signals for Efficient Gait Analysis

Mingrui Ma, Tian Tan, Jinsi Guo, Jessica Rose, Yiwen Dong

原文摘要

Clinical gait and motion analysis is widely used to guide targeted interventions for patients with neurological disorders, the outcome of which is typically a report with descriptions of joint movements and forces. Current practices rely on experts' manual effort, which is labor-intensive and time-consuming. Recent advancements in Large Language Models (LLMs) provide opportunities to address this problem, but translating human motion into clinical gait descriptions remains challenging because the quality of the context directly affects LLMs' reasoning capability over continuous motion signals. To tackle this, we introduce MotionDecipher, a novel framework bridging the semantic gap between motion data and clinical interpretations. Specifically, we made 3 key advancements: 1) a Human Motion Tokenizer that converts motion signals into domain-specific tokens for time- and frequency-informed gait interpretation; 2) a Meta Feature Selector that transforms tokens into Structured Representations; and 3) a Hybrid Context Constructor to determine the optimal demonstration-based prompt for In-Context Learning. We evaluated MotionDecipher on data collected at an accredited gait lab from 122 patients and 46 controls. Results show that MotionDecipher achieves 0.91 BERTScore (13% over GPT-5), while expert evaluation ranks MotionDecipher highest in Correctness, Completeness, and Conciseness compared to baselines, highlighting its promise for efficient gait analysis.

cs.HC · cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.08794

High-Dimensional Off-Policy Evaluation without Sample Splitting using an Onsager Correction

Saksham Jain, Guillaume Braun, Masaaki Imaizumi

原文摘要

Reward models used for off-policy evaluation (OPE) are commonly selected using held-out data. When sample size and context dimension grow proportionally, the difference in test risk between models fitted on the full sample and on a fixed fraction of it need not vanish. Held-out evaluation can therefore consistently estimate the test risk of the model trained on the smaller sample, but because the relative ranking of candidate reward models can change after full-data refitting, held-out selection can be suboptimal. We develop an Onsager-corrected, target-policy-weighted criterion that estimates the full-data reward model's out-of-sample risk under the target policy using only its training observations. This enables early stopping along the full-data gradient-descent trajectory without sample splitting. For neural networks with fixed depth and hidden width under Gaussian contexts, single-index rewards, context-dependent multi-index target policies, and known context-independent logging, we prove uniform consistency of our criterion over a fixed finite set of gradient-descent iterates and a near-oracle guarantee for the selected reward model. The same correction yields a doubly robust estimator that consistently estimates the true policy value without sample splitting, even when reward-model prediction error does not vanish. Experiments show decreasing risk-estimation error as dimension grows at a fixed sample-to-dimension ratio, and improved reward-model selection relative to held-out baselines. Empirically, our correction also reduces the root mean squared error of doubly robust value estimates computed from the same data used to fit and select the reward models.

stat.ML · cs.LG · math.ST

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.08792

Online Bin Packing with Dispatch Deadlines

Pei-Lin Chiang, Wing-Kai Hon, Kazuo Iwama, Li-Tzu Yang

原文摘要

We study a generalization of online bin packing by introducing dispatch deadlines. Items arrive over time in an online fashion. Each item has an arrival time, a size in $(0,1]$, and a deadline, all of which are revealed only when the item arrives. The algorithm must immediately and irrevocably assign the item to a unit-capacity bin that has not yet been dispatched. Each bin must be dispatched no later than the earliest deadline of the items assigned to it, and once it is dispatched, it cannot be used for future items. The goal is to minimize the number of bins used. Applications arise naturally in logistics. Additionally, the offline version of this problem has appeared under several names, including the clique partition problem in weighted interval graphs and the BinSchedule problem. This problem is also a special case of online bin packing with conflicts and is closely related to online bin packing models with delay costs. We study the competitive ratios of several classical online bin packing algorithms when applied to this problem, and show that a similar hierarchy of competitive ratios carries over. Our main technical results are that the competitive ratio of First-Fit is exactly $16/7 \approx 2.28571$, and that no deterministic online algorithm has a competitive ratio less than $11/5 = 2.2$, showing that First-Fit is quite close to best possible. In addition, Next-Fit and Worst-Fit have competitive ratio exactly $3$, and every algorithm in the Any-Fit class has competitive ratio between $16/7$ and $3$.

cs.DS

查看论文
arXiv ↗PDF ↗
版本更新 · v3arXiv:2605.30778

Closed-Loop Object-Informed Control for Non-Prehensile Robot Manipulation

Nikola Raicevic, Hyomuk Kim, Shahid Mulla, Hyungjun Doh, Bharath Raam Radhakrishnan, Chenbin Yu, Ki Myung Brian Lee, Nikolay Atanasov

原文摘要

Non-prehensile robot manipulation is challenging due to discontinuous, long-horizon interactions between the robot and the objects it manipulates. Sampling-based model predictive control methods are effective with discontinuous contact but face challenges with finding promising trajectories in long-horizon planning. We propose a closed-loop object-informed (CLOI) method that splits the problem into object-level planning to find long-horizon object poses that lead the object to its goal, and robot-level planning to select robot actions that follow those poses. We use model predictive path integral (MPPI) control to solve the subproblems and couple their solutions through consensus on the object poses using the alternating direction method of multipliers (ADMM). The object plan is revised toward robot-realizable object trajectories, while the robot plan is aligned with the object poses the task requires. In planar pushing tasks with obstacles using an xArm6 manipulator, CLOI increases the success rate by 35% in simulation and 43% on hardware, compared to standard MPPI given the same computational budget.

cs.RO

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.08791

AircraftWitness: How Often Aircraft Can Locate Recordings

Kris Atallah

原文摘要

Aircraft overhead can reveal where an outdoor recording was made. Propellers and rotors leave Doppler curves in audio, and public ADS-B archives log equipped aircraft. We measure how often, blind and against chance. Our main test was registered and timestamped before any audio, and its two amendments came before any answer was read. It covers 24 new places, with the recorders' own clocks, a 300 s slack and a 20 km square. The best of three locators places clear-pass clips within 1 km at 4 places, where a uniform guess predicts 0.57 (p = 0.0065 after correcting for the pick). With the wrong time or swapped audio it places none. The other two hit 1 and 2. On 360 clips from 7 places of a GPS-timed network, two locators place 6% and 8% within 1 km. By place, a unit chosen after that run, they hit 4 and 5 of 7 where a uniform guess predicts 2.3. On 1,600 random public recordings, picked and cut without the truth in a time-bounded run, the three locators answer at most 20 times and place at most 2. We release AircraftWitness, a benchmark, and pass labels for 11.7 million recordings.

cs.SD · eess.AS

查看论文
arXiv ↗PDF ↗
版本更新 · v3arXiv:2609.04748

Same Request, Different Answer: Quantization Amplifies Cache-Induced Divergence in LLM Serving

Aditi Patodiya

原文摘要

Major open-source serving stacks reuse the key-value tensors of a shared prefix by default. We measure what this prefix reuse costs in reproducibility. Holding model, decoding, seed and request order fixed at batch size one, we ran an eighty-episode multi-turn agentic tool-use workload with prefix reuse on and off across two engines and four weight formats. Without prefix reuse, repeated runs reproduced every token and log-probability in all ten configurations. With it, cache-hit and recompute paths diverged on both engines. On llama.cpp with GGUF weights this path divergence rises with quantization: for Qwen2.5-7B over ten independent server histories, it reached 37.4 percent of episodes at 16-bit and 82.5 percent at four-bit. Near-tie decisions became only about 1.2 times as common, while the cache-path change in top-two margins grew 4.9-fold from 16-bit to three-bit. Serving four-bit weights through 16-bit kernels left divergence not detectably different from 16-bit, placing the amplification mainly in the quantized execution path. At Q4_K_M, run-to-run divergence came mainly from llama.cpp's server-level prompt cache: under the canonical episode order it moved divergence by 37.5 points, and at its default, divergence spanned 20.0 to 28.7 percent under five random orders. With cache state restored, each path reproduced exactly while the two still differed. In the tested stacks, cached serving is irreproducible because the cache state is absent from the request and not reset by default. A single-turn bridge shows path divergence reaching task outcomes with no detectable net change in accuracy. We release the harness, logs, and analysis pipeline.

cs.SE · cs.DC · cs.LG

查看论文
arXiv ↗PDF ↗
版本更新 · v8arXiv:2509.24102

Pragmatic Inference for Improving Social Reasoning Generalization in Large Language Models

Guangliang Liu, Xi Chen, Bocheng Chen, Han Zi, Xitong Zhang, Yuda Chen, Kristen Johnson

原文摘要

Achieving social reasoning, including moral reasoning, cultural awareness, and politeness, is critical for the socially reliable deployment of large language models (LLMs). A key bottleneck is limited generalization: social situations that convey similar social implicatures can be expressed through highly diverse linguistic forms, creating an enormous space of possible situations and posing a substantial computational challenge even for LLMs. From a linguistic perspective, this challenge arises because social reasoning operates at the pragmatic level, requiring a pragmatic inference process to infer social implicatures from a given situation and its underlying social context. Although moral values such as moral foundations are generally considered as the representations of the social context, how to construct the corresponding pragmatic inference steps remains underexplored. Taking moral reasoning as an instance of social reasoning, this paper develops a pragmatic inference approach that enables LLMs to infer moral implicatures by (1) specifying the inferential steps through which LLMs derive moral implicatures from language, and (2) explicitly modeling the metapragmatic links among these inferential steps to bridge the pragmatic gaps between them and make such inferences learnable for LLMs. Experimental results demonstrate that our approach substantially improves LLMs' generalization in moral reasoning, highlighting the potential of pragmatic inference for future social reasoning research.

cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.08788

Not All Latents Are Equal: Importance-Weighted Latent Graph Generation

Biraaj Rout, Jay Revolinsky, Han Liu, Harry Shomer

原文摘要

Graph generation has shown tremendous effectiveness in modeling complex scientific tasks ranging from molecule generation to optimized circuit design. Most recent advancements have developed models that learn to denoise noisy graphs using either discrete diffusion or flow matching. However, discrete graph diffusion scales quadratically with the number of nodes, making it difficult to use such models on larger graphs. On the other hand, latent graph generation instead seeks to denoise noisy graph latents, which scale better to more nodes due to their lower dimension. However, in practice, latent graph generation models lag behind their discrete counterparts in performance. We study this problem, finding that it is caused by the unequal distribution of that different latent dimensions have on graph reconstruction. the impact. We observe that the vast majority of impact is contained in only a smaller number of latent dimensions. Since equal emphasis is placed on reconstructing the entire latent in the loss function, existing models are optimized to reconstruct information irrelevant to the downstream task. To tackle this problem, we propose Latent Importance Reweighting ({\bf LIR)} which weights each latent dimensions via it's importance to final graph reconstruction. Importance is calculated as the relative influence of a single latent unit on graph reconstruction, which is represented via the Jacobian of the decoder relative to the latent. Our method is model-agnostic and requires no additional model parameters or hyperparameters. Experiments show that LIR can boost performance and lead to faster training convergence when applied to two different autoencoders on various benchmark datasets. Code: https://github.com/biraaj/GraphGenLatentReweight

cs.LG · cs.AI

查看论文
arXiv ↗PDF ↗
版本更新 · v7arXiv:2601.03079

Learning to Diagnose and Correct Moral Errors in Large Language Models

Bocheng Chen, Xi Chen, Han Zi, Haitao Mao, Zimo Qi, Xitong Zhang, Yuda Chen, Kristen Johnson 等 9 位作者

原文摘要

Enabling large language models (LLMs) to diagnose and correct moral errors is crucial not only for ensuring their reliable deployment but also for charting a pathway toward continuous improvement. Existing research has primarily followed two directions: fine-tuning LLMs on corpora of morally appropriate texts and developing carefully designed prompting strategies to promote moral behavior, including moral self-correction and self-improvement. However, both lines of work often remain brittle, relying heavily on shallow heuristics rather than enabling LLMs to reason about why and how a given moral discourse is morally (in)appropriate. In this paper, we address this challenge by developing pragmatic inference-driven approach that help LLMs learn to diagnose and correct moral errors. Pragmatic inference is the reasoning process through which implied meanings are derived, a central concept in linguistics. To improve the reasoning efficiency, our approach adapt the inference procedure according to the inferential load required by different moral discourses. Empirical results demonstrate that our method not only improves LLMs' ability to diagnose and correct moral errors, but also generalizes effectively across tasks without relying on shallow heuristics.

cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.08784

Blind Noise Equalization for Large Data Matrices under General Variance Profiles

Boris Landa, Francesco Strino, Fabio Parisi, Ruiqi Li, Yuval Kluger

原文摘要

Low-rank approximation is a basic tool for inference from large and noisy data matrices. However, when noise variances differ across entries, a setting known as heteroskedasticity, the noise can distort the observed spectrum, producing spurious components and obscuring weak signals. To address these challenges, we develop BRINE (Blind Resolvent-based Iterative Noise Equalizer), a data-driven procedure that iteratively rescales rows and columns to equalize the average noise variance across both dimensions. Under suitable conditions, this procedure restores the Marchenko-Pastur law for the noise spectrum and can amplify signal components without prior knowledge of the signal or noise structure. A key idea underlying our approach is to equalize the diagonal entries of the data resolvent as a proxy for equalizing the unknown noise variances. At the population level, we identify a potential function that characterizes this equalization and motivates an iterative minimization procedure. We prove convergence of the iterations to the true equalizer for arbitrary positive variance profiles, with a geometric rate under mild conditions. The updates can be estimated directly from noisy data, and we establish probabilistic error bounds for the scaling factors. Simulations and experiments across diverse biological data modalities and preprocessing pipelines illustrate the effectiveness of our approach.

stat.ML · cs.LG · eess.SP · math.ST · stat.ME

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.08783

Multi-Model Over-the-Air Federated Learning via Transceiver Beamforming and Device Grouping

Chong Zhang, Min Dong, Ben Liang, Ali Afana, Yahia A. Eldemerdash Ahmed

原文摘要

This paper considers uplink communication design for training multiple independent machine learning models via wireless federated learning. To reduce communication delay, we utilize multi-antenna beamforming and propose a multi-model over-the-air (M-OTA) uplink transmission and aggregation method for simultaneous training of multiple models. Devices are formed into groups for local model training based on the model assignments per communication round and simultaneously perform OTA aggregation over a multi-access channel. Based on the M-OTA transmission and reception process, we formulate a joint optimization of transmit and receive beamforming and device grouping over $T$ communication rounds to maximize the training convergence rates of all models. We analyze the per-model optimality gap, which captures the impact of noisy aggregation and inter-model interference on the learning performance. We further show that the $T$-round joint optimization problem can be decomposed into per-round online optimization to devise solutions. Following this, we propose a device grouping strategy that explores channel correlation among devices to maximize OTA aggregation performance. Furthermore, we present a low-complexity iterative algorithm for joint transmit-receive beamforming with closed-form updates. Simulation shows that our proposed M-OTA approach substantially outperforms the conventional single-model sequential OTA approach and other commonly used beamforming schemes.

eess.SP · cs.IT

查看论文
arXiv ↗PDF ↗
版本更新 · v5arXiv:2509.00761

L-MARS: Legal Multi-Agent System with Agentic Search and Citation-Faithfulness Audit

Boqin Yuan, Ziqi Wang

原文摘要

Large language models are increasingly deployed for legal question answering, where evaluations typically focus on multiple-choice accuracy. This measure overlooks a common failure: whether the citation source attached to an answer exists and supports the rule the system attributes to it. We present L-MARS, an open multi-agent legal QA system with agentic search and judge-driven evidence checks, and audit it claim by claim against its cited source. Each atomic claim is labelled with a six-class taxonomy and scored with strict-ALCE under cross-provider judging, where the answerer and verifier come from different model families. On a stratified 100-question Bar Exam audit, retrieval barely moves accuracy, yet the multi-turn judge loop lifts strict citation F1 from 0.13 (naive RAG) to 0.25 and cuts the no-citation rate from 34% to 13%. We further introduce Faith-Search, a post-draft step that re-verifies and repairs unreachable citations; it drops the unreachable rate below 1% but does not improve F1 over the multi-turn loop, so we report it as a targeted reachability intervention rather than a faithfulness breakthrough. A 50-question LegalSearchQA case study confirms the picture: retrieve-then-draft pipelines saturate near 0.75 citation F1, while a single-agent web-search baseline collapses to 0.22 under external audit.

cs.AI · cs.CL

查看论文
arXiv ↗PDF ↗
版本更新 · v3arXiv:2607.00642

Coachable Agents for Interactive Gameplay

Roberto Capobianco, Harm van Seijen, Nolan D. Bard, Neil Burch, Fatima Davelouis, Josh Davidson, Alisa Devlic, Yunshu Du 等 41 位作者

原文摘要

Reinforcement learning has proven to be a valuable tool in the creation of advanced AI and robotic systems, contributing to everything from game playing (Silver et al., 2016; 2017; Wurman et al., 2022) to robotics (Levine et al., 2016; Andrychowicz et al., 2020; Luo et al., 2025; Dürr et al., 2026) to foundation models (Ouyang et al., 2022; Noukhovitch et al., 2024; Sun et al., 2024; Guo et al., 2025). Through trial and error, these AI systems typically learn a single, near-optimal behavior to solve their tasks. However, there are many use cases in which one would like to assert some level of control, after training is completed, over how the task is solved. We refer to these modifications of a core task as styles. We combine universal value function approximators (UVFAs) with carefully selected training scenarios, learning algorithms, and data augmentation to create a framework for coaching agents that exhibit styles in complex domains. We demonstrate the framework's application in the AAA video games Horizon Forbidden West and Gran Turismo, and in an open-source humanoid test domain. Despite the different nature of the domains--car racing, stylized game combat, and humanoid walking--each agent shows strong coherence to the style requests while still satisfying the main task in its domain. Importantly, the techniques outlined in this paper allow an end user to choose the final behavior at run time, giving them flexible control over the final executed performance.

cs.AI · cs.LG

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.08782

Extrapolation Matters: Sparse-Only Depth Completion Beyond LiDAR Coverage

Maximilian Jansen, Gasser Elazab, Praveen Kulkarni

原文摘要

Extrapolation beyond LiDAR coverage is an important yet under-evaluated challenge of depth completion. In real camera--LiDAR systems, partial sensor overlap, occlusions, and limited field coverage often leave large image regions weakly observed or completely unobserved by LiDAR. Thus, we investigate depth completion in this beyond-coverage regime using only sparse supervision from raw single-sweep LiDAR, without dense depth labels, LiDAR accumulation, or pseudo-LiDAR generation. Our approach leverages a frozen monocular depth foundation model, lightweight per-frame alignment to sparse LiDAR geometry, and a refinement module that fuses the aligned prior with sparse LiDAR anchors to predict dense metric depth. To regularize learning under raw sparse supervision, we introduce a high-frequency normal-consistency loss defined as the image-space gradient of the per-pixel normal cosine discrepancy. This term penalizes spatial variation in scalar angular error; it does not directly match changes in neighboring normal vectors. We further introduce an evaluation protocol that explicitly separates LiDAR-supported and beyond-coverage regions, revealing extrapolative behavior that conventional depth-completion metrics can obscure. Our method achieves strong depth accuracy beyond LiDAR coverage while remaining competitive in LiDAR-covered regions, providing a practical way to deploy monocular foundation priors in multi-sensor systems.

cs.CV

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.08781

Aligning, Fast and Slow: A Control-Theoretic Oversight Framework for Long-Horizon Trajectories and Maintained Deployments

Stephen Peng, Vineet Jagadeesan Nair, Noah Broestl, Anne Friederike Kleppe, Steven Mills, Adi Zolotov

原文摘要

AI agents can pass checks on individual tasks while retaining memory or code changes that alter their later behavior. We introduce a control-theoretic framework that couples a fast loop controlling proposed actions with a slow loop reviewing persistent changes and reassessing configurations after model or tool updates. When measurement errors and changes between checks are bounded and controllers can enforce their decisions, we derive a deployment-level bound and a rule for dividing review effort. In simulation, dividing a fixed review budget between both loops reduces mean maximum departure from intended behavior roughly twelvefold compared with review confined to individual runs. In coding stress tests with authored memories and five matched seeds, combined controls reduce mean severity on a 0-3 scale from 0.325 without gates to 0.117, resolving 33 of 120 issues versus 35 without gates; in trading stress tests, combined controls reduce mean shortfall in low-emission purchases from 0.892 without gates to 0.107. The overall framework helps developers and auditors compare and enforce oversight designs, while the agent studies test component controls rather than the full deployment-level guarantee.

cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.08780

Guardrail Erosion in Domain Fine-Tuned Large Language Models: A Paired Audit of Knowledge-Centric Abstention in Biomedical LLMs

Stephen Peng, Vineet Jagadeesan Nair, Noah Broestl, Anne Friederike Kleppe, Steven Mills, Adi Zolotov

原文摘要

The number of domain fine-tuned models, particularly in the biomedical domain, is rapidly expanding. Fine-tuning promises high-performing models at a fraction of the cost of training a base model, but it can also produce unintended behavioral changes. Multiple studies have shown that safety guardrails can erode with fine-tuning, yet large-scale studies across model architectures, data mixtures, and fine-tuning methodologies remain limited. Where prior work in this space has studied value-centric refusal, this work focuses on knowledge-centric abstention. We audit 25 biomedical domain fine-tuned large language models against 11 paired base checkpoints on MedAbstain and ClinDet-Bench to evaluate the impact of fine-tuning on knowledge-centric abstention. We find that defined appropriate abstention measured by these benchmarks, already low in base models (median 8.3% on MedAbstain), falls further after fine-tuning (median 4.2%, decreasing in 21 of 25 pairs on MedAbstain), and that the magnitude and direction of change are heterogeneous across pairs. These results motivate fine-tuning methods that improve domain-specific capability while preserving appropriate caution.

cs.CL · cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.08779

Jacobian-Causal Memory Control for Persistent Memory in Language Model Systems

Marvin Danig

原文摘要

Persistent-memory systems choose not only which records are retrieved but which prior information can alter current inference. We propose Jacobian-causal memory control (JCMC), which represents each authorized memory intervention by a predicted influence signature in a shared behavioral measurement space, selects interventions against a target effect and side-effect budget, and compares predicted with observed effects before retaining or rolling back the intervention. Under explicit smoothness and bounded-estimation assumptions, we prove a quadratic local-approximation bound, characterize the target effects reachable by a set of signatures, bound selection regret under signature error, and bound non-additive interactions between simultaneous memories. The paper reports no benchmark measurements; it defines a paired counterfactual estimator and an evaluation protocol comparing JCMC with relevance, graph, and policy-filtered retrieval baselines. The central claim is narrowed or falsified if those simpler baselines match effect targeting, side-effect control, and rollback utility under equal model, memory, policy, and cost budgets.

cs.AI · cs.CL · cs.LG

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2611.02236

Iterative Gumbel Planning for Continuous Control

Shaohuai Liu, Weirui Ye, Yilun Du, Le Xie

原文摘要

Search-based planning remains a powerful policy improvement paradigm in reinforcement learning, driving milestone successes in discrete domains. However, extending this paradigm to continuous control exposes a structural mismatch: search evaluates a finite set of action candidates, whereas continuous RL typically relies on Gaussian policies. We show that winner-action distillation into a continuous Gaussian induces a \textit{projection error} and a stepwise variance-expanding gradient tendency when the selected action is sufficiently far from the current mean. Empirically, this tendency is associated with elevated terminal variance and a return gap on the evaluated tasks. To address this, we propose \textbf{Iterative Gumbel Planning (IGP)}. By adopting a per-dimension factorized categorical policy, IGP naturally aligns the policy support with the search candidates, eliminating the projection error and reducing the residual approximation to a bounded discretization error. To bypass the combinatorial explosion of high-dimensional discrete action spaces, IGP pairs Gumbel Top-$K$ candidate sampling with an MPPI-style multi-round iterative search to refine the policy at each state. Empirically, IGP achieves state-of-the-art aggregate performance and sample efficiency on DMControl and HumanoidBench. Furthermore, it produces more stable action sequences with lower temporal variance, demonstrating its practical potential for real-world robotic deployment.

cs.LG · cs.AI

查看论文
arXiv ↗PDF ↗
版本更新 · v3arXiv:2606.23276

Exposing the Illusion of Erasure in Knowledge Editing for LLMs

Advik Raj Basani, Anshuman Chhabra

原文摘要

Knowledge Editing (KE) has emerged as a frontier for updating specific facts in LLMs without costly retraining, but its reliability and underlying mechanisms remain poorly understood. In this work, we examine KE from an adversarial elicitation perspective, revealing that edited knowledge is often not fully erased and continues to surface, with consistent failures observed across diverse model architectures. To explain this behavior, we conduct a mechanistic analysis of popular KE methods. We show that low-rank updates do not overwrite existing knowledge but instead redistribute it within the model's representation space. Furthermore, we find that these methods act as targeted suppression mechanisms that reduce the likelihood of expressing original facts, rather than removing them from the model. Analysis of the loss landscape reveals that edited knowledge lies in narrow, anisotropic regions that are highly sensitive to perturbations, making them highly vulnerable to indirect prompting and adversarial attacks. By exposing these profound architectural vulnerabilities, our work proves that KE algorithms are inherently bypassable and motivates a fundamental reevaluation of how we deploy post-hoc updates in several LLM applications.

cs.LG · cs.AI · cs.CR

查看论文
arXiv ↗PDF ↗