跳到正文
研踪PaperTrail

论文库

检索计算机领域论文,按研究方向、作者或关键词筛选。

部分采集异常
正在读取账号与研究记录…

全站更新概况

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

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

统计口径与数据新鲜度

统计 2026-11-15 21:17 — 2026-11-16 21:17(北京时间) · 点击数字查看对应简报;AI 相关包含在新增论文内。

最近完成检查:2026-11-16 01:34最近发现新增版本:2026-11-16 13:44库内论文最新更新时间:2026-11-14 02:59

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

23,827 已收录论文19,941 AI 相关6,921 有修订112 HF 社区发现

浏览论文库

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

找到 17,708 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-16 01:33(北京时间)
新论文 · v1arXiv:2611.15798

Bag of concepts can enhance other text representations for medical systematic review screening

Radosław Pytlak, Paweł Cichosz, Bartłomiej Fajdek, Bogdan Jastrzębski

原文摘要

The paper examines the choice of text representation for semi-automated systematic literature review (SLR) systems based on active learning. Motivated by the insufficient attention that the bag of concepts representation has received so far, the paper discusses the implementation and experimental evaluation of several variations thereof. The experimental study was performed on medical SLR case studies, with results reported for random forest classification, cluster initialization and uncertainty sampling. The utility of bag of concepts features obtained using different ontologies is evaluated by applying them within the active learning framework to sequentially create classification models based on training data iteratively expanded with expert annotations. Systematic experiments compare BoW, BoC, PubMedBERT, Jina, Gemma and Qwen3 representations, together with combinations of BoC and each of the other representations. The evaluation uses workload savings and performance profiles adapted from optimization benchmarking. The results suggest that concept-based features can provide complementary information, with the BoC-augmented representations achieving higher mean workload savings on the DERP cases than their corresponding base representations. The Qwen3+BoC representation gives the best average performance in this comparison. The effects of augmentation vary across case studies and base representations. The paper also concentrates on the impact of ontology choice on document classification outcomes. In particular, the paper compares the predictive value of bag of concepts representations based on chosen ontologies with that of representations obtained using MedCAT with a UMLS-based dictionary.

cs.IR

版本更新 · v6arXiv:2305.01420

An Improved Subquadratic Bound for Online Bisection

Marcin Bienkowski, Stefan Schmid

原文摘要

The online bisection problem is a natural dynamic variant of the classic bisection problem, where one has to dynamically maintain a partition of $n$ elements into two clusters of cardinality $n/2$. During runtime, an online algorithm is given a sequence of requests, each being a pair of elements: an inter-cluster request costs one unit while an intra-cluster one is free. The algorithm may change the partition, paying a unit cost for each element that changes its cluster. This natural problem admits a simple deterministic $O(n^2)$-competitive algorithm [Avin et al., DISC 2016]. While several significant improvements over this result have been obtained since the original work, all of them either limit the generality of the input or assume some form of resource augmentation (e.g., larger clusters). Moreover, the algorithm of Avin et al. achieves the best known competitive ratio even if randomization is allowed. In this paper, we present the first randomized online algorithm that breaks this natural quadratic barrier and achieves a competitive ratio of $\tilde{O}(n^{7/4})$ without resource augmentation and for an arbitrary sequence of requests.

cs.DS

新论文 · v1arXiv:2611.15794

On Multi-Constrained Path Finding utilizing Grover Search and Decoded Quantum Interferometry

Dmitrii Chemodanov

原文摘要

Finding a path whose totals under several additive metrics all fall inside prescribed ranges is weakly NP-complete. We ask what two quantum algorithms can do for it, against strong classical baselines. For directed acyclic graphs (DAGs) we give a Grover search whose state preparation builds the uniform superposition over all source-to-destination paths with one qubit per arc and no ancilla. We prove it correct, give its exact gate count, and show that measured strings form an error- detecting code whose distance is a girth-like graph parameter. We complete the oracle and count its gates: one search on our largest instance, 136 arc qubits and 262,144 paths, needs about 4.4 x 10^5 Toffoli gates, while classical dynamic programming decides the same instance with 18,754 labels. Grover search needs O(sqrt(N/M)) oracle calls for M good paths among N . Classical meet-in-the-middle matches this exponent in the worst case, and known quantum algorithms with quantum memory beat it. For a staged fabric with node-independent linear weights, maximizing the number of satisfied metrics is max-LINSAT, and with power- of-index weights it is Optimal Polynomial Intersection, where no known polynomial-time classical algorithm matches decoded quantum interferometry (DQI). We verify this translation in simulation and run two small binary instances on IBM Heron processors, which keep 67 % and 15 % of the ideal gain over uniform sampling.

cs.NI · quant-ph

新论文 · v1arXiv:2611.15792

Bounded Knapsack in Truly Subquadratic Time

Huanyi Zhou

原文摘要

Bounded Knapsack asks how many copies of each item to take to maximize profit under a capacity constraint. Given $n$ records with binary-encoded multiplicities, we return optimal counts in randomized time with a fixed power saving in the maximum item weight $w_{\max}$ or maximum item profit $p_{\max}$. Fix $1<ρ<2$ and suppose integer $(\min,+)$-convolution on length-$m$ arrays can be computed with failure probability at most $η$ in $\widetilde O(m^ρ)$ time, up to polynomial factors in the entry bit length and $\log(1/η)$. Put $f(ρ)=4(3-ρ)/(8-3ρ).$ For any fixed $ε>0$, on inputs whose integers have magnitude polynomial in $n+\min(w_{\max},p_{\max})$, our algorithm runs in time $$\widetilde O\bigl(n\min(w_{\max},p_{\max})^{f(ρ)+ε}\bigr).$$ It returns optimal counts with high probability. If convolution saves $δ=2-ρ$ in its exponent, knapsack retains $δ/(1+3δ/2)$. With the integer convolution exponent $1.9601397040925$, a separate parameter choice gives $\widetilde O(n+\min(w_{\max},p_{\max})^{1.961})$, including for 0-1 Knapsack. We keep the copies compressed. Witnessed convolutions generate residual queries, and local linear programs filter them by cost. Resilient lattice cores allow recursive states to coalesce. Progression reserves support rounding, and a thin affordable fiber yields a lift that reduces the normalized column volume. This volume bounds the number of LP regions a grid lookup examines.

cs.DS · cs.CG

版本更新 · v4arXiv:2605.09028

Diagnosing and Mitigating Domain Shift in Permission-Based Android Malware Detection

Md Rafid Islam

原文摘要

Machine learning-based Android malware detectors often fail in real-world deployment due to domain shift, where models trained on one data source perform poorly on applications from another. This paper presents a comprehensive study on the generalizability and interpretability of permission-based detectors under cross-domain conditions. Using two complementary datasets (PerMalDroid and NATICUSdroid) and five ensemble classifiers, we first establish an intra-domain baseline, where models achieve over 92% accuracy, and then quantify a severe asymmetric performance drop. While models trained on PerMalDroid generalize well to NATICUSdroid (86% accuracy), the reverse direction sees a drastic drop to 73% accuracy. Explainable AI analysis reveals bimodal feature distributions and shows that feature importance is highly unstable, with key permissions losing or gaining influence across domains. The predictive feature sets for different domains are fundamentally mismatched, as models rely on different, dataset-specific permissions. Most importantly, an ablation study demonstrates that for most models, training on a noisy feature set leads to poor generalization, confirming that domain-specific artifacts are a greater obstacle than missing features. To mitigate this, we validate a hybrid training strategy based on the intersection of common features and successfully recover cross-domain performance, achieving 88% accuracy on PerMalDroid and maintaining 97% on NATICUSdroid. These findings highlight the importance of explainable, cross-domain-robust malware detection systems and provide a practical pathway toward improving real-world deployment of permission-based Android malware detectors.

cs.LG

新论文 · v1arXiv:2611.15783

A Linear-Time Construction on $\mathbb{S}^2$ Achieving a Sublinear Energy Gap on Smale's Seventh Problem

Rishabh Kabra, Wim Dehaen, Andrea Karlova, Mark Hamilton, Joao Carreira, Maks Ovsjanikov, Jonathan T. Barron

原文摘要

Smale's seventh problem asks for a polynomial-time algorithm that places $N$ points on the unit sphere $\mathbb{S}^2$ with logarithmic energy (the sum of $-\log$ of all pairwise distances) within $O(\log N)$ of the minimum. The minimal energy has an asymptotic expansion with explicit terms of order $N^2$ and $N\log N$ and a linear term whose constant is unknown. Brauchart, Hardin, and Saff conjectured this constant; the resulting expansion, $E_{BHS}(N)$, is known to be an upper bound on the minimum, up to $o(N)$. All previously analyzed constructions on $\mathbb{S}^2$ have energy exceeding $E_{BHS}(N)$ by an amount that grows at least linearly in $N$, i.e. $Ω(N)$, far from the $O(\log N)$ accuracy sought by Smale. We present TerracedZonal, a deterministic construction that generates $N$ points in $Θ(N)$ time and memory. Points are placed on latitude rings whose populations are held constant across bands ("terraces") chosen by a one-dimensional dynamic program, so that adjacent rings within each terrace lock into a staggered triangular lattice. Under a locality assumption, we prove that the energy exceeds $E_{BHS}(N)$ by only $Θ(N^{2/3})$, with an explicit constant; exact evaluation up to $N = 10^9$ confirms this rate. Under the same assumption, its distance from the true minimum is sublinear if the conjecture holds. To our knowledge, this is the first algorithmic construction with a sublinear excess over $E_{BHS}(N)$ of explicit order (proven under the locality assumption), a necessary step toward Smale's seventh problem. With one parameter changed, the same algorithm also gives an equal-weight integration rule on $\mathbb{S}^2$ that runs just as fast and whose error decays at an empirical rate of $N^{-4}$ on smooth functions, versus $N^{-3/2}$ for every previous fast construction we tested.

cs.CG · math.MG

新论文 · v1arXiv:2611.15781

Prediction-Error-Tightened MPC for Socially Compliant Crowd Navigation

Alessandro Febbraro, Federico Rollo, Marcello Farina, Luca Bascetta, Navvab Kashiri

原文摘要

Autonomous navigation of mobile robots in crowded environments, while preserving pedestrians' psychological comfort, requires distinguishing pedestrians from other obstacles, predicting their future motion, and accounting for uncertainty in the predicted trajectories of both pedestrians and the robot. This work extends a previous MPC framework by integrating obstacle and pedestrian detection, Koopman-based motion prediction, and online prediction-error estimation directly within the collision-avoidance mechanism. To reduce collision-avoidance complexity in crowded scenarios, nearby obstacles are aggregated, thereby reducing the number of collision-avoidance constraints. Simulation studies evaluate the prediction-error margins and the computational benefits of semantics-aware aggregation. In a paired 300-seed closed-loop comparison with a moving generic obstacle, Markov residual tightening reduces the fraction of runs with physical contact from $10.0\%$ to $1.33\%$, at a mean travel-time cost of $0.68$ s. In a separate crowd-navigation semantic ablation, the offline robot-footprint/personal-space overlap occupies $2.11\%$ of observed time with pedestrian-aware handling versus $22.84\%$ with generic-obstacle handling.

cs.RO

新论文 · v1arXiv:2611.15779

Honesty Is a Race: Predicting When Verifiable-Reward RL Teaches a Model to Bluff

Nathaniel Bottman, Alexei Oblomkov

原文摘要

Two rewards with the same expected payoff at the start of training can end a reinforcement-learning run in opposite places, with an honest model or one that bluffs. We study rewards that pay $b$ for a verdict that a checker cannot verify ("this program is correct") and $r$ for a certificate that it accepts (an input revealing a bug). To first order, a GRPO step makes the model more honest exactly when $\tilde{p}\,r/b>1+κ$. Here $\tilde{p}$ is the rate at which the checker accepts the model's certificates, weighted toward the items on which the model still gives both verdicts, and $κ$ is the pull of the items on which the unverifiable verdict is right. Both terms move during training. A checker raises $\tilde{p}$, so near the threshold honesty is a race, which a run just below the threshold can win and a run under a lottery (random pay with the same mean) cannot. The term $κ$ starts large and falls as the other items settle, so a run just above the threshold first dips toward bluffing, on code nearly to zero, before climbing back. Paying the honest verdict also pulls the other items toward it, which predicts a ceiling on the payoff ratio, above which the model gives the honest verdict where it is wrong. On propositional Datalog, two code models, and abstention on grade-school math, with $p$ the model's success rate before RL, no run above $pr/b=1$ ends collapsed, and runs under the checker can survive slightly below it. At $pr/b=3$, above the ceiling, both code models call about half of correct programs buggy. On depth-3 Datalog, training generously under the checker and then cutting the payoff keeps the model at least 67% honest in three seeds, at a ratio where the same model trained from the start falls to 4-15% and is at most 47% honest after 160 steps. On depth-2 Datalog, when honesty collapses, a linear probe fit before RL still reads the truth.

cs.LG · cs.AI · cs.CL

新论文 · v1arXiv:2611.15778

From Link Prediction to Linear Community Detection: A Finite-Sample Guarantee for Graph Pretraining

Ying Chen, Zhangyang Wang, Yixuan He

原文摘要

Link prediction (LP) is a standard pretraining objective for graph neural networks (GNNs): hide edges, train an encoder-decoder to predict them, and reuse the node embeddings for downstream node tasks. Despite its empirical success, there is little quantitative theory explaining why an edge-level objective should produce embeddings that are linearly useful for node classification. We study this question in the finite-sample stochastic block model (SBM). For any bounded GNN encoder paired with a symmetric bilinear decoder, we prove a conditional reduction: small population held-out LP excess risk implies small community-classification error for a transductive linear classifier on the learned embeddings. The bound decomposes into an optimization residual, an architectural approximation residual, and explicit statistical terms of order $\tilde{O}(\sqrt{d/n})$ in bounded dense regimes. We further give a constructive few-label classifier: nearest-centroid classification under the positive semidefinite quadratic form $G = W Z^\top Z W$, whose label error is coupled to LP quality. The framework is architecture-agnostic through the approximation residual, and we instantiate this residual as $O(\log n / n)$ for a graph isomorphism network (GIN) with spectral positional encoding and a bilinear decoder. Synthetic SBM experiments validate the theorem chain from LP risk to score error to classification error, and show when the worst-case constants become conservative.

cs.LG · stat.ML

新论文 · v1arXiv:2611.15777

SO-Arena: Benchmarking the Capability Spectrum and Reliability of System One Models

Lichen Li, Ziping Dong, Sicheng Feng, Qi Li, Xinchao Wang

原文摘要

System One models promise fast, general-purpose decision-making without lengthy token generation. However, understanding their capability spectrum and decision reliability requires more than aggregate performance on current benchmarks. Existing evaluations often provide limited insight into the breadth of model capabilities and the reliability of their decisions. We introduce SO-Arena, a benchmark characterizing the capability spectrum and decision reliability of System One models across 11 domains and five capability families. For the capability spectrum, SO-Arena offers evidence-grounded evaluation through reliable sources and manual verification of reference labels against model-visible inputs; fine-grained capability assessment through domain-specific difficulty criteria; and realistic interactive evaluation, where models iteratively execute actions in sandboxes. Reasoning-intensive tasks further enable system one-two comparison. Beyond accuracy, SO-Arena assesses decision reliability through two complementary measures: decision stability under controlled option-set changes and missing-answer recognition when no valid option is available. Evaluations across 21 models reveal that the benefits of scaling and additional reasoning vary substantially across domains. More importantly, strong overall performance does not necessarily imply reliable decision-making: the best-performing System One model ranks only seventh in reliability. Together, these findings highlight the need to characterize capability spectrum and decision reliability as complementary dimensions, providing a systematic view of the strengths and limitations of System One models.

cs.CL · cs.AI · cs.LG

新论文 · v1arXiv:2611.15771

SAME: Semantic Watermark Embedding for Open-Source Large Pretrained Models

Xinyao Song, Jialiang Shen, Hui Kang, Huajie Chen, Ruiqi Liu, Jiyang Zheng, Huaxi Huang, Zhenliang Lu 等 10 位作者

原文摘要

Open-source large pretrained models (LPMs) have become increasingly valuable assets for model development and downstream applications, but their accessibility also facilitates unauthorised copying and modification. To protect their intellectual property (IP), existing approaches typically embed tailored trigger-response patterns as watermarks for ownership verification. However, such trigger designs need to be tailored to task-specific data that constrain their transferability across downstream tasks, and are not durable under subsequent benign fine-tuning. To address these limitations, we propose SemAntic waterMark Embedding (SAME), which introduces a semantically meaningful private vocabulary that enables flexible composition of adaptive and persistent watermarks. Specifically, SAME introduces a fixed private vocabulary that assigns each native token a private embedding counterpart, while aligning the semantic expressiveness of the two vocabularies within the model's representation space. This alignment allows private tokens to be flexibly composed into semantic triggers that elicit meaningful responses, enabling adaptive watermarks beyond predefined trigger-response pairs. By grounding the private vocabulary in the model's native semantics, SAME couples the watermark with its evolving semantic capabilities, allowing this connection to be preserved throughout downstream fine-tuning. Extensive experiments demonstrate that SAME provides reliable and robust watermarks across LPMs of varying scales and architectures, enabling effective ownership verification for IP protection.

cs.LG

新论文 · v1arXiv:2611.15770

Privacy-Preserving Spectrum Data Obfuscation in Next-Generation Spectrum Coexistence Systems

Yalin E. Sagduyu, Tugba Erpek, Sastry Kompella, Joseph A. Molnar

原文摘要

In spectrum coexistence systems, the primary user (PU) activity must be partially disclosed for interference avoidance and coordination, while the exact spectrum occupancy information is often restricted due to security and regulatory constraints. To this end, subset-based disclosures can be employed in which the true frequency is included while ambiguity is introduced. This paper shows that naive randomized subset disclosures remain highly vulnerable to information leakage through inference, as temporal structure can be exploited to recover spectrum usage information. Under Markovian dynamics of the PU's spectrum usage, hidden Markov model (HMM) filtering with online learning is applied, while under periodic dynamics, period and phase estimations with phase-aligned statistics are used for inference. To address this vulnerability, structure-aware obfuscation strategies are introduced. For Markovian dynamics, history-based and belief-based schemes are designed to induce model mismatch and limit posterior concentration, while for periodic dynamics, cycle-dominance and decoy-based schemes are designed to manipulate phase-wise statistics and bias the inference process. These obfuscation schemes significantly hinder inference of spectrum information and offer principled disclosure solutions for spectrum coexistence systems.

cs.NI · cs.CR · eess.SP

新论文 · v1arXiv:2611.15769

PDO Drug Screening Data for Evaluating and Training AI Drug Response Prediction Models

Y. Zhu, K. Marlor, V. Lavelle, T. S. Brettin, M. Sacco, C. Siniscalchi, H. W. Raeder, M. R. Weil 等 9 位作者

原文摘要

Anti-cancer drug response prediction (DRP) is critically important for precision oncology and therapeutic development. Currently, most DRP models are developed using drug screening data from immortalized cancer cell lines, which can differ substantially from patient tumors at the genomic and molecular levels, even though patient tumor response is the ultimate prediction target. To help bridge this gap and bring DRP models closer to their intended clinical application, we conducted four drug screening studies using patient-derived organoids (PDOs), which more faithfully recapitulate patient tumors than immortalized cell lines. These studies generated data for improving and evaluating DRP models, including multi-omics profiles of PDO lines, drug feature data, and treatment response measurements. Using these data, we retrained and evaluated multiple publicly available DRP models under three model-building scenarios: training models using cell line data only, training models using combined cell line and PDO data, and pretraining models on cell line data followed by fine-tuning on PDO data. We then evaluated and compared model prediction performance on PDOs across these analysis scenarios. In addition, we also assessed the consistency of drug screening viability data generated by different laboratories using different assay technologies and/or experimental protocols.

q-bio.QM · cs.AI

版本更新 · v4arXiv:2603.27998

HRIR-Former: Grid-Free Time-Domain Reconstruction of Head-Related Impulse Responses with a Spatially Encoded Transformer

Shaoheng Xu, Chunyi Sun, Jihui Zhang, Amy Bastine, Prasanga N. Samarasinghe, Thushara D. Abhayapala, Hongdong Li

原文摘要

Individualized head-related impulse responses (HRIRs) enable binaural rendering, but dense per-listener measurements are costly. We address HRIR spatial up-sampling from sparse per-listener measurements: given a few measured HRIRs for a listener, predict HRIRs at unmeasured target directions. Prior learning methods often work in the frequency domain, rely on minimum-phase assumptions or separate timing models, and use a fixed direction grid, which can degrade temporal fidelity and spatial continuity. We propose HRIR-Former, a time-domain, grid-free binaural Transformer for reconstructing HRIRs at arbitrary directions from sparse inputs. It uses sinusoidal spatial features, a Conv1D refinement module, and auxiliary interaural time difference (ITD) and interaural level difference (ILD) heads. On SONICOM, it improves normalized mean squared error (NMSE), cosine distance, and ITD/ILD errors over prior methods; ablations validate modules and show minimum-phase preprocessing is unnecessary.

eess.AS · cs.LG

版本更新 · v3arXiv:2606.06857

Interpreting Brain Responses to Language with Sparse Features from Language Models

Michael A. Lepori, Kendrick Kay, Greta Tuckute

原文摘要

A central goal of cognitive neuroscience is to characterize the features that are represented by human language cortex. Artificial language models (LMs) have emerged as a powerful tool to address this challenge, but studies relating biological and artificial representations are often criticized as relating one black box to another. The present work introduces Augmented Sparse Encoding Models, an encoding framework that replaces dense LM hidden states with hierarchically-organized sparse autoencoder (SAE) features, while explicitly including surprisal as a predictor. Using this approach, we (i) produce interpretations of neural responses and (ii) test whether model-brain alignment reflects primary or idiosyncratic variation in LM representations. Using a high-field 7T fMRI dataset of eight participants listening to 200 linguistically diverse sentences, we first validate our modeling framework by recovering previous interpretations of voxel populations tuned to processing difficulty and meaning abstractness. We then interpret a previously-uncharacterized (but reliable) voxel population and find that it is tuned to people-related content. Next, we show that the fronto-temporal human language network is predicted by a common set of features across its constituent regions, but find that frontal regions are relatively well-explained by surprisal alone, even in the absence of LM-based features. Finally, we show that brain responses are best explained by the features that tend to capture the most general information encoded in LM representations, suggesting a nontrivial correspondence between brain and LM language representation.

cs.CL

新论文 · v1arXiv:2611.15766

Prior-as-Skill: Remember Outside and Read Inside with On-Policy Self-Distillation for Self-Evolving VLM Agents

Yan Lin, Jingyu Sun, Yifei Qian, Zhengtao Yao, Yuxia Wang, Daizong Liu, Chun Pong Lau, Zhongliang Guo 等 9 位作者

原文摘要

Self-evolving visual agents keep what they learn as skills: key frames with step notes on a task's steps and their order. On-policy self-distillation (OPSD) is now a popular way to train such skills into a vision-language model that matches, on its own answers, a teacher that also reads the written order. At deployment, frames often go missing or show an unusual order, as when a person or a website reorders the steps. Existing internalization methods keep any order they store in the frame-reading weights: trained on full frames, the model guesses when they are missing, and trained with frames withheld, it stores the order, so memory can override what it sees. We propose PriorSplit, which fills in missing frames from an order kept outside the model, with no training beyond OPSD. It counts which of two neighboring steps comes first in training episodes, and the model checks if both asked steps are in view: if so, it answers from the frames, and otherwise the counts answer. Extensive experiments on synthetic skills with five models and on step-tagged order questions in new episodes of seen web, cooking and egocentric tasks demonstrate that, with frames missing, PriorSplit outperforms every compared method at OPSD's training cost. With Qwen3-VL-4B on synthetic skills, it also stays close to OPSD on unusual orders, where methods that gain most by storing the order fall behind.

cs.CV · cs.AI

新论文 · v1arXiv:2611.15764

AIFR: Adaptive Identity Feature Recovery for Diffusion-Based Makeup and Anime Portraits

Linyan Dai, Xinwei Zhang, Yaxin Xiao, Mengyu Sun, Xinwei Liu, Renyang Liu, Qingqing Ye, Haibo Hu

原文摘要

Diffusion-based makeup and anime portraits are increasingly popular in online content creation (e.g., virtual makeup editing and anime avatar generation), but their altered appearances can hinder face recognition (FR) in identity-based content screening. Existing methods mainly adapt FR encoders or restore input images. The former require costly gallery re-encoding, while the latter may be incomplete or distort identity-related facial details. To address these limitations, we aim to correct identity features within a frozen FR model to recover its output embeddings in its original space. Appearance changes vary in strength and affect facial regions differently. To this end, we propose Adaptive Identity Feature Recovery (AIFR), a framework that inserts a lightweight recovery module at a single intermediate layer to predict sample-level correction strength and spatially adaptive residual updates. Module training uses fused identity targets to guide embedding recovery in the original FR space and an identity-shift score to regulate correction strength and limit unnecessary changes. At inference, AIFR processes a single portrait in one forward pass without reference images, producing recovered embeddings that can be directly matched with existing gallery templates without re-encoding. With only 1.22M trainable parameters, AIFR improves the average TAR at FAR=$10^{-4}$ from 66.1\% to 90.2\% over the original ArcFace, and further experiments show improvements across unseen makeup styles, stylization strengths, and generators. Beyond recognition, the recovered embeddings can directly condition a pretrained Arc2Face model to generate faces with improved identity preservation, showing compatibility with embedding-conditioned downstream generation.

cs.CV

新论文 · v1arXiv:2611.15760

Image Deblending with Diffusion Models

Dan Botchan, Ohad Fried, Yacov Hel-Or

原文摘要

Image deblending is the task of decomposing a composite image back into its source images, which is an ill-posed problem, as a single blended image can be decomposed into multiple plausible source layers. While diffusion models are often used to solve inverse problems, they typically address single-output reconstruction problems. In this work, we formulate deblending as a joint inverse problem, employing diffusion posterior sampling to simultaneously reconstruct two source images, demonstrating that a single diffusion prior is expressive enough to disentangle two full images at once rather than just one. Drawing inspiration from human perception, where semantic understanding helps disentangle visual inconsistencies, we propose a training-free framework that leverages text prompts to constrain this solution space. We address the blind deblending task through a two-stage architecture: text-guided latent diffusion performs semantic separation using prompts as high-level cues, followed by a pixel-space refinement to recover fine details. Stochastic sampling produces multiple plausible decompositions that can be ranked without ground truth. Extensive experiments show that our training-free method achieves state-of-the-art results on unseen domains and extends to spatially-varying alpha maps, while remaining competitive with supervised baselines on standard scalar blending benchmarks.

cs.CV

新论文 · v1arXiv:2611.15758

Physics-based simulation unlocks scaling laws for generalizable MRI reconstruction

Hong Shang, Zijian Zhan, Sen Jia, Xiang Fan, Jintao Meng, Yang Zhou, Xiang Zhou, Tianyu He 等 13 位作者

原文摘要

Deep-learning-based magnetic resonance imaging (MRI) reconstruction holds great promise for accelerating acquisitions, yet its clinical translation remains bottlenecked by poor generalizability across unseen scanners, populations, anatomies, and protocols. Gathering sufficiently diverse clinical datasets to overcome this bottleneck is hindered by high acquisition costs, privacy restrictions, and institutional biases. Here, we introduce Scalable Inverse Model (SIM), a framework that systematically co-designs physics-simulated training data, model architecture, and learning paradigms around scalability--leveraging compute scaling to overcome historical data bottlenecks and achieve robust zero-shot MRI reconstruction. Across 14 public and in-house datasets--spanning seven international centers, five vendors, field strengths of 0.3-7 T, over ten anatomical regions, more than ten acquisition protocols, downstream quantitative and molecular tasks, and a 150-scan blinded radiologist study--SIM demonstrates robust zero-shot generalizability, recovering fine anatomical details while suppressing aliasing artifacts where conventional deep-learning methods fail. Attention-map analyses confirm that SIM learns physically explainable, dynamically adaptive reconstruction primitives rather than anatomy-specific priors. SIM establishes a scalable, privacy-preserving, and democratized infrastructure for universal MRI reconstruction, providing a blueprint for generalizable AI across inverse problems in medical imaging.

eess.IV · cs.CV

版本更新 · v2arXiv:2605.06367

The Interplay of Data Structure and Imbalance in the Learning Dynamics of Diffusion Models

Flavio Nicoletti, Chenxiao Ma, Enrico Ventura, Luca Saglietti, Stefano Sarao Mannelli

原文摘要

Real-world datasets are inherently heterogeneous, yet how per-class structural differences and sampling imbalance shape the training dynamics of diffusion models---and potentially exacerbate disparities---remains poorly understood. While models typically transition from an initial phase of generalization to memorizing the training set, existing theory assumes homogeneous data, leaving open how class imbalance and heterogeneity reshape these dynamics. In this work, we develop a high-dimensional analytical framework to study class-dependent learning in score-based diffusion models. Analyzing a random-features model trained on Gaussian mixtures, we derive the feature-covariance spectrum to characterize per-class generalization and memorization times. We reveal the explicit hierarchy governing these dynamics: class variance is the primary determinant of learning order---consistently favoring higher-variance classes---while centroid geometry and sampling imbalance play secondary roles. Specifically, sampling imbalance acts as a modulator that can reverse this ordering and, under strong imbalance, forces minority classes to acquire distinct, delayed inference times during backward diffusion. Together, these results suggest that diffusion models can memorize some classes while others remain insufficiently learned. We validate our theoretical predictions empirically using U-Net models trained on Fashion MNIST.

stat.ML · cond-mat.dis-nn · cs.LG

版本更新 · v2arXiv:2503.11321

Fractal-Informed Spatial-Frequency Adaptation for Texture Generation in Natural Image Compression

Lingyu Zhu, Baoliang Chen, Xiangrui Zeng, Chengyan Jiang, Liang Xie, Shiqi Wang

原文摘要

Generative image compression aims to reconstruct realistic images at low bitrates by balancing the trade-off among rate, distortion, and realism. However, recovering plausible textures while preserving reconstruction fidelity remains challenging, particularly under limited bandwidth. In this paper, we propose a diffusion-informed generative image compression framework that exploits frequency-domain representations and diffusion knowledge to improve texture reconstruction. Specifically, we approach the rate-distortion-realism trade-off from three complementary perspectives. First, motivated by the close relationship between image textures and frequency characteristics, we develop a Fractal Frequency-Aware Band Image Compression (FFAB-IC) network to capture directional frequency components in natural images. By integrating fractal band operations into nonlinear neural transformations, the network preserves essential image information while suppressing redundant details. Second, we incorporate diffusion knowledge into the encoder and introduce iterative diffusion refinement into the decoder, enabling the recovery of realistic textures under constrained bitrates. Third, we introduce frequency- and content-aware regularization to jointly exploit spatial structures and frequency characteristics during optimization. Extensive quantitative and qualitative experiments demonstrate that our method achieves a superior rate-distortion-realism trade-off, delivering higher realism at comparable distortion than existing approaches.

cs.CV · eess.IV

新论文 · v1arXiv:2611.15750

Sensitivity-Aware Surrogate Modeling with Random Feature Expansions

John E. Darges

原文摘要

Random feature expansions (RFEs) provide a flexible approach to surrogate modeling, but their accuracy depends on the distribution used to sample features. We propose spike-and-slab random feature expansions (SAS-RFE) to learn the feature sampling distribution from sensitivity analysis. The method employs a spike-and-slab distribution that assigns probabilities to the inclusion of individual inputs in randomly sampled Fourier features. These inclusion probabilities are determined by estimates of total Sobol' sensitivity indices, rather than by explicitly selecting important inputs or interaction terms. An information theoretic criterion shows that the total Sobol' indices yield a distribution that approximates the input interaction structure of the model. An iterative algorithm updates the sampling probabilities using sensitivity estimates from successive RFE surrogates. Analytical formulas for the Sobol' indices of random Fourier feature expansions enable these updates without additional evaluations of the target function. Numerical experiments on six benchmark problems, with up to 100 input dimensions, demonstrate improvements in surrogate approximation and sensitivity estimation.

cs.LG · stat.CO

版本更新 · v3arXiv:2601.19070

Critical Organization of p-Adic Deep Neural Networks

W. A. Zúñiga-Galindo

原文摘要

We study continuous neural networks whose neurons are the points of the ring of p-adic integers, a fractal tree-like space that encodes hierarchical architectures. The hidden states solve a nonlinear integral equation, which describes the equilibria of recurrent networks and deep equilibrium models; feedforward architectures appear as the special case of kernels without feedback. We prove that the state is unique and depends Lipschitz-continuously on the parameters whenever the product of the Lipschitz constant of the activation function and the operator norm of the connectivity is less than one, and that, when the parameters are locally constant, every state is exactly the state of a finite network with a tree-like topology. We introduce p-adic discrete deep networks, whose layers are the levels of the tree and whose weights are coarse-grainings of a single kernel, and show that, in the stable region, they converge to the continuous network as the depth tends to infinity. Feedforward networks always have a unique state; hence a multiplicity of states, and therefore critical organization in our sense, requires recurrence. For recurrent networks with radial kernels we prove a sharp transition: the zero state is the only state if and only if the total coupling does not exceed one. For hierarchical kernels, above the threshold, the states are indexed by sign patterns on the leaves of the tree, their number is computed exactly, and they form a tree-indexed partially ordered set graded by instability. Numerical experiments illustrate these results and test conjectures on general radial kernels and on random recurrent networks on the tree. Finally, we interpret the critical organization hypothesis for deep neural networks, originally formulated for brain organization, through the lens of multistability, and we discuss the continuous limits of families of finite networks.

cs.LG

新论文 · v1arXiv:2611.15748

Minimizing Monotone Norms in Fractional Load Balancing

Daniel Blankenburg, Antonia Ellerbrock, Thomas Kesselheim, Jens Vygen

原文摘要

We devise a fully polynomial time approximation scheme for minimizing a monotone norm $\left\lVert\cdot\right\rVert$ over a convex set $X\subseteq \mathbb{R}^d_{\geq 0}$. We study the multi-customer setting, where $X=\sum_{i=1}^n X_i$ and the solution set $X_i\subseteq \mathbb{R}^d_{\geq 0}$ is given by a linear minimization oracle for each customer $i$. Our algorithm computes, for any given $\varepsilon>0$, a $(1+\varepsilon)$-approximate solution. With high probability, it uses $\tilde{\mathcal{O}}\left( \varepsilon^{-2}(n+d) \right)$ single-customer linear minimization oracle calls and the same number of generalized KL projections to the dual space $Y=\{y\in \mathbb{R}^d_{\geq 0}: \left\lVert y \right\rVert^* \le 1\}$. We assume no prior knowledge on $X$, and our running time depends only on $n$, $d$, and $\varepsilon$. Our result is robust in the sense that it allows for approximate linear minimization oracles and approximate KL projections. We show that the (approximate) KL projection can be performed efficiently for large classes of monotone norms, demonstrating the applicability of our approach. While such a result (with exact projection) is well-known for the $\ell_{\infty}$ norm via multiplicative weight updates and was recently obtained for ordered norms via a smooth norm approximation and a novel randomization, existing techniques do not extend to general monotone norms. In particular, a key feature used in previous proofs was dual price stability, which was derived from a global contraction property of the KL projection to the dual space $Y$. We show that this property does not extend to general monotone norms. Therefore, we use a different price mechanism, based on the well-known mirror descent paradigm, for which a local contraction property holds and guarantees stability.

math.OC · cs.DS

新论文 · v1arXiv:2611.15744

Latent Imagination for Parameterized Action Spaces

Atahan Cilan, İhsan Mert Muhacıroğlu, Ümit Can Bekar

原文摘要

Parameterized-action MDPs (PAMDPs) ask an agent to choose a discrete action together with continuous parameters whose meaning depends on that choice. Model-free PAMDP methods act fast but need many interactions; model-based methods are sample-efficient but plan at every decision, which makes them slow to act. We introduce HyLI (Hybrid Latent Imagination), a model-based agent that learns a hybrid policy entirely in the latent imagination of a world model and acts with a single forward pass. To address learning problems that the PAMDP structure creates, HyLI adds counterfactual mode evaluation, which gives every discrete action a learning signal at every imagined state, and success-window replay, which keeps rare successes in training. On seven standard PAMDP benchmarks, HyLI is within one published standard deviation of, or above, the published score of the strongest model-free baseline in our comparison on all 7 tasks, reaches 90% of that score with 2.4x fewer environment steps than that baseline (5.0x when its reward-free pre-training is counted), and decides 67x faster than the model-based planner DLPA at its published planner settings. It is the only method in our comparison that is both more sample-efficient than that baseline and fast enough for real-time control.

cs.LG · cs.AI

版本更新 · v3arXiv:2606.15015

NEXUS: Neural Energy Fields for Physically Consistent Contact-Rich 3D Object Dynamics

Qizhen Ying, Guangming Wang, Yangchen Pan, Victor Adrian Prisacariu, Brian Sheil, Yixiong Jing

原文摘要

Physics-grounded video generation requires controllable 3D object dynamics that remain physically consistent under contact, deformation, and external forcing. Existing trajectory-based methods often model isolated physical effects, making it difficult to compose conservative and non-conservative dynamics in contact-rich 3D scenes. We present NEXUS, a neural energy-field framework for contact-rich 3D object dynamics. NEXUS represents each object as a structural graph and constructs dynamic object-object and object-environment contact graphs. Inspired by Hamiltonian Neural Networks, NEXUS formulates motion through scalar energy and dissipation terms rather than directly predicting states or accelerations. Conservative effects, including gravity and elastic deformation, are composed as additive energy terms, while non-conservative effects such as damping and impact-induced energy loss are modeled with learned Rayleigh-style dissipation. Forces are derived by differentiating the energy and dissipation functions and rolled out with a multi-substep semi-implicit integrator. Across controlled trajectory benchmarks, NEXUS improves long-horizon accuracy over representative learned and physics-structured dynamics baselines under varying mechanical properties and physical-effect compositions. We further show that NEXUS trajectories provide effective guidance for contact-rich video generation, improving physical plausibility while maintaining competitive visual quality.

cs.CV · cs.AI

版本更新 · v2arXiv:2603.10055

Training Language Models via Neural Cellular Automata

Dan Lee, Seungwook Han, Akarsh Kumar, Pulkit Agrawal

原文摘要

Pre-training is crucial for large language models (LLMs), as it is when most representations and capabilities are acquired. However, natural language pre-training has problems: high-quality text is finite, contains human biases, and entangles knowledge with reasoning. This raises a fundamental question: is natural language the only path to intelligence? We propose using neural cellular automata (NCA) to generate synthetic, non-linguistic data for pre-pre-training LLMs--training on synthetic-then-natural language. NCA data exhibits rich spatiotemporal structure and statistics resembling natural language while being controllable and cheap to generate at scale. We find that pre-pre-training on only 164M NCA tokens improves downstream language modeling by up to 6% and accelerates convergence by up to 1.6x in 400M-3B models trained with Chinchilla optimal data budgets. Surprisingly, this even outperforms pre-pre-training on 1.6B tokens of high-quality natural language data with more compute. Investigating what drives transfer, we find that attention layers are the most transferable, and that optimal NCA complexity varies by domain: code benefits from simpler dynamics, while math and web text favor more complex ones. These results enable systematic tuning of the synthetic distribution to target domains. More broadly, our work opens a path toward more efficient models with fully synthetic pre-training.

cs.LG · cs.AI · cs.CL

版本更新 · v2arXiv:2611.06486

Model Merging Does Not Require Uniform LoRA Ranks

Zirong Xie, Zainab Afolabi, Nikita Kozodoi, Jack Butler

原文摘要

Model merging combines separately fine-tuned experts into a single model without retraining. Standard practice trains every expert at the same LoRA rank, which couples the rank choice across experts even though each domain has its own best rank. We show this coupling is a property of the merge method, not of merging. We train 36 experts on three domains (mathematics, instruction following, multilingual) at four ranks and three model scales, then merge with six methods at uniform ranks and at each domain's own best rank. Giving experts different ranks costs averaging and utility-based selection nothing, and gains them up to 1.8 pp of multi-domain accuracy, while the conflict-resolution methods are inconsistent and can lose up to 4 pp. The split has a simple cause. Merging keeps only a fixed share of the pooled spectral components, and methods that choose them by size discard many that would have reduced loss. They also tend to discard the same expert: the instruction adapter has the smallest singular values, so it is dropped almost entirely and IFEval falls by 16 to 25 pp. Choosing components by their effect on loss instead keeps all three experts, and leaves accuracy nearly flat across an 8$\times$ rank range (0.7 pp, against 6.7 pp for TIES at 4B scale). Rank can therefore be chosen per expert whenever the merge operator does not select by magnitude.

cs.LG · cs.AI

版本更新 · v2arXiv:2609.13062

Anchoring Clinical Events in Time: UID-Preserving Multimodal Reconstruction and Source-Grounded Adjudication

Sayantan Kumar, Nicolas Grimaldi, Jack Cummins, Jeremy C. Weiss

原文摘要

Interpreting a patient's hospital course depends on when symptoms developed, tests were performed, and treatments began. Discharge summaries often leave this chronology unclear, while structured electronic health record (EHR) tables timestamp only some of these events. We reconstruct clinical timelines by retrieving structured evidence for narrative events and jointly revising their estimated times. Each extracted occurrence carries a persistent unique identifier (UID) linked to its source span, keeping repeated mentions distinct throughout reconstruction. We also introduce GAVEL, a large language model (LLM) judge that checks disagreements between two timelines against the note and structured record and cites evidence for its verdicts. We evaluated six open-weight models on 40 clinician-annotated discharge summaries. Our pipeline recovered more reference events than LLM-Timeline and TKW2 when each comparison used the same backbone. For GLM-5.2, adding structured evidence improved timestamp agreement with essentially unchanged event recovery; ablations showed that links to timestamped source rows contributed to this improvement. Results varied across the other models, with some timing gains accompanied by fewer recovered events. Both GAVEL judges, GLM-5.2 and MiMo-V2.6-Pro-RL, favored multimodal over text-only GLM-5.2 timelines, while neither clearly distinguished the DeepSeek V3.2 variants. Manual review upheld most sampled GAVEL findings. The source links allow reconstructed times to be checked, and GAVEL identifies disagreements for human review.

cs.AI