paperlist.ai

AI conference papers

AI conference paper library

Search papers from 30+ top conferences, including NeurIPS, ICLR, ICML, CVPR, ACL and AAAI. Filter by keyword, then open any paper for its abstract, authors and links.

arXiv

TreeHop: Efficient Embedding-Level Query Rewriter

Zhonghao Li, Kunpeng Zhang, Jinghuai Ou, Shuliang Liu, Xuming Hu

Retrieval-augmented generation (RAG) systems face significant challenges in multi-hop question answering (MHQA), where complex queries require synthesizing information across multiple document chunks. Existing approaches typically rely on iterative LLM-based query rewriting and routing, resulting in high computational costs due to repeated LLM invocations and multi-stage processes. To address these limitations, we propose TreeHop, an embedding-level framework without the need for LLMs in query refinement. TreeHop dynamically updates query embeddings by fusing semantic information from prior queries and retrieved documents, enabling iterative retrieval through embedding-space operations alone. This method replaces the traditional "Retrieve-Rewrite-Vectorize-Retrieve" cycle with a streamlined "Retrieve-Embed-Retrieve" loop, significantly reducing computational overhead. Moreover, a rule-based stopping criterion is introduced to further prune redundant retrievals, balancing efficiency and recall rate. Experimental results show that TreeHop rivals advanced RAG methods across four open-domain MHQA datasets, achieving comparable performance with only 2.2\%-29.4\% of the parameter size of concurrent solutions and reducing the query latency by 92.8\%-97.8\%. This makes TreeHop a faster and more cost-effective solution for low-resource or latency-sensitive deployment. For reproducibility purposes, codes and data are available here: https://github.com/allen-li1231/TreeHop-RAG

arXiv

Analogical Learning for Cross-Scenario Generalization: Framework and Application to Intelligent Localization

Zirui Chen, Hongning Ruan, Zhaoyang Zhang, Ziqing Xing, Ridong Li, Zhaohui Yang, Mérouane Debbah

Modern learning systems often struggle with joint learning across diverse scenarios and immediate adaptation to new ones, because they rely heavily on the scenario-dependent absolute data-label representations. Here, we propose analogical learning (AL), a learning framework that explores the inherent invariance of the underlying physical processes across scenarios, to improve the cross-scenario generalization. Specifically, we introduce the physical concepts of reference frames and relativity into the neural modeling. The resultant framework explicitly employs intra-scenario data-label pairs as reference anchors and enforces the network to mediate its data-to-label transformation through data-domain relative metrics that factor out the scenario-dependent variations. We instantiate AL with Mateformer, a bipartite Transformer-based neural architecture. Each layer of the auxiliary Transformer extracts certain feature space of the current data, while the corresponding layer of the primary Transformer computes attention among the data feature space and then use it as a relativity metric to weight the current label feature to synthesize the next label feature and, ultimately, the final prediction. We apply AL to intelligent wireless localization, a representative multi-scenario learning task. Across synthetic, real-world, and city-scale datasets, AL enables robust cross-scenario transfer and multi-scenario joint learning, achieving wavelength-scale localization accuracy that matches or surpasses state-of-the-art methods. This physics-inspired learning framework provides a promising alternative for other cross-scenario learning tasks and applications.

arXiv

Advancing Intelligent Sequence Modeling: Evolution, Trade-offs, and Applications of State-Space Architectures from S4 to Mamba

Shriyank Somvanshi, Md Monzurul Islam, Mahmuda Sultana Mimi, Sazzad Bin Bashar Polock, Gaurab Chhetri, Anandi Dutta, Amir Rafe, Subasish Das

Structured State Space Models (SSMs) have become a prominent class of sequence models, developed against two long-standing difficulties: the sequential computation and gradient propagation limits of Recurrent Neural Networks (RNNs), and the quadratic time and memory cost of self-attention in Transformers. By combining structured recurrence with state-space representations, SSMs attain linear or near-linear scaling in sequence length and hold a constant-size recurrent state during autoregressive decoding, so that no key-value cache grows with context. This paper is a structured review of the lineage running from the Structured State Space Sequence model (S4), through its diagonal and simplified successors S4D, Diagonal State Spaces (DSS) and S5, to the selective models Mamba and Mamba-2, and on to SSM-attention hybrids. Rather than describing models one at a time, the analysis is organized around cross-cutting design dimensions: input-dependent selectivity, the recurrent and convolutional views and when each is preferable, diagonalization, cache behavior at inference, hardware-aware kernels, and hybridization. Structured state-space duality is treated as a central thread, because it accounts for the equivalence between the recurrent and the attention-like form and for why hybrid designs work. Reported efficiency and accuracy results are presented with their measurement configuration and evidence level, because observed speedups reflect the combined effects of the algorithm, implementation kernel and hardware rather than an intrinsic property of the model alone. SSMs are competitive with Transformers and more efficient in specific long-context regimes, whereas attention retains an advantage on tasks dominated by exact retrieval and associative recall. The review closes with limitations, failure modes, deployment considerations and open problems.

arXiv

On the Effect of Sampling Diversity in Scaling LLM Inference

Tianchun Wang, Zichuan Liu, Yuanzhou Chen, Jonathan Light, Weiyang Liu, Haifeng Chen, Xiang Zhang, Wei Cheng

Large language model (LLM) scaling inference is key to unlocking greater performance, and leveraging diversity has proven an effective way to enhance it. Motivated by the observed relationship between solution accuracy and meaningful response diversity, we systematically study the effect of prompt diversity in scaling inference. We theoretically explain why diversified sampling improves Best-of-$N$ scaling, showing that responses generated from diverse prompts after Best-of-$N$ selection exhibit significantly lower error rates than those produced from stationary prompts. Building on this analysis, we derive a diversity-fidelity trade-off principle, that guides the design of sampling strategies introducing diversity. From this guidance, we instantiate a family of effective perturbation styles. We theoretically and empirically characterize \textbf{when} diversified exploration remains effective, demonstrating that it works under a variety of conditions, and we further show that under majority voting, diversity may vanish. Finally, we systematically evaluate the effectiveness of sampling diversity and show that, when applied appropriately in different contexts, meaningful perturbations yield stronger, task-dependent gains as diversity increases. Overall, this work provides a systematic analysis that offers a theoretical and empirical foundation for understanding how sampling diversity affects LLM inference-time scaling.

arXiv

Machine Learning and Data Analysis Using Posets: A Survey

Arnauld Mesinga Mwafise

Partially ordered sets (posets) are discrete mathematical structures that formalize the notion of comparison without forcing every pair of objects to be comparable. This makes them a natural representation for the many machine learning and data-analysis settings in which objects are related by dominance, containment, priority, or refinement relations rather than by a single scalar score. Over the past two decades, a substantial and fragmented literature has connected posets and lattice theory to ranking, clustering, formal concept analysis, multidimensional and multi-criteria data analysis, structured and safe learning, graph and topological deep learning, and explainable artificial intelligence, spanning a wide range of application domains. Despite this activity, the field has lacked (i) an organizing taxonomy that relates these disparate strands of work, and (ii) an up-to-date account extending through 2025--2026 that incorporates recent developments in poset-structured learning methods, including poset-structured safety layers for reinforcement learning, poset-valued neural pooling operators, functor-calculus approaches to multiparameter persistent homology. This survey addresses both gaps. We propose a four-axis taxonomy of poset-based methods (representation, learning paradigm, data modality, and task), provide a comprehensive and comparative review of representative models and algorithms organized along this taxonomy, curate an extensive collection of datasets, software packages, and algorithmic resources, and close with a critical discussion of open theoretical and practical problems -- including model depth and expressivity, scalability--fidelity trade-offs, heterogeneity of order-structured data, and dynamicity of posets over time -- that we argue define a research agenda for the next generation of order-aware machine learning.

arXiv

Matrix Completion via Nonsmooth Regularization of Fully Connected Neural Networks

Sajad Faramarzi, Farzan Haddadi, Sajjad Amini, Masoud Ahookhosh, Symeon Chatzinotas

Conventional matrix completion methods approximate the missing values by assuming the matrix to be low-rank, which leads to a linear approximation of missing values. It has been shown that enhanced performance could be attained by using nonlinear estimators such as deep neural networks. Deep fully connected neural networks (FCNNs), one of the most suitable architectures for matrix completion, suffer from over-fitting due to their high capacity, which leads to low generalizability. In this paper, we control over-fitting by regularizing the FCNN model in terms of the $\ell_{1}$ norm of intermediate representations and nuclear norm of weight matrices. As such, the resulting regularized objective function becomes nonsmooth and nonconvex, i.e., existing gradient-based methods cannot be applied to our model. We propose a variant of the proximal gradient method and investigate its convergence to a critical point. In the initial epochs of FCNN training, the regularization terms are ignored, and through epochs, the effect of that increases. The gradual addition of nonsmooth regularization terms is the main reason for the better performance of the deep neural network with nonsmooth regularization terms (DNN-NSR) algorithm. Our simulations indicate the superiority of the proposed algorithm in comparison with existing linear and nonlinear algorithms.

arXiv

Motion Artifact-Aware Self-Supervised Representation Learning for 3D Brain MRI Motion Artifact Reduction

Mojtaba Safari, Shansong Wang, Zach Eidex, Matthew Goette, Tonghe Wang, Zhen Tian, Xiaofeng Yang

Patient motion remains a source of image degradation in brain MRI, leading to signal loss, blurring, and geometric distortion that compromise quantitative analysis. Existing deep learning methods for motion correction typically rely on paired clean-corrupted data or k-space acquisitions, which are rarely available in clinical settings. We propose SSRL-MAR, a motion artifact-aware unpaired representation learning framework for motion artifact reduction that requires neither paired training data nor explicit motion labels. SSRL-MAR employed a three-stage training strategy: (1) contrastive learning on 3D patches to extract motion representations by contrasting clean and synthetically corrupted images, (2) a motion artifact-aware synthesis network to generate motion artifacts from clean scans, and (3) a motion artifact-aware generator to restore clean volumes using the learned degrader for self-supervised supervision. On in-silico dataset, SSRL-MAR achieved PSNR 23.81dB, SSIM 91.55%, and NMSE 0.79%. On in-vivo MR-ART dataset, the pretrained model reduced motion distortion, and unsupervised domain adaptation further improved anatomical fidelity. Against a source-only supervised model trained on the same simulated pairs, SSRL-MAR improved PSNR by up to 2.0 dB on MR-ART after unsupervised domain adaptation, and remained within 0.25-0.47 dB of an oracle supervised model that requires real paired data unavailable in practice. At the milder motion level, volumetric error in structures such as the corpus callosum and ventricular system decreased by more than 50%, confirming improved neuroanatomical consistency. These results indicate that SSRL-MAR provides a robust and scalable image-domain solution for 3D brain MRI motion correction, enabling reliable structural quantification in large-scale neuroimaging studies without requiring prospectively acquired pairs or acquisition-specific calibration.

arXiv

On Cross-Validation for Hyperparameter Optimization of Deep Learning Image Classifiers

Ljubomir Buturovic

Hyperparameter optimization (HPO) can materially affect the performance of deep learning (DL) image classifiers, but there is little empirical guidance on how to derive the validation signal that drives it, especially for the small sample sizes common in fields such as medical imaging. We compared three HPO protocols in terms of {\em absolute performance-estimation error} (AEE; the absolute difference between the winning configuration's validation AUROC and its test AUROC): fixed holdout (F), reshuffled holdout (R), and 5-fold cross-validation (C). The search space, sampler, training procedure, architecture, and test set were held identical across protocols. We evaluated the protocols on three public datasets spanning two regimes: binary medical imaging (RSNA pneumonia radiographs and binarized HAM10000 skin lesions) and 200-class natural imaging (Tiny ImageNet), across a range of development set sizes $n$ and two backbones (ResNet-18 on all datasets, Vision Transformer (ViT-S/16) on RSNA). On the medical datasets, every point estimate favored cross-validation over both holdout protocols, with reductions in AEE largest at small sample sizes and diminishing as $n$ increased. This pattern remained robust under conservative family-wise adjustment. On Tiny ImageNet, AEE was negligible under all three protocols. Test AUROC was generally similar among protocols. Fixed holdout had lower mean AEE than reshuffled holdout in 11 of 12 medical conditions, although this secondary finding was less uniformly supported. For small-sample medical image classification, we recommend cross-validation-based HPO when computational resources permit because it trades additional computation for a more reliable development-time estimate of subsequent test performance.

arXiv

Deep Analog: Open-Set Film Emulation with Reference-Conditioned 3D LUTs

Yitong Mu

Film emulation reproduces the look of an analog film stock on a new digital photograph. We target its open-set form -- matching any reference film frame from a single example -- with a 3D lookup table (LUT) predicted from that reference. Real-time image enhancement predicts per-image weights over a fixed bank of 3D LUTs and blends them. We show this is a gated mixture of experts and inherits its failure: trained end-to-end against reconstruction, the gate collapses onto a single expert, so a bank of K LUTs delivers the capacity of one. An entropy term, the enhancement-setting analogue of mixture-of-experts load balancing, restores utilization and recovers about 1 dB PSNR. The deeper constraint survives: a fixed LUT basis is closed-set, freezing the achievable looks at training time. We therefore discard the basis and predict a single 3D LUT as a residual from a reference image (StyleLUTNet), trained by self-supervision on procedurally generated color transforms. The conditional design removes the gate and generalizes open-set to unseen film stocks without paired data or retraining. Around this color backbone we build Deep Analog, a film-emulation pipeline that adds histogram-based tone matching and a physics-informed optical renderer -- multi-scale grain and per-channel halation driven by parameters an inverse network regresses from the reference. On 350 self-supervised pairs the color stage reaches 22.05 dB PSNR / 0.925 SSIM and the full pipeline 21.72 dB / 0.923; the color path runs in 5.2 ms at 1080p (192 FPS) and exports a portable .cube LUT for standard editing tools. A second degeneracy in conditional LUT training -- residual-scale collapse -- shares the root cause and yields a general principle: auxiliary regularization must stay subordinate to reconstruction.

arXiv

Periocular Soft Biometrics: A Survey and Applications to Multimedia Forensics and Disinformation Detection

Fernando Alonso-Fernandez, Kevin Hernandez-Diaz, Josef Bigun

Soft-biometric attributes such as gender, age, and ethnicity provide valuable ancillary evidence when full identity recognition is not feasible, supporting applications in forensic investigation, identity verification, surveillance, or detection of synthetic and manipulated media. Among biometric modalities, the periocular region is a robust source of soft-biometric cues, as it often remains visible when other parts of the face are occluded, a frequent condition in forensic evidence and surveillance footage, and can be captured across a wide range of acquisition conditions. In this paper, we provide a survey of demographic attribute estimation from periocular images, covering publicly available datasets, methodological trends from handcrafted descriptors to deep learning architectures, and the state of the art in gender, age, and ethnicity prediction. We discuss use cases relevant to multimedia forensics and disinformation-detection applications, including demographic filtering in surveillance footage, age verification, and the detection of demographic inconsistencies in synthetic data. We also highlight open challenges, including dataset bias, cross-domain generalisation, fairness, ethical aspects, and the lack of forensic-oriented benchmarks.

arXiv

Xemo-Talker: Unlock Emotions Explicitly for Audio-Driven Talking Portrait Synthesis

Chaolong Yang, Yinuo Guo, Kai Yao, Yuyao Yan, Jie Sun, Guangliang Cheng, Shibin Wu, Bin Dong, Kaizhu Huang

Precise emotion control in audio-driven talking heads remains a challenge due to the reliance on implicit emotion regulation in existing systems, which often leads to indirect and insufficient control. Additionally, training with explicit emotion-related losses across the entire motion space poses significant difficulties due to the inherent trade-off between accurate lip synchronization and fine-grained emotion control. In this paper, we reveal a key finding: although emotional cues are distributed throughout the motion space, concentrating discriminative supervision on less-principal components achieves a better emotion-lip synchronization balance, as principal components mainly encode high-energy articulation and pose variations. Building on this insight, we propose Xemo-Talker, which first learns a neutral speech-to-motion mapping for stable articulation and lip synchronization, and then introduces a lightweight emotion branch guided by less-principal subspace supervision. To enhance emotion control, we design a Tri-Loss consisting of inter-class separation, intra-class compactness, and less-principal contrastive learning. Given an audio input, a reference image, and an emotion label, Xemo-Talker achieves state-of-the-art emotion classification accuracy while maintaining competitive lip synchronization and high inference efficiency, with performance approaching that measured on real videos.The source code is publicly available at https://github.com/chaolongy/Xemo-Talker.

arXiv

Sycophants in the Courtroom: Are LLMs Fragile to Juridical Authority and Evolving Legal Standards?

Lorenzo Molfetta, Alessio Cocchieri, Luca Ragazzi, Ilaria Bartolini, Marco Patella, Gianluca Moro

In medicine, claims remain valid when supported by empirical evidence grounded in stable biological reality. In law, by contrast, truth is contingent, defined by jurisdiction, temporal validity, and the hierarchy of authoritative sources. The recent success of large language models (LLMs) on medical licensing examinations has encouraged an expectation of comparable legal competence. This analogy, however, obscures a critical distinction between domains. Unlike in medicine, legal performance often depends less on inference than on determining when external authority is applicable, valid, and non-contradictory. We introduce a comparative diagnostic framework evaluating legal reasoning against medical baselines along four axes (knowledge recall, grounding, confidence, and robustness), uncovering a sharp domain asymmetry when applied to a new benchmark that encodes temporal validity and normative relationships. While medical LLMs reliably benefit from verified sources, legal LLMs struggle to assess when retrieved citations are useful or misleading, exhibiting overconfidence in perturbed contexts and sensitivity to superficial formatting cues. Increased model scale amplifies this tendency, revealing that stronger instruction following can coincide with weaker resistance to authoritative perturbations. These findings show that LLMs treat law as unstructured text rather than binding precedent, while revealing a tendency to over-trust authoritative but false information when external references conflict with a model's internal knowledge.

arXiv

Unsupervised Domain Adaptation for Multitask Image Analysis in Realistic Context with Extreme Label Shift; Application to the CTAO first Large Sized Telescope

Michaël Dell'aiera, Thomas Vuillaume, Alexandre Benoit

Unsupervised domain adaptation is a widespread set of methods that leverages the knowledge of a labeled source domain to train a model to perform well on a related unlabeled target domain. They generally introduce an auxiliary adaptation-related task that can be integrated into the multitask paradigm, which aims to merge multiple single-task models into a unified architecture. In this paper, we propose to associate domain adaptation and multitask balancing in the realistic context of an extreme class imbalance. Therefore, we propose a combined framework to cover and validate these approaches, and evaluate its performance in the physics-based context of the Cherenkov Telescope Array Observatory (CTAO). Along with a comparative study of some relevant adaptation techniques, we highlight the impact of extreme label shift and extend the investigations on importance weighting to rectify it. The complete code and results are published and available as open-source resources on Zenodo.

arXiv

Decodable but Not Accessible: Auditing Distance-Based Reliability Estimation on Disentangled Skin-Lesion Representations

Duc-Vinh Tran

Distance-based reliability estimation assumes that a representation's geometry reflects its trustworthiness, yet this assumption is rarely tested under training interventions that reshape geometry directly. We audit this assumption under domain-adversarial representation learning using a disentanglement dose-response ladder. Three checkpoint families share the same architecture and a 16-dimensional representation, differing only in orthogonality strength (lambda = 0, 1, 5). Representation geometry changed substantially with disentanglement strength: the condition number shifted by two orders of magnitude (Kendall tau = 0.84, exact p = 2.8e-5). This change was not accompanied by improved reliability estimation: Mahalanobis-distance AUROC (ISIC-test vs. PAD-UFES) remained flat and below chance (about 0.40) at every level, with no significant association with any of five geometry metrics tested. The same failure was observed for cosine-to-centroid and pooled k-nearest-neighbor scorers, plus three non-distance-based scorers: an energy-based confidence score, Virtual-Logit Matching, and a kernel density estimator. Seven of eight scorers converged on the same result; the energy-based score showed an isolated upward trend that we report but do not treat as evidence against the overall pattern. A supervised probe with no access to the training objective recovered domain membership from the identical embeddings at 0.72-0.81 AUROC across every level, showing that the relevant information was not absent from the representation. These findings indicate that classification performance alone can overlook whether information in a learned representation is organized in a form that downstream reliability estimators can use. Information can remain decodable while becoming largely inaccessible to non-probing reliability estimators.

arXiv

GeoUniPR: A Geometry-Consistent Unified Framework for Cross-Modal Place Recognition

Wonbong Kim, Jiatong Xiao, Rui Li, Xufei Wang, Qiwen Gu, Junqiao Zhao, Chen Ye, Guang Chen

Cross-modal place recognition (CMPR) aims to identify the same location across heterogeneous sensing modalities, such as vision and LiDAR. Existing methods commonly bridge the modality gap using complex alignment modules, multi-stage training, or full fine-tuning of pretrained backbones. In this work, we revisit CMPR from the perspective of geometric consistency and propose GeoUniPR, a unified and concise geometry-consistent framework. GeoUniPR reduces cross-modal discrepancy at the representation level by projecting LiDAR point clouds into the camera perspective to construct Geometry-Consistent depth image views (DIV), which establish direct RGB-LiDAR correspondence. We further augment DIV with native LiDAR cues, including intensity and surface-normal information, yielding a multi-channel geometric representation that improves structural consistency. Based on this representation, GeoUniPR learns a unified embedding space using two modality-specific ViT-based encoders with identical architectures, trained through parameter-efficient adaptation without auxiliary alignment modules, multi-stage training, or full backbone fine-tuning. In addition, we introduce Spatially-Consistent InfoNCE (SC-InfoNCE), a CMPR-specific contrastive objective that suppresses distance-induced false negatives under spatial continuity. Extensive experiments on KITTI and KITTI-360 demonstrate that GeoUniPR achieves state-of-the-art (SOTA) performance in both same-modal and cross-modal place recognition, with strong cross-dataset generalization.

arXiv

World Tokens: Enhancing Embodied Policies with Training-Time World Modeling

Qu Tang, Benhui Zhuang, Bo Yuan, Xue Yu, Longteng Guo, Junlan Feng

Vision-language-action (VLA) models are a widely adopted paradigm for embodied policies. They excel at efficient closed-loop control but do not explicitly model how physical scenes evolve as a task unfolds. Recently emerging world-action models (WAMs) leverage pretrained video world models to capture spatiotemporal evolution, yet retaining future generation or a large video backbone in the control loop substantially increases inference cost. We introduce World Tokens, an embodied policy architecture built around a World Adapter that bridges visual-language understanding, world-dynamics modeling, and action generation. It uses world modeling during training to enhance the action policy while preserving efficient deployment. Specifically, the World Adapter transforms VLM features into a fixed set of world tokens, which condition a jointly fine-tuned future-video denoiser and simultaneously serve as the action expert's sole visual-language context. This shared conditioning allows gradients from future-video denoising to directly shape the representation used for action prediction, while exclusive routing prevents the policy from bypassing that representation. At deployment, the world-model branch is removed, leaving only the VLM, World Adapter, and action expert, with no online video-model inference. With a 2B backbone and no embodied action pretraining, World Tokens is highly competitive on LIBERO, attains the best reported averages on SIMPLER, substantially improves real-world R1 Pro success over a matched action-only baseline, and generates each action chunk at VLA-level latency.

arXiv

A Hybrid Neural-Microfacet BRDF Model for Real-Time Rendering

Louis De Oliveira, Anastasia Karpova, Georges Nader, Antoine Houdard, Pierre Mezieres, Damien Rioux-Lavoie, Romain Pacanowski

Over the past decade, microfacet-based BRDF models have formed the foundation of real-time rendering pipelines. Despite their widespread use, they often fail to reproduce subtle appearance effects arising from complex light-surface interactions, which have led to the emergence of specialized physics-based models for specific optical phenomena (e.g., diffraction, iridescence, multilayers). Although more accurate, these models lose versatility and lack performance for real-time rendering. Recently introduced, neural models have demonstrated their ability to approximate BRDF reference data coming from measurements, simulations, or even complex shading networks. However, most current neural models require relatively large networks, making them costly for real-time rendering. In this paper, we introduce a hybrid model that combines a GGX-type microfacet model and a neural model to leverage the best features of both representations. The neural component corrects the appearance approximated by the microfacet component, allowing much smaller network than in existing neural models. We show that, at identical memory cost, our model approximates measurements better than state-of-the-art neural models for a low evaluation overhead compared to a microfacet-based model. Furthermore, our hybrid model remains easily editable by artists and benefits from an important sampling scheme, making it attractive for both offline and real-time rendering.

arXiv

From Detection to Understanding: TAR and TAR-Bench for Multi-Task Traffic Anomaly Reasoning

Han Zhang, Yilin Zhao, Zaid Pervaiz Bhat, Zheng Tang, Varun Praveen, Vidya N. Murali, David C. Anastasiu, Tomasz Kornuta

We present TAR (Traffic Anomaly Reasoning) and TAR-Bench datasets, resources for training and evaluating video-language models beyond anomaly detection. TAR contains 44,040 chain-of-thought training annotations across 10 tasks for 3,670 CCTV videos ($\sim$26 hours) from eight public datasets. Its evaluation component, TAR-Bench, contains 960 human-curated test annotations for 80 held-out clips trimmed from 17 public YouTube videos. TAR's training annotations are produced with MAVEN, which consolidates multi-scale video evidence into structured event descriptions before generating question-answer pairs and reasoning traces. On TAR-Bench, eleven vision-language models reveal that strong question-answering accuracy does not reliably predict temporal or scene reasoning ability. Multi-task fine-tuning on TAR yields consistent gains, with the full 10-task model improving aggregate score by 21.4 points over its zero-shot baseline. TAR and TAR-Bench provide the official training and in-domain evaluation data for AI City Challenge 2026 Track 3. The dataset is available at https://huggingface.co/datasets/nvidia/PhysicalAI-Traffic-Anomaly-Reasoning

arXiv

UniMod: Enhancing Multi-Modal Medical Diagnosis through Cross-Modality and Within-Modality Alignment

Zijian Gu, Weikai Lin, Shuang Zhou, Zihan Chen, Song Wang

Multi-modal learning combining medical images and clinical text is promising for disease diagnosis. However, standard multi-modal training leads to shortcut learning: models exploit the easier modality (e.g., diagnostic cues in text) while neglecting harder-to-learn features (e.g., subtle visual patterns). We propose UniMod, a framework that mitigates shortcut learning by requiring each modality to predict the diagnosis on its own. It supervises image-only, text-only, and multi-modal classification simultaneously, so each modality must extract diagnostic features. We add cross-modality alignment for knowledge transfer and within-modality supervised contrastive alignment over same-diagnosis patients. On Harvard-Glaucoma, UniMod reaches 0.850 AUC, outperforming OGM-GE and Gradient Blending by 1.6-1.8%; on CheXpert Plus, it reaches 0.966 AUC, surpassing them by over 5%. UniMod also extends to 5-class multi-label diagnosis without architectural change, improving mean AUC by 0.097 over CGGM.

arXiv

Frozen Brain-MRI Foundation Models Are Site Fingerprints

Saman Rahbar

Frozen foundation-model (FM) embeddings are increasingly used as off-the-shelf brain-MRI representations, on the assumption that they capture anatomy. We audit what they actually encode and find that acquisition site is a large, intrinsic component of the representation. Across two independent cohorts (ABIDE-I, ABIDE-II), three frozen 3-D encoders (brain-pretrained, CT-pretrained, and randomly initialized), and every network depth, site is linearly decodable at roughly 0.9 balanced accuracy at deep layers, exceeding the decodability of every clinical or demographic variable (sex, age, autism diagnosis) at every layer. The effect is intrinsic rather than learned: a randomly initialized encoder is already a ~0.9 site classifier on both cohorts and across three architecture families (Swin, ViT, ResNet), and site is decodable at ~0.95 directly from the raw downsampled image with no encoder, so the fingerprint reflects low-level image statistics that any encoder preserves rather than a product of pretraining. Residualizing measured population covariates leaves site decodability essentially unchanged, indicating an acquisition- rather than population-driven effect. A nonlinear probe matches the linear one, so the fingerprint is fully linearly accessible. The site subspace is removable post hoc by iterative null-space projection or ComBat (site decodability 0.94 -> 0.07/0.00), and is a site-attribution concern for shared or federated embeddings; but for dense segmentation this removal is not free, because site and anatomy occupy an entangled linear subspace (a matched-rank random-direction projection is Dice-neutral, whereas removing the site subspace is destructive). We recommend site-audited use of frozen brain-MRI FMs and release an open audit toolkit.

Keep browsing

More papers from top AI conferences

30+ top conferences, by conference, year, and topic.