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

+979新增论文
+781其中 AI 相关
+119已有论文修订
+95新增 AI 导读

arXiv 最近完成扫描:2026-11-20 08:48 · 全站最近入库:2026-11-20 19:58各信源进度 →

统计口径与数据新鲜度

统计 2026-11-20 08:14 — 2026-11-21 08:14(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:2026-11-20 08:48最近发现新增版本:2026-11-20 19:58库内论文最新更新时间:2026-11-20 02:59

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

28,961 已收录论文24,030 AI 相关8,912 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 21,358 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-20 08:46(北京时间)
版本更新 · v2arXiv:2601.21359

Graph-Free Root Cause Analysis

Luan Pham

原文摘要

Failures in complex systems demand rapid Root Cause Analysis (RCA) to prevent cascading damage. Graph-free RCA methods typically assume that the root cause has the highest anomaly score. This assumption fails when faults propagate, because a small deviation at the root cause can grow into a larger anomaly downstream. Our key observation is that a root cause deviates in both its internal properties (e.g., resource usage) and external properties (e.g., latency), whereas components affected by propagation deviate only in external properties. We propose CIE, a framework that ranks components by combining their internal and external anomaly scores. We formalize a class of component-based systems and state conditions under which CIE provably ranks the root cause first. The guarantee holds under arbitrary amplification for a conjunctive score, and under bounded amplification and a stronger root-cause condition for the default additive score. On 735 failures in nine datasets from three microservice systems, CIE achieves 68% Top-1 accuracy averaged over 11 fault types, a 21% relative improvement over the best of twelve baselines, while requiring 8 ms per diagnosis.

cs.LG · cs.SE

版本更新 · v3arXiv:2603.25507

Lightweight GenAI for Network Traffic Generation: Fidelity, Augmentation, and Classification

Giampaolo Bovenzi, Domenico Ciuonzo, Jonatan Krolikowski, Antonio Montieri, Alfredo Nascita, Antonio Pescapè, Dario Rossi

原文摘要

Network Traffic Classification (NTC) increasingly relies on data-driven models, yet its practical deployment is often constrained by limited labeled data, strict privacy requirements, and the cost of collecting representative traffic traces. While Network Traffic Generation (NTG) provides an effective means to mitigate data scarcity, conventional generative methods struggle to model the complex temporal dynamics of modern traffic and often incur high computational costs. In this article, we investigate lightweight Generative Artificial Intelligence (GenAI) architectures for practical NTG. Rather than generating raw packet bytes or relying on large foundation models, we synthesize compact flow-level traffic representations derived from early packet-header information, enabling transformer-based, state-space, and diffusion models with only a few million parameters. We present a modular GenAI pipeline for NTG and evaluate it along four complementary axes: (i) synthetic traffic fidelity, (ii) synthetic-only training for reduced raw-data sharing in NTC, (iii) data augmentation under low-data regimes, and (iv) computational efficiency. Experiments on two heterogeneous datasets show that lightweight transformer-based and state-space models preserve both static and temporal traffic characteristics, while providing useful synthetic data for downstream NTC. Among them, transformer-based models offer the best fidelity-efficiency trade-off, combining high-quality traffic generation with moderate computational overhead.

cs.NI · cs.AI · cs.LG

新论文 · v1arXiv:2611.15577

Orthogonal Code-Based Parallel Signaling for PLCA in Shared-Medium Ethernet

Mohamed Elattar, Ahmed Abdelkader, Zhe Lou, Georg Carle

原文摘要

Physical Layer Collision Avoidance (PLCA) enables collision-free channel access in shared-medium Ethernet by granting each node an exclusive transmission opportunity (TO) in a sequential, cyclic manner. This sequential, node-based access structure, however, is the root cause of PLCA's fundamental limitations, causing access latency to grow with network size and idle TOs to accumulate when traffic is sparse, a limitation inherited by existing PLCA extensions. To address this, we propose OC-PLCA, an orthogonal code-based parallel signaling mechanism for PLCA. Rather than signaling sequentially, all nodes with pending traffic transmit their unique Walsh-Hadamard code simultaneously to signal their intent to transmit. Since these codes are mutually orthogonal, each node independently identifies the complete set of active nodes, allowing the transmission cycle to include only active nodes and eliminating idle TOs entirely. We further investigate the impact of quantization on active node detection, and propose a grouping-based extension that preserves unambiguous detection for networks of an arbitrary number of nodes. Theoretical analysis shows that OC-PLCA reduces signaling overhead by 88% for an 8-node network, with the reduction growing as network size increases. Simulation results confirm significant reductions in observed access latency across a range of network loads and heterogeneous node activity.

cs.NI

新论文 · v1arXiv:2611.15576

Can Model Answer Probabilities Represent Survey Populations? Conditional Prediction and Joint Validity in Jev Survey Simulation

Zengqing Wu, Chuan Xiao

原文摘要

Survey simulations require realistic combinations of answers, but predictive accuracy alone does not establish that a model can generate them. We evaluate the option probabilities returned by Jev, a TypeSafe decision model, using linked responses to 20 binary items from five archived survey waves. We compare population-only prediction, prediction given recorded answers, and sequential completion that uses generated answers to predict later responses. Joint comparisons hold the starting population information fixed, and two three-question cases are completed in both question orders. We evaluate predictive scores, answer frequencies, and associations against the survey records. Recorded answers can improve predictive scores while associations remain distorted. Large conditional errors also occur. For example, among men who described themselves as having traditional values, the model predicted that 16% would see men and women as similar in workplace abilities, compared with 56% in the survey. Sequential completion has mixed effects on joint validity, and reversing question order produces different joint distributions in both tested cases. Order checks and classical marginal bounds clarify what can be assessed before linked validation answers are available. The findings support validating the joint events a simulation uses and separating empirical constraints from model assumptions. The evidence concerns selected binary tasks and one model version.

stat.AP · cs.AI · cs.CY

新论文 · v1arXiv:2611.15574

RipBench: A Unified Benchmark for Multi-Level Rip Current Detection, Classification and Segmentation

Andrei Dumitriu, Aakash Ralhan, Florin Miron, Florin Tatui, Radu Tudor Ionescu, Radu Timofte

原文摘要

Rip currents are a serious and often under-addressed threat to beach safety and the leading cause of coastal drownings worldwide. They are difficult to detect due to their amorphous structure, similarity to the background, and large variability in viewpoints and environments. Despite their dangers and growing interest in automated detection, existing research remains fragmented across datasets, metrics, models, and approaches, leading to an incomplete understanding of model performance. We introduce RipBench, a unified benchmark that enables controlled evaluation of rip current detection spanning multiple tasks, including classification, axis-aligned and oriented object detection, and instance, semantic, and panoptic segmentation, on the same data with standardized splits. This allows direct comparison of model performance on all levels of visual abstraction. Across $304$ videos ($303{,}491$ frames, $51{,}011$ of them annotated for all six tasks) collected from diverse coastlines, our results expose a clear performance gap between coarse recognition and precise spatial understanding: frame-level classification reaches $F_2 = 0.93$, while the best detector and the best instance segmenter reach $0.72$ and $0.66$, with model rankings that change from task to task. Video-level confidence intervals, a blind inter-annotator study of the spatial labels, stratified analyses by viewpoint and visual signature, and video views with temporal baselines (single-seed, short-schedule reference runs) accompany the benchmark. All tasks are supported with carefully curated annotations and evaluated using standard metrics, with a focus on the $F_2$ score to emphasize recall in this safety-critical setting. RipBench, along with multiple baselines per task, is publicly available at https://RipBench.ai to support progress in real-world, multi-task vision for beach safety.

cs.CV

版本更新 · v3arXiv:2609.02753

The Import Tax: A Longitudinal Measurement of Startup Cost in the Python Ecosystem

Trinath Sai Subhash Reddy Pittala

原文摘要

Python programs pay for their imports at every process start, a cost that is invisible in steady-state benchmarks but dominant for command-line tools, test workers, and serverless cold starts. Python 3.15 adds explicit lazy imports (PEP 810) on largely anecdotal evidence. We measure the import cost of the 500 most-downloaded PyPI packages, every quarterly release since mid-2021, under six CPython versions (3.9-3.14) on two platforms (Apple M5/macOS and Intel Xeon/Linux), 66,816 cells in all, with targeted experiments that isolate individual effects. Values are for macOS (Linux in parentheses). Import cost is heavily skewed (median 12.3 ms, 99th percentile 400 ms). Among packages that import in more than 2 ms, namespace distributions excluded, the median package spends 68% of its import time in the standard library (68% on Linux with the interpreter on local storage, 79% on network storage). Precompiling bytecode removes almost all of the cold-start penalty on Linux and 92% of it for the median package on macOS, but only a third of the summed macOS penalty: packages with native libraries pay a security assessment of each library that the system has not seen before, again in every fresh environment. A package's own import code grows by a median 4.0% (5.0%) per year; installing each release with the dependencies of its time changes its trend by a median of only 0.15 (0.35) percentage points (CPython 3.9). CPython 3.14 runs package code faster than 3.9 but imports more and slower standard-library modules, a net 1.16x (1.07x) slowdown. Submodules that dependents import beyond the top level add nothing for the median package and at most 11.7 (49.5) ms for nine packages in ten. On CPython 3.15, global lazy mode cuts the imports that dependents perform to 0.41x (0.39x) of their cost, and 38 of 461 packages fail under it, many only on first use of a name. Harness, code, and dataset are public.

cs.SE

版本更新 · v2arXiv:2608.24763

MoTE: Mixture of Task Experts for Multi-Task Video Understanding

Muhammad Asad Ali, Umar Khan, Nadia Robertini, Didier Stricker

原文摘要

Procedural video-language models must solve heterogeneous tasks from the same visual evidence, including action recognition, forecasting, and procedure prediction. Dense transformer decoders share the same feed-forward networks across tasks, which can entangle task behavior and make controlled capability expansion difficult. Sparse Mixture-of-Experts (MoE) decoders provide conditional computation, but token-level learned routing is not naturally aligned with task-level procedural objectives. We propose MoTE (Mixture of Task Experts), a decoder architecture that converts large language model feed-forward networks into task-specific experts while keeping the multimodal backbone shared. Each example follows one sample-level task route, so active task-expert computation remains independent of the number of stored task experts. We instantiate this design as VideoLLM-MoTE and evaluate it on five COIN benchmarks using explicit task routes. The five-expert model activates ~2B LLM parameters per sample and achieves higher average top-1 accuracy than recent VideoLLM baselines. Under the same expert topology, it improves over dense all-expert activation and learned sparse-routing controls. These results show that task-structured routing provides an interpretable and compute-efficient decoder alternative for multi-task video-language learning.

cs.CV · cs.LG

版本更新 · v2arXiv:2605.07665

Debiased Counterfactual Generation via Flow Matching from Observations

Hugh Dance, Johnny Xi, Peter Orbanz, Benjamin Bloem-Reddy

原文摘要

Estimating the full counterfactual distribution under an intervention is central to treatment risk assessment and counterfactual generation tasks. Existing approaches model the counterfactual distribution as a standalone generative target, without exploiting its relationship to the observational data. In this work, we show that under standard assumptions, observational distributions and marginal counterfactual distributions over outcomes are tightly linked: they have identical support and tail behavior, remain statistically close under weak confounding, and share any outcome features which are invariant to confounders. These properties motivate learning counterfactual distributions not from scratch, but via a deconfounding flow from the observational distribution. We formulate this problem via flow-matching and derive a semiparametrically efficient estimator based on a novel efficient influence function correction. We subsequently extend our estimator to target minimal-energy flows in high-dimensions, which we show can be especially simple targets between observational and counterfactual distributions. In experiments, deconfounding flows outperform existing debiased counterfactual distribution estimators, while also mitigating known failure modes of flow-based methods.

stat.ML · cs.LG

新论文 · v1arXiv:2611.15572

Learning a calibrated decision head from logged bandit feedback

Ye Yuan

原文摘要

A decision head reads a state and a list of options and returns a probability per option in one forward pass. Once deployed, it usually observes only whether the option it chose was right: logged bandit feedback. We adapt such a head, LoRA on Qwen3-0.6B-Base and 1.7B-Base, from that feedback and report three results. First, the Brier or log score of the sampled action's own probability is not a usable REINFORCE reward: every wrong action earns at least gold's reward, with a mean baseline the lowest-probability wrong action gets the largest advantage, and at two options the signal is zero; in one run it cut accuracy on ordered-scale questions from .636 to .111. Scoring the probability as a loss by the pathwise gradient avoids the collapse, as do the REINFORCE rewards r and r minus p. Second, with full labels, none of four tested bandit arms beat cross-entropy plus one temperature on mean expected calibration error (ECE) over four unseen label schemas. Third, from logged feedback alone, Chosen-Action Proper Scoring (CAPS), the direct method with a proper score on the chosen action's probability, beats fine-tuning on confirmed answers only, which discards negatives: on Banking77 by 2.6 accuracy points, .053 lower ECE and lower selective risk (three seeds against one baseline seed), and by 21.7 points on one seed on emotion data with an often-wrong logger. At N 5000, CAPS stays 1.5 points below full-label training at 0.6B and .6 points at 1.7B against one full-label seed, and propensity weights add nothing on this log. Under feedback flipped at a known rate, one seed per rate, CAPS's confidence tracks the noisy channel on average, backward correction recovers most of the clean calibration, and early stopping on held-out logged outcomes prevented memorising noisy draws in all three runs. Code, adapters, logs and per-record results are public.

cs.LG · cs.CL

新论文 · v1arXiv:2611.15568

Can We Make an LLM Dumber, or Is It Just Playing Dumb? An Empirical Study of Believable Performance Throttling

Chengguang Gan, Yunhao Liang, Hanjun Wei, QingHao Zhang, Shiwen Ni

原文摘要

A benchmark score is commonly read as a proxy for model capability. We study the inverse problem: can a model be driven to a target score, and do its failures resemble those of a naturally weaker model? We operationalize believable performance reduction along three axes: target calibration, item-difficulty structure, and error-distribution similarity to a real weak reference. We introduce CapLadder, an IRT-calibrated suite of 1,985 items spanning mathematics, code, multiple-choice reasoning, and tool use. A simple weak-teacher LoRA dial reaches unseen target levels on reasoning with a 1.9-point held-out calibration error, whereas prompting, temperature, and weight noise cannot combine comparable range and precision. Yet on multiple choice, the dial converges to an answer-token attractor: up to 97% of its newly introduced errors land on one letter, with the attractor changing from D in Qwen to A in Llama. Logit interventions identify a dominant near-constant shift: subtracting the estimated shift on held-out items recovers 52 to 59% of lost accuracy and removes the dominant attractor, while option permutation confirms that the bias tracks token position rather than option content. With high inter-annotator agreement ($κ=0.94$), human-coded error-category distributions nonetheless match the weak reference on reasoning (JS = 0.00), showing that local plausibility and aggregate detectability differ. Structured tool calls exhibit more natural degradation in the Qwen family (JS = 0.09), while a globally calibrated dial transfers poorly across benchmarks (14.2-point error on two usable benchmarks). Precise score control, therefore, does not imply a portable or believable reduction in skill.

cs.CL

新论文 · v1arXiv:2611.15566

Observations Drift, Structures Remain: Structural Pretraining with Time Alignment for Electromagnetic Signals

Wenjin Gui, Junyu Shen, Haibo Xu, Yuchuang Sun, Luqing Luo, Wupeng Xie, Qian Huang, Fengxiang Wang 等 10 位作者

原文摘要

Electromagnetic signal modeling is increasingly moving toward self-supervised pretraining on heterogeneous I/Q observations. However, an observed waveform is jointly shaped by signal generation, propagation environments, and acquisition processes, which mix reusable structural regularities with observation-specific variations. Existing methods usually pursue unified pretraining with reconstruction objectives on mixed observations. This heterogeneity creates two key mismatches for reconstruction pretraining: (1) appearance mismatch, induced by propagation-dependent waveform variations; and (2) temporal mismatch, induced by acquisition-dependent sampling coordinates. These mismatches make observation-specific appearances and sampling coordinates easier to reconstruct than stable signal structures, inducing observational shortcuts. We identify this failure mode as observation-biased pretraining. To address this, we propose Structural Pretraining with Time Alignment for heterogeneous EM signals. It mitigates temporal mismatch with RoPE-fs, which maps token positions to sampling-rate-normalized physical-time positions, and reduces reliance on appearance shortcuts by replacing waveform reconstruction with masked prediction of discrete structural units. These units are organized into atomic structural units and emergent structural units that capture variable-length compositional patterns beyond fixed patch boundaries. Across diverse EM benchmarks, our method improves both full-supervision and few-shot adaptation over existing EM pretraining baselines, showing that the learned representations better preserve structural consistency when observations drift in waveform appearance and temporal scale.

cs.LG

版本更新 · v2arXiv:2610.21820

A Closed Form Depth-Width Law for Neural Collapse in Deep Linear Networks

Pankhil Gawade, Kirsten Fischer, Guy Wolf, Ewa Szczurek

原文摘要

How the depth and width of neural networks shape learned representations is difficult to predict from the data and architecture. We use neural collapse (NC) as a probe of the representation geometry and express its first three properties, within-class variability collapse (NC1), simplex class means (NC2) and classifier alignment (NC3), through the learned feature kernel, extending a known identity for NC1 to NC2 and NC3. For deep linear networks with an exchangeable input kernel, the kernel equations yield a geometric law, exact within this theory, for the layerwise decrease of NC1. At large width $N$ the accumulated decrease becomes a function of the depth-to-width ratio $L/N$, with its rate determined self-consistently by the data and architecture, yielding a scaling law for representation geometry.

cs.LG

新论文 · v1arXiv:2611.15565

Generalized Inverse Kinematics via Conditional Flow

Kaitlyn Zhang, Tomás Lozano-Pérez, Leslie Pack Kaelbling

原文摘要

Inverse kinematics (IK) is a foundational primitive in robotics, mapping the task-space objectives expressed by higher-level planners to robot configurations that realize them. Traditional IK methods generally treat each query as an isolated pose-to-configuration problem, but planners often need a distribution over valid realizations that accounts for future reachability and preferences such as collision avoidance and stability. We introduce GenGenIK, a generative modeling approach that supports mobile bases, multiple end-effectors and targets, scene awareness, and preferences over robot behavior within a common interface. Experiments show that GenGenIK outperforms existing methods in both solve rate and runtime across tasks such as bimanual lift and flip-up, while allowing flexible control over the sampled distribution and remaining effective under partial scene observations. Website: https://icra-2027-gengenik.github.io/.

cs.RO

版本更新 · v2arXiv:2507.04575

Lilith: Developmental Modular LLMs with Chemical Signaling

Mohid Farooqi, Alejandro Comas-Leon

原文摘要

Current paradigms in Artificial Intelligence rely on layers of feedforward networks which model brain activity at the neuronal level. We conjecture that expanding to the level of multiple brain regions with chemical signaling may be a productive step toward understanding the emergence of consciousness. We propose LILITH, a novel architecture that combines developmental training of modular language models with brain-inspired token-based communication protocols, mirroring chemical signaling in the brain. Our approach models distinct brain regions as specialized LLM modules including thinking, memory, sensory, and regulatory components that communicate through emergent token-based signaling protocols analogous to neurotransmitter networks. Unlike traditional pre-trained systems, LILITH would employ developmental training where untrained LLM architectures learn through simulated life experiences, developing communication pathways and cognitive abilities through environmental interaction and evolutionary optimization. This framework would enable direct empirical investigation of consciousness emergence using Integrated Information Theory metrics while providing unprecedented insight into inter-module signaling patterns during development. By optimizing for consciousness emergence rather than task performance, LILITH could provide insight into different emergent phenomena at multiple levels of neural correlates, contrasting neuronal-level processing with multi-region coordination dynamics. The goal of this paper is to put the idea forward while recognizing the substantial challenges in implementing such a system.

q-bio.NC · cs.AI

版本更新 · v2arXiv:2611.03767

Quantifying out-of-distribution data severity with Wasserstein distances: benchmarking large language models for alloy property prediction

YuJie Liu, Dezhen Xue, Jun Sun, Xiangdong Ding, Zhibin Gao, Turab Lookman

原文摘要

General-purpose Large Language Models (LLMs) increasingly rival domain-specific pipelines on scientific prediction, but their performance under distribution shift is untested. Across six alloy and crystal-structure datasets, we benchmark LLMs against traditional, BERT-based, TabPFN, and text-regression baselines under random cross-validation and six out-of-distribution (OOD) protocols shifting input and target distributions, with severity quantified via a standardized Wasserstein distance. We isolate the LLM's advantage over k-nearest-neighbor (k-NN) retrieval: it beats k-NN on all eight tasks at a fixed 5% reference budget, on seven of eight when independently optimized, and in 72.5% of 120 OOD task--protocol configurations against an optimized baseline. No model family is uniformly best; the LLM is often strongest under severe shift, and error increases with target-space but not representation-space severity. Self-reported LLM confidence tracks both shift and error, more consistently than a density-only control from the same retrieved samples, and a matched-cohort check finds no substantial evidence of memorization.

cs.LG · cond-mat.mtrl-sci

新论文 · v1arXiv:2611.15555

What Conformal Selection Can Certify in CRISPR Off-Target Design:Benchmark Leakage, Free Baselines, and a Finite-Sample Cohort Requirement

Mohammad Malekpouri, Somayeh Lotfi

原文摘要

Conformal selection promises distribution-free, finite-sample control of the error rate of a decision, and has been proposed as a safety layer for machine-learning CRISPR guide design. We ask whether such a certificate is attainable from the validated off-target data that exist. The constraint is arithmetic: a conformal p-value cannot fall below one divided by the calibration cohort size plus one, so a Benjamini-Hochberg selection certifies a single candidate only when that cohort is at least the candidate pool size divided by the tolerated error rate. Public cohorts supply ten validated guides, whereas a pool of two thousand at a ten percent error rate needs twenty thousand; the accepted set is consequently empty in over ninety-six percent of splits. Two benchmarking findings explain why this went unnoticed. The standard CIRCLE-seq file lists ten guides under eighty on-target strings, so grouping cross-validation on the sequence column trains and tests on one guide; corrected, pair-level discrimination and per-guide fidelity collapse. With that leakage removed, a training-free position-weight score and a bare count of near-match genomic sites are not beaten by a trained oracle. We give the corrected protocol, the baselines a new model must beat, and the cohort size a certified pipeline needs.

cs.LG

版本更新 · v3arXiv:2601.21688

XFACTORS: Disentangled Information Bottleneck via Contrastive Supervision

Alexandre Myara, Nicolas Bourriez, Thomas Boyer, Thomas Lemercier, Ihab Bendidi, Auguste Genovesio

原文摘要

Disentangled representation learning aims to map independent factors of variation to independent representation components. On one hand, purely unsupervised approaches have proven successful on fully disentangled synthetic data, but fail to recover semantic factors from real data without strong inductive biases. On the other hand, supervised approaches are unstable and hard to scale to large attribute sets because they rely on adversarial objectives or auxiliary classifiers. We introduce \textsc{XFactors}, a weakly-supervised VAE framework that disentangles and provides explicit control over a chosen set of factors. Building on the Disentangled Information Bottleneck perspective, we decompose the representation into a residual subspace $\mathcal{S}$ and factor-specific subspaces $\mathcal{T}_1,\ldots,\mathcal{T}_K$ and a residual subspace $\mathcal{S}$. Each target factor is encoded in its assigned $\mathcal{T}_i$ through contrastive supervision: an InfoNCE loss pulls together latents sharing the same factor value and pushes apart mismatched pairs. In parallel, KL regularization imposes a Gaussian structure on both $\mathcal{S}$ and the aggregated factor subspaces, organizing the geometry without additional supervision for non-targeted factors and avoiding adversarial training and classifiers. Across multiple datasets, with constant hyperparameters, \textsc{XFactors} achieves state-of-the-art disentanglement scores and yields consistent qualitative factor alignment in the corresponding subspaces, enabling controlled factor swapping via latent replacement. We further demonstrate that our method scales correctly with increasing latent capacity and evaluate it on the real-world dataset CelebA. Our code is available at \href{https://github.com/ICML26-anon/XFactors}{github.com/ICML26-anon/XFactors}.

cs.LG · cs.AI

版本更新 · v2arXiv:2605.12476

Routers Learn the Geometry of Their Experts: Geometric Coupling in Sparse Mixture-of-Experts

Sagi Ahrac, Noya Hochwald, Mor Geva

原文摘要

Sparse Mixture-of-Experts (SMoE) models enable scaling language models efficiently, but training them remains challenging, as routing can collapse onto few experts and auxiliary load-balancing losses can reduce specialization. Motivated by these hurdles, we study how routing decisions in SMoEs are formed mechanistically. First, we reveal a geometric coupling between routers and their corresponding experts. For a given token, the router weights for the selected expert and the expert weights processing it receive gradients along the same input direction, differing only in scalar coefficients. Thus, matched router--expert directions accumulate the same routed token history. This theoretical coupling also appears empirically in routing dynamics. In a $1$B SMoE trained from scratch, higher router scores predict stronger expert neuron preactivations, showing that routing decisions are mirrored inside the selected expert. Next, we analyze the effects of auxiliary load balancing on the router--expert geometric coupling, showing that such losses break this structure by spreading input-directed gradients across router weights, making distinct router directions nearly three times more similar to each other. Last, we demonstrate the centrality of geometric coupling for effective routing with a parameter-free online K-Means router, in which each expert maintains a running average of the hidden states routed to it and tokens are assigned based on cosine similarity. Compared with auxiliary-loss and loss-free balancing, this router achieves the lowest load imbalance with only a modest perplexity increase, indicating that geometric coupling captures a substantial part of what the router learns. Overall, our results explain how routers form assignment geometry that supports an effective division of labor.

cs.LG · cs.CL

新论文 · v1arXiv:2611.15538

Faithful in Steps: Improving Generalization and Citation in RAG via Query Decomposition

Yue Liu, Zhongying Ru, Shimin Di, Jipeng Zhang, Ruiyuan Zhang, Xiaofang Zhou

原文摘要

Retrieval-augment generation is a prevalent strategy to mitigate hallucinations of LLMs. The attributable RAG (RAGQ) generates quotes for its answers. The quotes indicate which input contexts support the RAG to derive the answers, enhancing the answer's verifiability and trustworthiness. However, existing RAGQs exhibit significant degradation when dealing with questions that require multi-hop reasoning and multi-modal understanding, suffering from over-citation, implicit entity identification failure, and poor generalization. In this paper, we propose a novel RAGQ framework, namely QDRAG. QDRAG breaks down the input question into atomic subquestions to identify the implicit entities. Then, the reranker prunes context distractors to eliminate the downstream over-citation. To facilitate query decomposition, we propose two zero-shot approaches: QD-C and QD-R, which guide the QD MLLM to decompose the question based on context knowledge and retrieval rewards, respectively. One interesting finding is that finetuning on the QD task shows better generalizability compared to directly finetuning on the downstream RAGQ task. Experiments on four multi-modal QA benchmarks demonstrate QDRAG's efficacy in grounding answers and generating faithful citations. The framework significantly outperforms all the baselines on both in-domain and out-of-domain tests, even surpassing Gemini-Pro.

cs.CL · cs.AI

版本更新 · v2arXiv:2611.13525

EJAI: Judgment-Preserving Civic AI for Data Center Environmental Justice

Evan Pačes-Wiles, Bryen Alvarez, Sara Gulabani, Akshara Kolipaka, Snehalkumar 'Neil' S. Gaikwad

原文摘要

Communities in the United States and around the world are contesting data centers, the physical infrastructure of AI. These disputes pose environmental justice issues: residents must weigh unevenly distributed burdens using evidence scattered across metrics, corporate claims, policy documents, and public testimony. A formative study with 10 users showed that dashboards and single-agent AI can settle questions before people examine that evidence. Here we present EJAI, a source-grounded system that links an interactive map, retrieved evidence, missing-data statements, and two stakeholder personas that allow people to inspect evidence and form their own policy recommendations. In a small functional benchmark of 21 queries, retrieved evidence supported 98.4% of factual claims and the system acknowledged missing evidence in 32 of 33 cases. Persona alignment carried a grounding cost: the Corporate Executive omitted some evidence that conflicted with its assigned priorities. In a within-subjects study, 15 participants acting as residents prepared for a simulated town hall. Some checked cited AI claims while treating map metrics as official, some made conditional recommendations with safeguards such as water-use disclosure, and some found the personas simplified or biased. Our results suggest that civic AI can support preparation for public reasoning when it makes evidence, gaps, and disagreement open to inspection, and that corpus imbalance and two-sided personas remain open justice concerns.

cs.HC · cs.CL · cs.CY

新论文 · v1arXiv:2611.15533

Analysis choices decide whether untrained networks beat trained ones at V1: a multiverse analysis of model-brain RSA

Nils Leutenegger

原文摘要

Untrained convolutional networks have been reported to rival trained networks as models of visual cortex, and we had reported an untrained-network advantage over backpropagation at V1 in THINGS fMRI, together with a dependence on evaluation resolution. Under a pre-specified ridge readout, the untrained network scored below the trained one at every pooling grid (median r difference -0.043 at g = 4, 0 of 5 seeds positive), and also in MSE skill and per-fold r. In exploratory analyses, across 36 combinations of unit normalisation, distance measure and voxel normalisation (28 distinct analyses), the sign of the V1 contrast depended on the analysis (intervals above 0 in 5 and below 0 in 9 distinct analyses), and in our networks the resolution dependence of the original convention appeared in only one cell. Raw correlation-distance representational similarity analysis (RSA) covaried with the alignment of model and voxel patterns with the unit-mean profile (Spearman 0.97 across our networks, 0.79 across six ImageNet architectures); the advantage was absent from the component orthogonal to the mean profile, and the parallel component alone did not resolve it. On the unweighted angular component of centred patterns, trained networks were ahead at V1 (our network and 5 of 6 ImageNet models) and at LOC (all seven); the norm-weighted angular term was not resolved. A luminance-and-contrast bank scored above every network under ridge (point estimates), and adding network features under a shared penalty did not improve prediction. Intervals are stimulus-bootstrap intervals for three subjects, and 6 pre-stated predictions were not met. We recommend reporting sensitivity to centring over the evaluated stimuli on both sides, reporting components separately, and including a small multiverse and simple baselines.

q-bio.NC · cs.LG

新论文 · v1arXiv:2611.15530

MAGIC: Modality-Aware Gini Inspection for Clinical Hallucination Detection in LMMs

Sachin Acharya, Niyoj Oli, Shishir Adhikari, Aminu Lawal, Yash Raj Shrestha, Prashnna K Gyawali, Binod Bhattarai

原文摘要

Recent advances in medical large multimodal models (LMMs) have brought promising progress to clinical decision support, yet hallucination, the generation of confident but clinically unverifiable outputs, poses a critical threat to the reliability demanded in high-stakes medical settings. Existing hallucination detection methods, predominantly designed for unimodal settings, fail to capture the visual-linguistic dynamics inherent to medical imaging. We formulate the problem of hallucination detection in LMMs as disproportionate distribution of attention weights across the modalities during answer token generation. We quantify modality-specific attention concentration across layers and heads using the Gini coefficient and demonstrate statistically significant redistribution of attention mass between visual and textual tokens in hallucinated responses. We leverage these features to train a logistic regression-based classifier and evaluate it on two medical LMMs across three medical imaging modalities, demonstrating that our method achieves competitive or superior performance against contemporary hallucination detection baselines.

cs.CV · cs.AI

版本更新 · v4arXiv:2605.02171

QuIVer: Efficient Graph Indexing with Two-Bit Construction and Navigation

Wenxuan Xiao, Peidong Zhu, Chengcheng Li, Xu Cao

原文摘要

High-dimensional vector indices must balance construction time, memory, and query throughput. Two-bit comparisons reduce computation and storage, but preserving useful graph connections and search candidates requires scores suited to each stage. We present QuIVer, a graph index that uses a training-free sign-magnitude representation for construction and navigation while reserving original vectors for bounded final reranking. Construction discovers candidates with Hamming scores, orders them with an integer proxy, and defers reverse-edge pruning across parallel insertions. Search uses reconstruction norms and bitwise inner products to order both the frontier and retained candidates, without query calibration or an exact anchor. Experiments cover 21 datasets and compare 30 external baseline modes on three core workloads. QuIVer's validation-selected 95% configurations lie on the measured Pareto frontiers of achieved recall, throughput, index preparation time, and memory, whether memory counts navigation state alone or also the original vectors. On Cohere-5M, QuIVer reaches 95.90% Recall@10 at 3,075 single-thread queries/s, $11.74\times$ the throughput of Lucene BBQ with FP32 reranking at 95.83% recall. Against SymphonyQG at 95.87%, it retains 81% of single-thread and 97% of 32-thread throughput while reducing index preparation time by 54%, total logical memory by 47%, and navigation storage by 87%. An SSD mode with INT8 reranking serves Cohere-5M at about 5 GiB of steady-state process RSS. Controlled crossovers and candidate-coverage diagnostics connect these results to construction and retention quality. Across 21 datasets, high recall holds for most text and image embeddings, while non-negative descriptors, cross-modal queries, and random data expose the limits of the representation.

cs.DB

新论文 · v1arXiv:2611.15529

One Hierarchy, Many Sports: Hierarchy-Aligned Multi-Task Learning for Next-Shot Prediction

Chao-Wei Cai, Min-Te Sun, Sheng Kuang Wu, Hung-Hsuan Chen

原文摘要

Next-shot prediction in racket sports requires learning from event data with different match structures, recorded attributes, and label vocabularies. Table tennis, badminton, and tennis nevertheless share a hierarchy in which shots form rallies and rallies form higher scoring units. We present MT-HTA, a hierarchy-aligned multi-task Transformer that uses this structure to jointly predict the attributes of the next shot. A temporal Transformer captures event progression, a feature-token path models interactions among semantic fields rather than along the time axis, and task-specific attention retrieves information from hierarchical summaries and player context. The architecture supports three- and four-level hierarchies and sport-dependent task sets. Across four datasets from three sports, the final MT-HTA achieves the highest mean task-averaged F1 among the compared main models. Ablations on table tennis support the combination of hierarchical encoding, task attention, and feature-token interaction. Additional modules alter the balance between classification scores and spatial error rather than improving every metric. These results support a reusable architecture for heterogeneous racket-sport event data. We release the code for reproducibility. https://github.com/gndd1221/For-Sports-event-modeling-with-rally-and-score-hierarchical-A-Multi-Task-Hierarchical-Transformer

cs.LG

新论文 · v1arXiv:2611.15526

SCR-Mamba: Skeleton-conditioned Mamba with cross-scale incremental refinement for efficient 3D human pose estimation

Jie Wang, Xuejing Gu, Congzhe Zhang, Zibo Zhao

原文摘要

Mamba-based methods for video-based 3D human pose estimation can model long sequences with linear complexity, but their selective state updates usually lack direct exploitation of skeletal relations. Meanwhile, stacking modeling blocks at the full temporal resolution leads to high computational overhead, while compressing the temporal sequence makes it difficult to retain per-frame details. To address these issues, we propose SCR-Mamba, whose core consists of Skeleton-Conditioned State Modeling (SCSM) and Cross-Scale Incremental Refinement (CSIR). SCSM generates a skeleton context from bone connections and left-right symmetry relations and generates the selective parameters of the spatial and temporal branches conditioned on it, so that structural relations are involved in the writing, propagation, and reading of state information. CSIR constructs multi-scale motion representations through content-adaptive aggregation and progressively refines the retained per-frame features with the increments produced by coarse-scale modeling, integrating long-range motion context while reducing full-resolution computation. On Human3.6M with Stacked Hourglass (SH) 2D detections and 243-frame inputs, SCR-Mamba-L achieves 36.8 mm MPJPE (P1) with 7.72 M parameters and 15.10 G MACs, less than 1/7 of the computational cost of StructMamPose-L with similar accuracy; SCR-Mamba-X further reaches 36.6 mm P1 and 30.9 mm P-MPJPE (P2), achieving state-of-the-art performance. On MPI-INF-3DHP, SCR-Mamba-X achieves a percentage of correct keypoints (PCK) of 99.3%, an area under the curve (AUC) of 88.1%, and an MPJPE of 14.2 mm.

cs.CV

新论文 · v1arXiv:2611.15524

Reading the Table: Sim-to-Real Game State Detection for the Chef's Hat Card Game

Connor Gäde, Matthias Kerzel, Hauke Bünning, Jan-Gerrit Habekost, Mariana Barros, Stefan Wermter

原文摘要

Tabletop games offer a compelling scenario to study the interaction between artificial and human agents, combining social interaction with game state tracking and physical manipulation. To assess the game state, a robot needs to precisely detect closely related game objects, such as different card designs, requiring dedicated training data that is costly to annotate in the real world. Instead, we develop a fully automated synthetic data collection pipeline, using re-projection of known 3D coordinates to generate labels automatically and domain randomization to bridge the reality gap. This approach is used to collect a dataset of 25413 samples to finetune an object detection and a segmentation model, which are combined to assign detected cards to game-relevant zones and infer the game state. When evaluated on a real NICO humanoid robot, our pipeline achieves an F1-score of 0.906 for card-to-zone assignment, thus providing the perceptual foundation for a robot that can autonomously play Chef's Hat against human players.

cs.RO · cs.CV

新论文 · v1arXiv:2611.15522

GHAgentCI: A Dataset of Coding-Agent Invocations in CI Workflow Histories

Daniel Ogenrwot, Jorge Gonzalo Delgado Cervantes, Woojoong Kim, John Businge

原文摘要

Coding agents are increasingly integrated into GitHub Actions to review pull requests, triage issues and modify code. Identifying these integrations is challenging because agentic workflows often resemble ordinary CI files, with evidence appearing in action references, shell commands and workflow conventions. We present GHAgentCI, a dataset of 15K+ workflow histories from 6K+ of 196K+ mined GitHub repositories, comprising 159K+ revisions and 205K+ agent-evidence records linked to workflow triggers and declared execution constraints. Our open-source extractor, cai-miner, detects agent-related evidence using an extensible catalogue. Manual validation of 576 stratified records confirmed 96.1% of dedicated-action matches and 98.8% of GitHub Agentic Workflow matches, with lower rates for other evidence kinds. Exploratory analyses reveal a shift toward command-line and GitHub Agentic Workflow integrations, differences in declared safeguards, and frequent removal of detected agentic workflows. GHAgentCI supports longitudinal studies of coding-agent integration into CI.

cs.SE · cs.AI

新论文 · v1arXiv:2611.15521

DecArena: Evaluating Large-Scale Decentralized Multi-Agent Collaboration

Jiali Cheng, Oleg Rokhlenko, Joo-Kyung Kim, Hadi Amiri, Xingzhi Guo

原文摘要

Emerging personal AI agents raise fundamental scalability and privacy concerns for centralized multi-agent systems (MASs), which rely on a global orchestrator with broad access to agents, resources, and communication. In decentralized environments, global visibility is unavailable. Agents need to proactively discover non-local resources (e.g. tools, skills), and collaborate with other agents to solve tasks. We study decentralized multi-agent collaboration under graph-constrained communication, a natural reflection of human structures. Each agent observes only its local state and can interact only with neighboring agents. We introduce a benchmark with synthetic scalable distributed tool calling with 100K+ tools. Our experiments reveal that recent LLM agents struggle to discover non-local relevant resources and collaborate in decentralized setting. Moreover, standard reinforcement learning fails to reliably overcome this limitation, suggesting that decentralized MAS needs more explicit harness mechanisms for effective collaboration, even using SOTA LLMs. We then present a strong baseline \textbf{GraphHarness}, combining topology-aware warmup with self-improving LLM-based routing. Results show that GraphHarness substantially improves task success (+48.5) and produces higher quality collaboration (+30.6), while using fewer costs (+34.0) than baselines.

cs.MA · cs.AI · cs.CL

新论文 · v1arXiv:2611.15515

Incentive-Responsive Agents: A Principal-Agent Framework, Metric Suite, and Runtime Architecture for Long-Lived Personal AI Agents

Jipu Su

原文摘要

Personal AI agents now persist: they hold accounts, mailboxes, long-term memory, and open-ended task streams, yet the dominant way to improve them remains capability-side: prompt engineering, larger models, longer thinking, and retrieval. We argue that long-lived agents are missing an incentive layer. We formalize an incentive not as text in a prompt but as a runtime contract over the agent's operating point (model tier, test-time compute, memory budget, tool authority, task priority, verification, and compensation) that the agent's policy must causally consume. We develop a principal-agent model with hidden effort, a contractible operating point, and costly verification. We derive (i) a non-consumption invariance principle separating genuine incentives from prompt-level persuasion, (ii) a persistence-as-discount theorem making identity continuity a precondition for sustained effort, (iii) an incentive-capability exchange-rate frontier yielding a per-task routing rule, and (iv) a Goodhart bound for multi-channel verification, plus a three-channel decomposition of mechanism gains (allocation, verification, continuation). A 240-run real-LLM experiment with two models and 60 verifiable tasks shows that prompt-level incentive framing has no detectable effect (p = 0.62), whereas runtime allocation raises success on the hardest task family from 0.60 to 1.00 (p = 0.008) at 4.8x price-weighted cost; the exchange-rate theorem predicts the break-even baseline at 0.21. A 20-seed, 3,000-task simulation shows resource ROI rising from 7.09 to 16.67 and gaming falling from 0.333 to 0.115. Code, simulator, and all results are released; live deployment remains future work.

cs.AI · cs.MA

新论文 · v1arXiv:2611.15514

Real Pairs Are Not Enough: Physically Anchored Supervision for Volumetric CT Super-Resolution

Haoran Liu, Zipei Yan, Mingzhe Liu, Jizhou Li, Guibin Zan

原文摘要

Volumetric CT super-resolution (SR) models are commonly trained on synthetic or physically acquired pairs of low-resolution (LR) and high-resolution (HR) volumes. However, synthetic pairs miss real acquisition effects, whereas registered physical pairs retain inter-scan noise, artifact, and microstructure differences---causing direct voxelwise regression, and even naive synthetic-real mixing, to oversmooth fine porous structures. We introduce the Physically-anchored Implicit Network (PIN) with Real-Pair Dual-Anchor learning, which decouples physical-pair supervision into two complementary roles. The HR anchor teaches fine structure from registered HR through pair-conditioned spectral replay, while the LR anchor constrains the prediction from acquired LR via affine-calibrated low-pass projection. Together with exact synthetic supervision in one shared volumetric implicit model, PIN learns from both physical scans without forcing the acquired-LR reconstruction to match registered HR voxel by voxel. Inference requires only acquired LR. Across 30 external baseline configurations on real CT, PIN achieves the lowest full-volume LPIPS, reduces average symmetric surface distance by 42.3-49.5% and bone-volume-fraction error by 52.2-53.4% relative to the three strongest baselines for each metric, and lowers the erased reference-pore volume fraction from 37.3-72.9% to 10.4%. Code, pretrained models, and evaluation scripts will be publicly available.

cs.CV · eess.IV