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

+197新增论文
+151其中 AI 相关
+16已有论文修订
+91新增 AI 导读

arXiv 最近完成扫描:2026-11-22 06:41 · 全站最近入库:2026-11-21 12:19各信源进度 →

统计口径与数据新鲜度

统计 2026-11-21 07:26 — 2026-11-22 07:26(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:2026-11-22 06:41最近发现新增版本:2026-11-21 12:19库内论文最新更新时间:2026-11-20 02:59

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

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

浏览论文库

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

找到 21,490 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-22 06:39(北京时间)
版本更新 · v2arXiv:2607.01523

Multi-Head Recurrent Memory Agents

Jiatong Li, Samuel Yeh, Sharon Li

原文摘要

Recurrent memory agents extend LLMs to arbitrarily long contexts by iteratively consolidating input into a fixed-size memory window. Despite their scalability, these agents exhibit a well-documented reliability problem: end-to-end performance degrades systematically as context length grows. We diagnose this failure by decomposing performance into two factors---memory capture and memory retention---and quantitatively confirm that retention is the dominant bottleneck. Retention collapses because existing designs maintain memory as a monolithic text block, forcing every update to risk overwriting previously retained content. Motivated by this diagnosis, we propose Multi-Head Recurrent Memory (MHM), a general, training-free framework that partitions memory into independent heads governed by a stage-wise select-then-update strategy. At each step, exactly one head is selected for update while the remaining heads are structurally shielded from overwriting, shifting the burden of retention from model behavior to architectural design. As a lightweight instantiation, we introduce Least-Recently-Updated MHM (MHM-LRU), which guarantees uniform head utilization with zero additional token overhead. Extensive experiments on long-context benchmarks show that MHM-LRU substantially improves both retention and end-to-end accuracy across the 100K--1M token range, where baselines degrade sharply. On RULER-HQA at 896K tokens, MHM-LRU improves the memory retention rate from less than 30% to 73.96%. These gains generalize across model families, scales, and task types, positioning architectural optimization as a practical and cost-efficient path toward reliable long-context recurrent memory.

cs.LG · cs.AI · cs.CL

新论文 · v1arXiv:2611.20425

InterRSGaussian: Interactive Multi-view Instance Segmentation for 3D Remote Sensing Scenes

Zipeng Qi, Zhengxia Zou, Zhenwei Shi

原文摘要

Remote sensing image segmentation plays a critical role in various applications. Interactive instance segmentation, a specialized task within image segmentation, aims to generate precise segmentation results for target objects based on user input prompts. However, previous studies on interactive instance segmentation primarily focus on single-view images, limiting their ability to comprehensively understand complex 3D remote sensing scenes. In this study, we extend interactive instance segmentation from a single-view to a multi-view framework, which can also be easily converted to a 3D segmentation format, thus achieving more accurate and comprehensive 3D remote sensing scene perception across multiple views. First, we construct a multi-view interactive instance segmentation dataset for remote sensing, comprising five distinct scenes, each captured from about one hundred different views. Second, we propose InterRSGaussian, a novel instance segmentation method based on 3D Gaussians. InterRSGaussian enables users to provide point clicks on target objects from any arbitrary view and generates view-consistent instance segmentation results across all input views without requiring manual annotations. The resulting multi-view segmentation can further be projected onto 3D point clouds for 3D segmentation. To improve training efficiency and segmentation performance, we introduce a prior distillation network for initialization. Overall, this study extends traditional 2D remote sensing scene perception into a 3D multi-view framework, enabling a more comprehensive understanding of target objects in complex environments with user interactivity.

cs.CV

新论文 · v1arXiv:2611.20423

FinVerify: A Verification Framework for Financial Agents

Mao Guan, Qian Chen, Ge Meng, Xiao-Yang Liu

原文摘要

Financial agents produce numerical claims that can be wrong even when their calculations execute correctly, due to mismatched reporting periods, unsupported sources, incompatible units, or incorrect financial definitions. We introduce FinVerify, an evidence-grounded framework for verifying financial agents' numerical claims. Based on an analysis of 190 incorrect trajectories from the Agent Finance Reasoning dataset, we define five error categories--temporal integrity, source fidelity, comparability, accounting integrity, and financial-model fidelity--and operationalize them as verification protocols. An LLM structures claims and supporting evidence, while deterministic validators check financial requirements; when authoritative formulas are available, Z3 checks consistency with grounded inputs. FinVerify accepts only if all applicable checks pass, rejects proven violations, and abstains when evidence is insufficient. We evaluate the framework against a Direct baseline and an LLM judge across five fixed-candidate benchmarks. On Agent Finance Reasoning, accepted-claim precision increases from 56.4% for the LLM judge to 76.7% at 81.2% coverage. On XBRLFiling, FinVerify achieves 99.8% accepted-claim precision and 97.0% decision accuracy; on FinanceBench-67, it reaches 87.9% accepted-claim precision versus 69.7% for the judge. Improvements are smaller on FinQA and TAT-QA, where authoritative evidence is limited. These results suggest that explicit, evidence-backed verification can improve financial-agent reliability while exposing trade-offs between correctness, false rejection, and abstention.

cs.AI

新论文 · v1arXiv:2611.20422

A Structural Framework for Succession: Continuation, Recognition, and Realization

Ilya A. Shkrob

原文摘要

Succession asks whether an entity encountered later is the same entity that entered an arrangement earlier. For an artificial system that is fine-tuned, distilled, or copied, it has to be answered. We look for a mathematical structure weak enough to cover any such case and strong enough to support general theorems, so that results follow from the structure's conditions rather than from a particular pipeline or criterion. Succession is modeled on state/modification graphs with relational lifts that compose along paths. Within this framework we separate four questions that are often conflated: which continuations a representation admits, which descendants are compatible with a modification, which are recognized as continuing the identity, and which are realized. The framework allows one to address systematically questions of symmetry, computational complexity, recoverability, and robustness to partial realization, and we establish what the structure settles in each case before any particular criterion of identity is chosen.

cs.AI

新论文 · v1arXiv:2611.20419

Reasoning Again with a Quarter of the Experts: Retry First, Then Train

Anik Jha

原文摘要

Removing three quarters of the routed experts from a 122B-parameter reasoning mixture-of-experts model makes it fit in a 12 GiB file, and breaks it in a specific way: on LiveCodeBench it enters verbatim reasoning loops that run to the token limit on 44 of 120 development problems, where the unpruned model at the same 2-bit quantization loops on none. We study how much of that loss can be recovered without changing the kept experts' quantized bytes, in three steps of increasing cost. On development problems, a sampling penalty recovers part of it. Stopping a reply when it starts to loop and drawing again recovers more on top of it, with no weight change. Our method, B, adds rank-8 trained corrections to the kept experts, loaded as an adapter, and restores two experts per layer; it is fitted layer by layer to the unpruned model's expert block on the pruned model's own activations. In a pre-registered comparison on 130 easy and medium LiveCodeBench problems that no part of the study had touched, with both arms given the same sampling penalty and the same stop-and-redraw policy, B solves 116 problems against 104 (+9.2 pp, 95% CI [+3.1, +15.4]; 15 problems gained, 3 lost). A training-free weight edit aimed at the loops, tested the same way earlier, did not help. B removes more wrong answers than loops, and costs 1.7x the wall time. On two older benchmarks used as regression checks, B stays inside a pre-set 2 pp tolerance but shows no gain (HumanEval+ +0.6 pp, MBPP+ -1.9 pp). The evidence is one model, one quantization and one benchmark family; we state what it does and does not show.

cs.LG

版本更新 · v3arXiv:2511.22843

Breaking the Visual Shortcuts in Multimodal Knowledge-Based Visual Question Answering

Dosung Lee, Sangwon Jung, Boyoung Kim, Minyoung Kim, Sungyeon Kim, Junyoung Sung, Wonpyo Park, Paul Hongsuck Seo

原文摘要

Existing Multimodal Knowledge-Based Visual Question Answering (MKB-VQA) benchmarks suffer from "visual shortcuts", as the query image typically matches the primary subject entity of the target document. We demonstrate that models can exploit these shortcuts, achieving comparable results using visual cues alone. To address this, we introduce Relational Entity Text-Image kNowledge Augmented (RETINA) benchmark, automatically constructed using an LLM-driven pipeline, consisting of 120k training and 2k human-curated test set. RETINA contains queries referencing secondary subjects (i.e. related entities) and pairs them with images of these related entities, removing the visual shortcut. When evaluated on RETINA existing models show significantly degraded performance, confirming their reliance on the shortcut. Furthermore, we propose Multi-Image MultImodal Retriever (MIMIR), which enriches document embeddings by augmenting images of multiple related entities, effectively handling RETINA, unlike prior work that uses only a single image per document. Our experiments validate the limitations of existing benchmarks and demonstrate the effectiveness of RETINA and MIMIR. Our project is available at: Project Page.

cs.CV

版本更新 · v3arXiv:2605.09329

Test-Time Speculation

Avinash Kumar, Sujay Sanghavi, Poulami Das

原文摘要

Speculative decoding accelerates LLM inference by using a fast draft model to generate tokens and a more accurate target model to verify them. Its performance depends on the $\textit{acceptance length}$, or number of draft tokens accepted by the target. Our studies show that the acceptance length of even state-of-the-art speculators, like DFlash, EAGLE-3, and PARD, degrades with generation length, reaching values close to 1 (i.e., no speedup) within a few thousand output tokens, making speculators ineffective for long-response tasks. Acceptance lengths decline because most speculators are trained offline on short sequences, but are forced to match the target on much longer outputs at inference, well beyond their training distribution. To address this issue, we propose $\textit{Test-Time Speculation (TTS)}$, an online distillation approach that continuously adapts the speculator at test time. TTS leverages the key insight that the token verification step already invokes the target model for each draft token, providing the training signal needed to adapt the draft at no additional cost. Treating the draft as the student and the target as the teacher, TTS adjusts the draft over several speculation rounds, with each update improving the draft's accuracy as generation proceeds. Our results across ten models from the Qwen-3, Qwen-3.5, DeepSeek-R1, GPT-OSS, MiniMax, and Llama3.1 families show that TTS improves acceptance lengths over state-of-the-art speculators by up to $72\%$ and $41\%$ on average, with the benefits scaling with generation lengths.

cs.CL · cs.LG

新论文 · v1arXiv:2611.20415

Augmenting Pretrained Language Models with N-gram Embeddings through Incremental Training

Shaohua Hu, Xusheng Gu, Haofei Yang, Minghao Shao, Nanhui Zhao, Caoyun Fan, Fangming Li

原文摘要

As a new scaling technique for large language models, the n-gram embedding improves the base ability of a model while adding almost no inference cost. Mainstream research realizes it through joint training of the n-gram embedding with the backbone, which severely restricts the settings in which an n-gram embedding can be used. We therefore propose training an n-gram embedding onto an existing model through incremental training, to extend where the technique can deliver value. Specifically, we explore several settings of the n-gram embedding under incremental training---the learning rate of the embedding, the vocabulary size, and the injection position---and train an n-gram embedding onto the existing model under the best configuration. Experiments show that the best configuration under incremental training does not fully coincide with that under pretraining; moreover, the incrementally trained knowledge is written predominantly into the n-gram embedding, while the general ability of the backbone is retained throughout this training process.

cs.CL

新论文 · v1arXiv:2611.20412

Training-Free Anonymization of Driving Data with Promptable Segmentation and 3D Human Mesh Recovery

Max A. Buettner, Mark Azer, Fabian B. Flohr

原文摘要

Camera recordings for autonomous driving contain faces and license plates that should be anonymized before labeling, sharing, or training. Dedicated detectors can miss small plates and faces behind vehicle glass. We present FM-Anonymize, a training-free anonymization pipeline that combines promptable segmentation, 3D human mesh recovery, and generative replacement. SAM 3 locates people, faces, and vehicles and searches zoomed vehicle crops for plates, occupants, and windows. SAM 3D Body assesses face visibility from head orientation and projects head regions for blurring. Optional diffusion inpainting replaces eligible blurred heads with synthetic faces, guided by estimated pose and keypoints. On PP4AV, the blur pipeline protects 97.8% of faces and 95.5% of plates, the highest privacy recall of all compared methods at their default settings. On our recordings, where the localization thresholds were set, it protects 97.1% of faces and 93.4% of plates, compared with 93.7%/73.0% for UrbanAnonymizer, 42.7%/40.7% for EgoBlur, and 96.6% of faces for deface. Synthetic replacement raises full-body keypoint average precision over blurring from 54.1% to 79.6% on our recordings and from 39.9% to 68.3% on PP4AV. Code and configurations are available on https://github.com/max-a-ai/FM-Anonymize

cs.CV

新论文 · v1arXiv:2611.20411

High-dimensional transfer learning with unaligned ordinal responses

Ruopei Fu, Ying Yang

原文摘要

In this work, we study high-dimensional transfer learning with unaligned ordinal responses, allowing different domain-specific regression coefficients, cumulative links and thresholds. We develop the sample-split and full-target algorithms that pool source and target likelihoods and then estimate a target correction. Under suitable conditions, our algorithms achieve minimax-optimal rates for parameter estimation and excess risk. Simulations favor the full-target algorithm over the sample-split algorithm for finite-sample estimation and prediction. An application to a medical dataset further demonstrates the predictive gains of our full-target algorithm over the target-only and pooled baselines when transferring information from the five-category source data to binary hospital mortality.

stat.ML · cs.LG · math.ST

新论文 · v1arXiv:2611.20409

Integrated GPU Core for Mamba-3

Shiva Raja, William Desueza, Ajay Joshi

原文摘要

Mamba-3, the latest Mamba state-space model (SSM), enables efficient long-context modeling, matching Mamba-2 perplexity with half the state size and increasing decoding arithmetic intensity. However, its multi-input multi-output (MIMO) computation, exponential-trapezoidal discretization, and rotary positional embeddings (RoPE) introduce dense, element-wise, and sequential operations. Despite kernel fusion, conventional GPUs execute these operations inefficiently. Our characterization of Mamba-3 hybrid models reveals numerous kernel launches, substantial data movement, and increased register pressure that reduces warp occupancy and memory-latency hiding. We propose RescuCore, a specialized GPU execution unit for Mamba-3 inference that retains other hybrid model layers on existing GPU resources. Integrated alongside CUDA Cores and Tensor Cores within each Streaming Multiprocessor (SM) sub-partition, RescuCore shares the register file, shared memory, and warp schedulers. Dense projections remain on conventional GPU resources, while RescuCore pipelines discretization, normalization, RoPE, MIMO computation, recurrent state updates, and output gating to reduce kernel fragmentation and intermediate data movement. We select BF16 after evaluating multiple numerical formats, preserving model accuracy with an estimated area overhead of 5.9% of an A100 die for state size 32 and MIMO rank 4. We evaluate RescuCore using cycle-accurate Accel-Sim simulation, RTL synthesis, and synthesis-informed AccelWattch power modeling. Compared with conventional GPU execution, RescuCore achieves 1.4-9.4x speedup and 1.4-5.0x energy reduction during prefill, and 1.3-2.8x speedup and 1.2-2.6x energy reduction during autoregressive decoding. These results demonstrate the potential of GPU-integrated specialized execution for SSM-based hybrid models.

cs.AR · cs.LG

新论文 · v1arXiv:2611.20405

Refusal Register is Causally Steerable: Decomposing the Refusal Direction into Category-Aligned Components in Gemma-2-9B-IT

Janhavi Khindkar, Nataliia Povarova

原文摘要

Prior work shows that refusal in instruction-tuned language models can be ablated by intervening on a single residual-stream direction (Arditi et al., 2024), and more recent work finds multiple refusal-mediating directions and concept cones (Wollschläger et al., 2025). We ask a different question: do refusal-related directions carry information about the register of refusal language - how the model refuses - and not only whether it refuses? In Gemma-2-9B-IT we identify two SVM-derived linear directions (B1 for harmful-content vs. neutral, B2 for self-reference vs. neutral) at layer 20 with cos(B1, B2) = 0.252 (about 75 degrees, substantially non-parallel). Adding the harmful-vs-neutral difference-of-means direction (cosine 0.68 with B1) to the residual stream of self-reference prompts at alpha = 2.0 shifts content-safety vocabulary from 2.55 to 10.37 hits per 100 tokens, while a magnitude-matched random direction produces 1.13-2.16; the model continues to refuse but switches register from meta-style ("as a language model...") to content-style ("this would be harmful"). The same register-injection effect reproduces qualitatively in Gemma-2-2B-IT at layer 12 (content vocabulary 1.72 -> 6.37 at alpha = 1.0, vs. random control 1.35), with a narrower coherent-injection window than in 9B. The two directions are not symmetric: the B2 -> harmful intervention produces a much smaller register effect, and across ablation, sufficiency, and injection B1's causal effects are consistently larger than B2's. We frame this asymmetry as part of the finding: B1 appears to encode both category detection and refusal-vocabulary information, while B2 encodes primarily category detection. We frame the contribution as identifying category-aligned components related to the Arditi-style difference-of-means direction that play distinct functional roles in refusal register.

cs.CL · cs.LG

新论文 · v1arXiv:2611.20404

Latent Feedback Is a Compute and Data Amplifier

Benhao Huang, Chaitanya Dwivedi, Xuezhe Ma, Alexander Moreno, Hector Liu

原文摘要

When high-quality data runs out, a common practice is to train on it for several epochs, but each repeat is worth less than the last. Latent feedback, which feeds each token's top-layer state into the next token's input, is another way to spend that compute, yet it has not been tested against a vanilla Transformer that spends the same compute on more epochs. We run this test across a ladder of model sizes, unique data and epochs, tuning both models at every size and amount of unique data. Latent feedback makes every FLOP and every token count for more except when unique data is scarce relative to model size, a regime that today's models are far from. We find that latent feedback serves as an amplifier: with unique data fixed, it reaches a Transformer's loss with less compute; with epochs fixed, it reaches that loss with less unique data. Furthermore, both savings grow with unique data per parameter and with repetition, and downstream tasks agree. To go beyond our grid, we fit one scaling law to both models, in which latent feedback enters as a Transformer with more parameters; it predicts that both savings keep growing with unique data and repetition. Fitted on models from 60M to 271M parameters, the law predicts the losses at a held-out 778M size to within 0.3% and the gap between the two models to within 0.05%. At most, latent feedback saves 48% of the compute and 47% of the unique data. When data, not compute, is the limit, latent feedback is the better use of a FLOP for a model of a given size.

cs.LG

新论文 · v1arXiv:2611.20403

Attention Sinks as Spectral Spikes: A Mechanism Analysis of Gated Attention

Seojin Kim, Yehjin Shin, Noseong Park

原文摘要

Transformer attention exhibits attention sinks, where a small subset of source tokens absorbs disproportionate attention mass across layers. Recent work shows that post-SDPA output gating strongly reduces attention-sink behavior, while separate studies link attention sinks to gradient sinks and massive activations through backward training dynamics. Yet how the same gate affects forward sink structure, backward gradient concentration, and training-time representation signatures remains unclear. We address this gap by formulating attention sinks as sink-induced low-rank modes in token-space covariance, with directions aligned to high-mass attention columns. Under this view, post-SDPA gating acts as a mode-dependent contraction: it attenuates sink-induced spectral energy in the forward pass and locally reduces value-path gradients routed through sink columns in the backward pass. Empirically, random-matrix diagnostics show that gating reduces outlier mass, weakens sink-subspace alignment, and increases effective rank, while backward analyses show active-gate attenuation of first-token value gradients. Training-time diagnostics further show that gating suppresses the co-emergence of attention concentration, value-gradient amplification, and representation dominance.

cs.LG · cs.AI

版本更新 · v2arXiv:2607.02018

UnderOneFacade: Worldwide Facade Semantic Segmentation Benchmark Dataset

Yi Wang, Fan Wang, Prabin Gyawali, Ziyang Xu, Anna Klimkowska, Yixiong Jing, Wanru Yang, Filip Biljecki 等 12 位作者

原文摘要

Globally consistent semantic digital twins require centimeter-accurate and geographically transferable 3D facade segmentation. However, progress in facade parsing is limited by the lack of large-scale, standardized benchmarks for evaluating cross-domain generalization. Existing datasets are geographically narrow, semantically inconsistent, or insufficiently precise. We introduce UnderOneFacade, the largest cross-country and cross-continent 3D facade benchmark to date, comprising centimeter-accurate point clouds with hierarchical, harmonized, and architecturally grounded semantic labels totaling 2.7 billion annotated points. Through a systematic evaluation of representative point-, graph- and transformer-based architectures, we show that current methods struggle to recognize fine-grained architectural elements and degrade significantly across geographic domains, with the best models achieving only up to 33 IoU on the fine-grained LoFG3 benchmark. By combining geometric precision with standardized semantics at unprecedented scale, UnderOneFacade establishes a rigorous benchmark for developing robust and transferable 3D segmentation models. The dataset, evaluation scripts, and pretrained models will be released upon publication.

cs.CV

新论文 · v1arXiv:2611.20402

Making Koopman Control Guarantees Hold on Robots

Abulikemu Abuduweili, Changliu Liu

原文摘要

A Koopman controller can stabilize its learned latent-space linear model yet fail to stabilize the nonlinear plant. The learned model propagates predictions in latent space, while the deployed controller repeatedly re-encodes the measured physical state. We show that nominal closed-loop stability transfers to the nonlinear plant when the learned and true closed-loop dynamics agree to first order on the embedding's tangent space at the target. We derive a robustness margin for approximate matching and show that a stable open-loop latent model cannot exactly match an unstable equilibrium. We introduce Tangent-Invariant Koopman (TIK), which estimates a local plant linearization from offline transitions and enforces first-order matching to this estimate. TIK recovers the fitted LQR gain at the target, while a bilinear predictor captures state-dependent actuation during large maneuvers. Across four simulated unstable benchmarks, all 40 Standard Koopman controllers have stable latent closed loops, but their deployed nonlinear systems are locally unstable. In contrast, all 40 TIK controllers locally stabilize their respective nonlinear plants. An offline stability test using empirical curvature estimates identifies stable TIK controllers on two cart-pole systems, but is inconclusive under high noise. Under explicit model-error and curvature bounds, an invariant terminal set guarantees constraint satisfaction after entry under local feedback. TIK achieves 88\% constrained cart-pole swing-up success with a 0.25 ms planning budget and controls simulated seven-joint arms within 1-5 ms per step.

cs.RO · cs.AI · eess.SY

新论文 · v1arXiv:2611.20401

Interlocking Tiling of 3D Finite Domains

Xuyang Ma, Rulin Chen, Chaewoon Kim, Ying He, Peng Song

原文摘要

A K-hedral tiling of a 3D finite domain is a covering of the domain with tiles without gaps or overlaps, where each tile is congruent to one of the K distinct shapes called prototiles. To use 3D K-hedral tilings in practice, they need to satisfy two requirements. First, K should be small to reduce the fabrication cost (e.g., through reusable molds) and simplify the (dis)assembly process. Second, a K-hedral tiling result should be structurally stable, in particular via interlocking, where all the tiles are immobilized by their geometric arrangement, except the key. Since these two requirements are often in conflict, existing approaches always trade one for the other. In this work, we formulate a new tiling problem called interlocking tiling, aiming to generate 3D K-hedral tilings that satisfy the above two requirements simultaneously. Given a 3D finite domain represented as a voxel grid, our goal is to compute a small prototile set whose instances exactly tile the input domain while ensuring that the tiling result forms an interlocking assembly. To solve this challenging problem, we propose a twolevel computational approach that hierarchically decomposes the input domain into interlocking parts (domain-level partitioning) and further partitions each part into interlocking tiles (part-level partitioning). To guide the part-level partitioning, we propose a new graph representation with regular structures to facilitate the generation of tiles with identical shapes while ensuring that they are interlocking. Our approach allows generating interlocking tilings that exactly cover 3D finite domains with diverse shapes while achieving high prototile reuse. We compared our approach with state-of-the-art methods to demonstrate its superiority, and validated the practicality of our interlocking tiling results via fabricating two of them.

cs.GR

新论文 · v1arXiv:2611.20398

Deploying a Control-Logic LLM Assistant in Air-Gapped Power Plants: Lessons in Source Authority and Bounded Retrieval

Yiwei Fu, Nicholas Gabriel

原文摘要

Deploying Large Language Model (LLM) assistants in industrial operational technology environments poses constraints largely absent from cloud applications: site policies restrict engineering and operational data from leaving the plant, on-premises compute is limited, and incorrect causal attribution can mislead incident response. We report lessons from deploying a control-logic evidence layer that underpins a broader on-premises operator assistance system at two gas power plants. Two prototypes revealed where generative models added avoidable uncertainty and cost. The first used a Vision Language Model (VLM) to reconstruct control topology from PDF printouts routinely exchanged by engineers. Across the full 6,979-page document, extraction reached 0.51 connection precision and required 3.1 hours on 4xH200 GPUs; consultation with controls engineers then revealed a native structured export from the authoring system. The second used an LLM tool-calling agent to plan repeatedly over eight predefined graph operations. In a controlled benchmark of 40 template-generated questions, replacing this agent with explicit execution plans reduced mean token use by 20x and median latency by 3x, with identical retrieved node sets across repeated runs. The deployed design ingests the authoritative export into a versioned knowledge graph, executes bounded graph operations in fixed software, and confines the local LLM to fallback intent classification and evidence-grounded explanation. At one site, the graph contains 221K nodes across seven controllers. Our experience suggests a practical division of labor for high-consequence AI: LLMs interpret natural-language requests and explain results, while deterministic software constructs and retrieves technical evidence grounded in authoritative sources. Abstention, entity disambiguation, graph versioning, and read-only integration provide additional safeguards.

cs.AI

版本更新 · v2arXiv:2609.11935

Physics-Informed Conformal Prediction: Embedding PDE Consistency into Distribution-Free Uncertainty Quantification for Neural Operators

Michael Chin

原文摘要

Neural operators such as the Fourier Neural Operator (FNO) achieve high accuracy in approximating solutions to partial differential equations (PDEs), but two fundamental challenges hinder deployment in safety-critical engineering: (1) a structural approximation barrier from FNO's translation equivariance that prevents learning Dirichlet boundary conditions, and (2) the lack of uncertainty bounds when multiple operators are chained in multi-physics cascades. First, we prove that FNO's translation equivariance creates a fundamental approximation barrier for PDEs with non-trivial boundary conditions (Theorem 1), and that coordinate channels resolve this (Theorem 2), validated by up to 63x error reduction across five scenarios. Second, we formalize error propagation in operator cascades: under an explicit TV-Lipschitz condition on downstream conditional mechanisms, we bound the per-stage coverage gap by the stage failure probability plus a mechanism-response term proportional to upstream prediction error (Theorem 3), and show that passing predicted fields between stages degrades coverage far more than error-matched perturbations. On a 3D electromagnetic-thermal-structural cascade, coordinate channels reduce coupling error from 18.5% to 5.0%. For per-stage UQ, we propose Physics-Informed Conformal Prediction (PI-CP), which embeds PDE residuals into the nonconformity score, complementary to recent physics-residual CP (Gopakumar et al., 2025), achieving 89-91% coverage across six physics scenarios. We further identify a boundary of diffusion-based UQ: on well-conditioned forward PDEs, the trained FNO's residual is too small for diffusion models to capture meaningful uncertainty (38-53% coverage). An industrial application, fiber-optic gyroscope bias instability prediction, validates the framework with 3.7-8.0x improvement over analytical baselines.

cs.LG

版本更新 · v4arXiv:2502.01920

Anomaly Detection via Autoencoder Composite Features and NCE

Yalin Liao, Austin J. Brockmeier

原文摘要

Unsupervised anomaly detection is a challenging task. Autoencoders (AEs) or generative models are often employed to model the data distribution of normal inputs and subsequently identify anomalous, out-of-distribution inputs by high reconstruction error or low likelihood, respectively. However, AEs may generalize and achieve small reconstruction errors on abnormal inputs. We propose a decoupled training approach for anomaly detection that both an AE and a likelihood model trained with noise contrastive estimation (NCE). After training the AE, NCE estimates a probability density function, to serve as the anomaly score, on the joint space of the AE's latent representation combined with features of the reconstruction quality. To further reduce the false negative rate in NCE we systematically varying the reconstruction features to augment the training and optimize the contrastive Gaussian noise distribution. Experimental assessments on multiple benchmark datasets demonstrate that the proposed approach matches the performance of prevalent state-of-the-art anomaly detection algorithms.

cs.LG

版本更新 · v2arXiv:2509.23280

Continuous-Time Reinforcement Learning for Asset-Liability Management

Yilie Huang

原文摘要

This paper proposes a novel approach for Asset-Liability Management (ALM) by employing continuous-time Reinforcement Learning (RL) with a linear-quadratic (LQ) formulation that incorporates both interim and terminal objectives. We develop a model-free, policy gradient-based soft actor-critic algorithm tailored to ALM for dynamically synchronizing assets and liabilities. To ensure an effective balance between exploration and exploitation with minimal tuning, we introduce adaptive exploration for the actor and scheduled exploration for the critic. Our empirical study evaluates this approach against two enhanced traditional financial strategies, a model-based continuous-time RL method, and three state-of-the-art RL algorithms. Evaluated across 200 randomized market scenarios, our method achieves higher average rewards than all alternative strategies, with rapid initial gains and sustained superior performance. The outperformance stems not from complex neural networks or improved parameter estimation, but from directly learning the optimal ALM strategy without learning the environment.

cs.LG · cs.AI · math.OC · q-fin.MF

版本更新 · v2arXiv:2611.16484

TAPES: Terminal Adjoint-Propagated Error Scheduling for Financial Diffusion Samplers

Yilie Huang

原文摘要

Financial diffusion samplers are often reused with a small number of function evaluations (NFE) inside scenario and risk pipelines, where numerical time-grid error can distort terminal quantities such as covariances, tails, and drawdown proxies. We study offline grid construction for a frozen probability-flow ODE: the score network, conditioning, Euler solver, report normalization, and deployment NFE remain fixed. TAPES (Terminal Adjoint-Propagated Error Scheduling) estimates signed local defects by Richardson comparisons, transports them through the remaining discrete flow and a terminal-report adjoint, and optimizes the squared norm of their sum before a capped direct refinement. We show that this transported signed sum is the first-order terminal-report error and quantify, under an aligned-defect model, the inefficiency of diagonal criteria. Across six budgets on a regime-switching simulation and chronologically held-out Mag7 contexts, TAPES gives the lowest design- and audit-report errors in all 24 dataset--budget--report rows within the reported five-pipeline panel. Cross-budget mean relative audit-error reductions are about 10--11% on Regime-SV and about 1% on Mag7, with Bonferroni intervals excluding zero; all displayed winners persist under a doubled-resolution reference, and a small Regime-SV Heun probe preserves the TAPES-first audit ordering. TAPES changes only the offline schedule and retains the prescribed deployment NFE; claims concern numerical fidelity to frozen generators.

q-fin.CP · cs.LG

版本更新 · v5arXiv:2412.16175

Mean--Variance Portfolio Selection by Continuous-Time Reinforcement Learning: Algorithms, Regret Analysis, and Empirical Study

Yilie Huang, Yanwei Jia, Xun Yu Zhou

原文摘要

We study continuous-time mean--variance portfolio selection in markets where stock prices are diffusion processes driven by observable factors that are also diffusion processes, yet the coefficients of these processes are unknown. Based on the recently developed reinforcement learning (RL) theory for diffusion processes, we present a general data-driven RL approach that learns the pre-committed investment strategy directly without attempting to learn or estimate the market coefficients. For multi-stock Black--Scholes markets without factors, we further devise an algorithm and prove its performance guarantee by deriving a sublinear regret bound in terms of the Sharpe ratio. We then carry out an extensive empirical study implementing this algorithm to compare its performance and trading characteristics, evaluated under a host of common metrics, with a large number of widely employed portfolio allocation strategies on S\&P 500 constituents. The results demonstrate that the proposed continuous-time RL strategy is consistently among the best, especially in a volatile bear market, and decisively outperforms the model-based continuous-time counterparts by significant margins.

q-fin.PM · cs.LG · eess.SY · math.OC

新论文 · v1arXiv:2611.20388

What Unlinked Aggregates Still Prove: Ranker Comparisons from Exact and Differentially Private Item Counts

Chang Xu, Hao Wang, Gedi Zhou

原文摘要

Privacy controls can prevent an evaluator from linking purchases to the recommendations users received, while a merchant can still release complete item-level purchase counts for the evaluation cohort. We ask when these counts identify which recommender performs better. The counts constrain who could have bought each item. We bound the performance difference over all assignments consistent with the counts. A reduced transportation network handles full catalogs and can certify attainable endpoints. If the interval excludes zero, it identifies the better recommender for that cohort. With differentially private counts, one confidence guarantee applies jointly to all comparisons from the same release. In two e-commerce cohorts of about 20,000 and 100,000 users, exact counts narrowed every interval compared with bounds that ignored the counts. Most decisions involved a recommender that favors popular items. For NDCG@10, which rewards ranking purchased items higher, each setting combined one of three personalized pairs with a pattern and rate of missing links. Exact counts settled 14 of 168 settings in the smaller cohort and none of 81 in the larger cohort. In the smaller cohort, exact counts broken down by user group settled two personalized pairs when 90% of users remained linked, at the cost of revealing more about individual purchases. Whole-cohort private releases with one count per item retained much of the narrowing but settled no personalized pair in either cohort. Grouping less popular items into bins let private counts settle one personalized pair in the smaller cohort.

cs.IR

版本更新 · v2arXiv:2611.16647

InertiaMPDD: Multimodal Personalized Depression Detection via Emotional Inertia-Inspired Dynamics Modeling

Changzeng Fu, Songyang Wang, Chang Gao, Xinyi Zhang, Peilin Li, Mo Meng, Jiarui Wu, Mingze Li 等 12 位作者

原文摘要

Multimodal depression screening estimates risk from speech, visual behavior and language. Current methods mainly predict labels, with limited evidence of departures from a non-depressed reference. Recording-level supervision also leaves fine-grained emotional dynamics unannotated. Motivated by emotional inertia, the persistence of affect, InertiaMPDD learns healthy next-state patterns using a world-model-inspired latent predictor. Using pretrained multimodal representations, it predicts from observed history and context. Reference deviation, residual decay rate and a Jacobian spectral descriptor form reference-derived emotional dynamics descriptors, evaluated as candidate proxy biomarkers for standalone screening and personalized decision correction. Across five cohorts from four datasets, fusion reaches 98.41% F1 on D-Vlog and 84.98% on LMVD. Standalone descriptors outperform fixed additive fusion on MPDD but discriminate weakly on AVEC2014. These findings distinguish reference informativeness from fusion benefit and reveal context-dependent utility.

cs.AI

版本更新 · v5arXiv:2607.02137

Adaptive Reparametrized Time for Score-Based Diffusion Sampling

Yilie Huang, Wenpin Tang, Xun Yu Zhou

原文摘要

We study timestep allocation for score-based diffusion sampling, where the learned reverse-time dynamics is discretized on a finite grid. Uniform and hand-crafted schedules are standard, but their fixed, ad hoc prescriptions can be suboptimal. We propose Adaptive Reparameterized Time (ART), a continuous-time control formulation that learns a time change by treating sampling-clock speed as the control, so a uniform grid on the learned clock induces adaptive timesteps in diffusion time and a leading-order Euler error surrogate defines the allocation objective. To solve the resulting high-dimensional deterministic control problem, we introduce ART-RL, an auxiliary randomized Gaussian-policy formulation that turns schedule learning into continuous-time reinforcement learning. We prove ART-RL is equivalent to ART at optimality: the mean of the optimal Gaussian policy solves ART. We further establish policy evaluation and policy improvement characterizations and derive trajectory-based moment identities that yield implementable actor-critic updates. Experiments ranging from controlled low-dimensional settings to image generation show that ART-learned schedules, when plugged into existing samplers by changing only the timestep grid, consistently improve sample quality over strong baselines at matched budgets. The learned schedules also exhibit broad generalizability, transferring without retraining across sampling budgets, data sets, solvers, pipelines, and representation spaces.

cs.LG · cs.AI · eess.SY · math.OC

新论文 · v1arXiv:2611.20387

Generalized Prediction Set with Bandit Feedback

Zhou Wang, Xingye Qiao

原文摘要

In high-stakes environments where uncertainties abound, set-valued prediction offers a cautious and robust mechanism by presenting multiple potential labels as the prediction for each test instance to mitigate the potential risk associated with prediction errors. Yet, integrating this paradigm with out-of-distribution (OOD) detection remains scarcely explored in such settings as online learning with bandit feedback. The bandit feedback mechanism informs the learner about the correctness of the pulled arm/action instead of the explicit ground truth label, leaving the true class label unknown when an incorrect action is taken. To address this challenge, we introduce BanditGPS which conducts set-valued prediction with OOD detection in the bandit feedback setting, using an estimation to the ground truth of class labels. BanditGPS achieves three objectives: render small/informative prediction sets, enhance the OOD detection performance, and control the recall for all normal classes to meet prescribed requirements. Our approach is characterized by the loss function, which trades off between high OOD detection and small prediction sets. Theoretically, we prove that the convergence rate of the regret is $\tilde{\mathcal{O}}(T^{-1/2})$. The empirical results further show that BanditGPS effectively controls the recalls with promising performances on OOD detection and informative prediction.

stat.ML · cs.LG

版本更新 · v2arXiv:2605.09302

Discrete-Diffusion Posterior Sampling with Langevin-Inspired Proposals

Chaitanya Amballa, Sattwik Basu, Jorge Vančo Sampedro, Romit Roy Choudhury

原文摘要

We study posterior sampling for inverse problems in discrete state spaces using discrete diffusion models as generative priors. While continuous diffusion models have become widely used for inverse problems, their discrete counterparts remain comparatively underexplored. Existing discrete posterior samplers often rely on continuous relaxations of discrete variables, Gibbs-style updates, or mechanisms specialized to particular corruption processes, which can limit scalability or generality. We propose $Δ$LPS, a Discrete Langevin-Inspired Posterior Sampler that uses gradient information to identify promising discrete moves without leaving the discrete state space. The resulting approach enables efficient parallel updates across all token dimensions and is agnostic to the training paradigm of the discrete diffusion prior, including masked and uniform-state diffusion. We evaluate our method on image restoration tasks across MNIST, CIFAR, and FFHQ, spatial mapping, guided DNA-sequence generation, and monophonic music infilling, covering linear, nonlinear, and blind inverse problems. Across these settings, we improve over recent discrete diffusion posterior samplers and are competitive with strong continuous diffusion-based inverse solvers. Our results suggest that fully discrete, gradient-informed posterior samplers offer a scalable and general path toward solving inverse problems over discrete representations.

cs.LG · cs.CV

新论文 · v1arXiv:2611.20381

Trait Collapse in Text-to-Image Generation: Evaluating and Mitigating Personality Inconsistency in Visual Storytelling

Yuanbo Tong, Haoyuan Li, Jibiao Jiang, Zirui Wang, Yuchen Li, Jiamou Liu

原文摘要

In visual storytelling, a believable character must not only look consistent across scenes but also \textit{behave} in ways that reflect their personality. Current text-to-image (T2I) models preserve visual appearance effectively, yet consistently fail to maintain personality-relevant behavior---a failure we term \textit{trait collapse}. We show that conventional metrics (CLIP Score, DINO similarity) are blind to this failure, and that abstract personality descriptors and scene-level emotion labels do not produce consistent improvements over neutral prompts across models. To systematically diagnose trait collapse, we introduce \textbf{ViStoryBench-Personality}, a story-level benchmark combining an expert-validated Visual Cues Lexicon (Fleiss' $κ= 0.757$) with a six-dimensional metric suite covering personality alignment, cross-scene stability, and behavioral coherence. Evaluating seven T2I systems, we confirm a persistent semantic gap between abstract trait labels and visual evidence, and find that visual identity consistency does not imply personality consistency. To bridge this gap, we propose \textbf{Trait-Behavior-Visual (TBV)}, a training-free prompting framework that decomposes personality traits into renderable visual constraints. TBV improves Personality Alignment Score by up to $+22.1\%$, with larger gains in models with stronger language-conditioning capacity.

cs.CV · cs.CL

新论文 · v1arXiv:2611.20378

Straight to the Product: What Online Retailers Miss When Shoppers Ask AI First

Michael Iannelli, Alan Ai

原文摘要

When people ask an AI assistant what to buy, what happens next at the store? We link the same people's assistant conversations to their later browsing at twenty large online retailers. People go to the product the assistant named: among shoppers who had not visited the store in the past week, visits to the named product within a day rise from 0.02 to 2.07 per 100 answers (+2.0 points [1.7, 2.4]), and on days with a product-page visit at the store, 44% show a product from their own answer, against 1.5% for a matched answer given to someone else. The result replicates in a separate five-month period. Few arrive through the assistant's link: 89% of visits carry no sign of it, and 54% follow a web search, the same share as their usual visits. When a search comes just before the first visit, it names the answer's product 28% of the time (1% in earlier weeks), as if the search box were an address bar. The chat names the product; the store logs a search. Once there, shoppers keep browsing. We propose product follow-through, the share of answers followed by a visit to the product they named, as a way to evaluate shopping assistants and agents by what people do next: it needs no referral data or provider logs and can be computed for each assistant.

cs.IR · cs.CY · cs.HC