论文检索

Deep Learning · Large Language Models

Jian Chen, Yesheng Liang, Zhijian Liu

Autoregressive large language models (LLMs) deliver strong performance but require inherently sequential decoding, leading to high inference latency and poor GPU utilization. Speculative decoding mitigates this bottleneck by using a fast draft model whose outputs are verified in parallel by the target LLM. However, existing methods still rely on *autoregressive drafting*, which remains sequential and constrains practical speedups. Diffusion LLMs offer a promising alternative by enabling parallel generation, but current diffusion models typically underperform compared with autoregressive models. In this paper, we introduce **DFlash**, a speculative decoding framework that employs a lightweight block diffusion model for parallel drafting. We show that speculative decoding provides a natural and effective setting for diffusion models. By generating draft tokens in a single forward pass, DFlash enables efficient drafting, and by conditioning the draft model on context features extracted from the target model, it achieves high-quality drafts with improved acceptance rates. Experiments demonstrate that DFlash achieves more than 6$\times$ lossless acceleration across a range of models and tasks, delivering up to 2.5$\times$ higher speedup than the state-of-the-art speculative decoding method EAGLE-3.

Deep Learning · Large Language Models

Chanuk Lee, Minki Kang, Sung Ju Hwang

Recent studies observe that reinforcement learning with verifiable rewards (RLVR) reliably improves pass@1 on reasoning tasks, yet often fails to yield comparable gains in pass@k, raising the question of whether RLVR genuinely enables large language models to acquire novel reasoning abilities or merely enhances the efficiency of sampling reasoning modes already present in the base model. Prior analyses largely support the latter view, attributing this limitation to structural properties of standard RLVR objectives that result in insufficient exploration pressure. In this work, we argue that a central structural constraint arises from reverse-KL regularization, which stabilizes training but inherently anchors the policy to the reference distribution, thereby suppressing the emergence of alternative reasoning modes. However, we show that neither removing the KL term nor replacing it with forward-KL provides a satisfactory solution, as both disrupt the efficiency–coverage trade-off by either inducing reward hacking or allocating probability mass to off-target regions. To resolve this tension, we propose SAGE, a principled framework that enables controllable empirical support expansion by reshaping the reverse-KL anchor distribution itself through a guide function $q(x,y)$, achieving consistent improvements in both pass@1 and pass@k across challenging mathematical reasoning benchmarks.

Reinforcement Learning · Online

Parnian Behdin, Kevin Roice, Golnaz Mesbahi

Reinforcement Learning (RL) has received increasing attention and adoption in real-world use cases. Most of these systems follow a train-then-fix paradigm, where trained agents do not learn while interacting with the world, until performance degrades and retraining becomes necessary. In this position paper, we argue that deploying an agent that is incapable of optimality, but receives an evaluative reward signal, is inherently a continual RL problem. We identify four sources of non-stationarity after deployment that necessitate never-ending learning, and highlight why the best deployed agents never stop adapting. We analyze successful examples of continual RL in the real world, and present the community with the advantages and measures to move away from the current train-then-fix paradigm.

Deep Learning · Generative Models and Autoencoders

Qianxin Xia, Zhiyong Shu, Wenbo Jiang, Jiawei Du, Jielei Wang, Guoming Lu

Dataset distillation aims to synthesize a compact proxy dataset that is unreadable or non-raw from the original dataset for privacy protection and highly efficient learning. However, previous approaches typically adopt a single-stage distillation paradigm, which suffers from learning specific patterns that overfit on a prior architecture, consequently suppressing the expression of semantics and leading to performance degradation across heterogeneous architectures. To address this issue, we propose a novel dual-stage distillation framework called ${\textbf{DIVER}}$, which leverages the pre-trained diffusion model to dive deeper into $\textbf{DI}$stilled data $\textbf{V}$ia $\textbf{E}$xpressive semantic $\textbf{R}$ecovery, a process of semantic inheritance, guidance, and fusion. Semantic inheritance distills high-level semantic knowledge of abstract distilled images into the latent space to filter out architecture-specific ``noise" and retain the intrinsic semantics. Furthermore, semantic guidance improves the preservation of the original semantics by directing the reverse procedure. Ultimately, \textcolor{red}{semantic fusion is designed to provide semantic guidance only during the concrete phase of the reverse process, preventing semantic ambiguity and artifacts while maintaining the guidance information.} Extensive experiments validate the effectiveness and efficiency of our method in improving classical distillation techniques and significantly improving cross-architecture generalization, requiring processing time comparable to raw DiT on ImageNet (256$\times$256) with only 4 GB of GPU memory usage.

Applications · Chemistry, Physics, and Earth Sciences

Yeqiu Chen, Ziyan Liu, Hong Wang, Lei Liu

Solving large-scale Generalized Eigenvalue Problems (GEPs) is a fundamental yet computationally prohibitive task in science and engineering. As a promising direction, contour integral (CI) methods offer an efficient and parallelizable framework. However, their performance is critically dependent on the selection of \textit{integration contours}---improper selection without reliable prior knowledge of eigenvalue distribution can incur significant computational overhead and compromise numerical accuracy. To address this challenge, we propose **Deepcontour**, a novel hybrid framework that integrates a deep learning-based spectral predictor with Kernel Density Estimation (KDE) for principled contour design. Specifically, Deepcontour utilizes its specialized Eigen-Neural-Operator (ENO) to provide rapid spectral distribution priors, driving a KDE module to automatically construct the optimized integration contours, which guide the CI solver to efficiently find the desired eigenvalues. Deepcontour achieves up to a 5.63x speedup across diverse scientific datasets while maintaining strict numerical rigor. By merging the predictive power of deep learning with the numerical rigor of classical solvers, this work establishes an efficient and robust paradigm for solving large-scale GEPs.

Yichong Xia, Yimin Zhou, Zongyu Li, Shiyu Qin, Mingyao Hong, Bin Chen, Haoqian Wang

Stereo vision is widely utilized in automotive imagery and 3D reconstruction, creating a demand for compressing stereo images. Existing methods for stereo image compression often employ VAE-like architectures based on distortion optimization, leading to subpar perceptual quality at low bitrates. While generative compression excels in high perceptual fidelity at low bitrates, it struggles to maintain consistency across viewpoints, making decoded images less useful for critical downstream tasks. To address this, we introduce SDiD, a distributed stereo image compression architecture based on shared pre-trained diffusion priors. We employ a diffusion prior alignment module to efficiently obtain the main-view-prior from the foundation diffusion, and utilize a prior transformation structure to enable the auxiliary view to achieve reliable and fast perceptual enhancement while maintaining consistency. Through extensive experiments, we demonstrate that SDiD outperforms existing methods in perceptual quality across multiple datasets. Even at extremely low bitrates, SDiD can accurately recover depth information between decoded images. On the InStereo2K dataset, SDiD requires only one-third of the bits compared to the state-of-the-art baseline (0.02 bpp vs. 0.06 bpp) to reconstruct image pairs with similar depth information.

Applications · Health / Medicine

Weiyi Wu, Xinwen Xu, Xingjian Diao, Siting Li, Zhi Wei, Alma Andersson, Jiang Gui

Spatial transcriptomics (ST) links gene expression with tissue morphology but remains expensive and low-throughput, motivating surrogates that infer expression from routine histology. Whole-slide H&E-to-ST inference pairs a gigapixel image with gene measurements at a sparse, irregular set of locations, making multiscale modeling challenging without incurring dense-grid overhead or quadratic token mixing. We propose HiST, a hierarchical sparse transformer that treats measured locations as a lattice-indexed sparse field and builds a dyadic encoder--decoder directly on the active tissue footprint. HiST combines sparse window attention for local geometric correspondence with resolution-changing operators for rapid multiscale context integration. For a fixed window size, the dominant runtime and memory scale with the number of observed locations rather than the dense slide area. To mitigate slide-specific acquisition variation, HiST adds a bottlenecked global conditioning pathway via a \emph{slide calibration token} that summarizes slide-level context and conditions local representations. On a multi-organ benchmark spanning diverse tissues and acquisition sources, HiST improves predictive performance over recent baselines while reducing runtime and peak memory.

Social Aspects · Everything Else

Pedram Bakhtiarifard, Pınar Tözün, Christian Igel, Raghavendra Selvan

Sustainability encompasses three key facets: economic, environmental, and social. However, the nascent discourse that is emerging on sustainable artificial intelligence (AI) has predominantly focused on the environmental sustainability of AI, often neglecting the economic and social aspects. Achieving truly sustainable AI necessitates addressing the tension between its climate awareness, which emphasizes the need to mitigate AI's environmental impacts, and its social sustainability, which hinges on equitable access to AI development resources. The concept of resource awareness advocates for AI sovereignty through broader access to the infrastructure required to develop AI. Yet, this push for improving accessibility often overlooks the environmental costs of expanding such resource usage. This position paper argues that reconciling climate awareness and resource awareness is essential to realizing sustainable AI and neglecting these factors fuelling the global AI arms race. By applying the base-superstructure framework from historical materialism, we analyze how the material conditions are shaping the current AI progress and the discourse surrounding it. We also introduce the Climate and Resource Aware Machine Learning (CARAML) framework to address the conflict between climate and resource awareness of AI, with actionable recommendations spanning individual, community, industry, government, and global levels to achieve sustainable AI.

Applications · Language, Speech and Dialog

Yash Saxena, Ankur Padia, Mandar Chaudhary, Kalpa Gunaratna, Srinivasan Parthasarathy, Manas Gaur

Retrieval-Augmented Generation (RAG) systems deployed in sensitive domains must provide interpretable evidence selection and robust safeguards against data poisoning, yet current approaches rely on opaque similarity-based retrieval with arbitrary top-k cutoffs that offer no explanation for their selections and remain vulnerable to adversarial manipulation. We propose METEORA, a rationale-driven RAG framework that addresses these fundamental limitations through interpretable, adaptive evidence retrieval. Our framework introduces three synergistic contributions. First, we preference-tune a general-purpose LLM to generate explicit rationales that articulate why specific evidence is needed for a given query. These rationales then guide adaptive evidence selection through a two-step process: rationale-chunk pairing for query-specific relevance assessment, followed by dynamic cutoff detection that eliminates the need for arbitrary k heuristics. Finally, the same rationales enable a verification stage that filters poisoned or misleading evidence before generation. Evaluation across six datasets demonstrates substantial improvements on three critical dimensions. For retrieval quality, METEORA achieves **21.05\%** higher precision than the best-performing baseline, while its variant with context expansion achieves **13.41\%** higher recall. In terms of efficiency, the framework reduces the volume of evidence required to reach comparable recall by **80\%**, which directly translates to a **33.34\%** improvement in downstream answer generation accuracy. Most notably for adversarial robustness, METEORA increases the F1 score from **0.10 to 0.44** under poisoning attacks, a 4.4$\times$ improvement that makes RAG systems substantially more resilient to adversarial manipulation. Human evaluation with four experienced annotators confirms genuine interpretability, achieving a mean confidence score of **3.64/5** and demonstrating that humans can reliably reconstruct evidence-level decisions with **86\% accuracy**. These results demonstrate that rationale-driven retrieval can simultaneously enhance interpretability, efficiency, and safety in RAG systems for sensitive domains. The code is available in the anonymous GitHub repository \url{https://anonymous.4open.science/r/METEORA-DC46/README.md}

Theory · Online Learning and Bandits

Kanad Pardeshi, Samsara Foubert, Aarti Singh

In many real-world settings, a centralized decision-maker must repeatedly allocate finite resources to a population over multiple time steps. Individuals who receive a resource derive some stochastic utility; to characterize the population-level effects of an allocation, the expected individual utilities are then aggregated using a social welfare function (SWF). We formalize this setting and present a general confidence sequence framework for SWF-based online learning and inference, valid for any monotonic, concave, and Lipschitz-continuous SWF. Our key insight is that monotonicity alone suffices to lift confidence sequences from individual utilities to anytime-valid bounds on optimal welfare. Building on this foundation, we propose SWF-UCB, a SWF-agnostic online learning algorithm that achieves near-optimal $\tilde{\mathcal{O}}(n+\sqrt{nkT})$ regret (for $k$ resources distributed among $n$ individuals at each of $T$ time steps). We instantiate our framework on three normatively distinct SWF families: Weighted Power Mean, Kolm, and Gini, providing bespoke oracle algorithms for each. Experiments confirm $\sqrt{T}$ scaling and reveal rich interactions between $k$ and SWF parameters. This framework naturally supports inference applications such as sequential hypothesis testing, optimal stopping, and policy evaluation.

Deep Learning · Other Representation Learning

Chen Huang, Xianhang Li, Vimal Thilak, Etai Littwin, Joshua M Susskind

Image-based Joint-Embedding Predictive Architecture (I-JEPA) offers a promising approach to visual self-supervised learning through masked feature prediction. However with the inherent visual uncertainty at masked positions, feature prediction remains challenging and may fail to learn semantic representations. In this work, we propose Text-Conditional JEPA (TC-JEPA) that uses image captions to reduce the prediction uncertainty. Specifically, we modulate the predicted patch features using a self-supervised, image-grounded text conditioner that computes sparse cross-attention over input text tokens. With such conditioning, patch features become predictable as a function of text, thus are more semantically meaningful. We show TC-JEPA improves downstream performance and training stability, with promising scaling properties. TC-JEPA also offers a new vision-language pretraining paradigm based on feature prediction only, outperforming contrastive methods on diverse tasks, especially those requiring fine-grained visual understanding and reasoning.

Applications · Computer Vision

Weiyan Chen, Weijian Deng, Yao Xiao, Weijie Tu, ZiYi Dong, Ibrahim Radwan, Liang Lin, Pengxu Wei

Preference alignment aims to guide generative models by learning from comparisons between preferred and non-preferred samples. In practice, most existing approaches rely on preference pairs constructed from model-generated images. Such supervision is inherently relative and can be ambiguous when both samples exhibit artifacts or limited visual quality, making it difficult to infer what constitutes a truly desirable output. In this work, we investigate whether real data can serve as an alternative source of supervision for preference alignment. We adopt a data-centric perspective and study a curation strategy that treats real images as reference points and constructs preference signals by contrasting them with generated or perturbed samples, without requiring manually annotated preference pairs. Through empirical analysis, we show that real-data-based supervision provides effective guidance for aligning diffusion models and achieves performance comparable to existing preference-based methods. Our results suggest that real data offers a practical and complementary source of supervision for preference alignment and highlight directions of label-efficient alignment strategies.

Applications · Computer Vision

Zijie Cao, Weijie Tu, Yao Xiao, Weijian Deng, Weiyan Chen, Liang Lin, Pengxu Wei

Detecting AI-generated images (AIGI) remains challenging because detectors often fail to generalize to unseen generators. Although existing methods are trained on large datasets, their performance still degrades when generation settings change, indicating that data scale alone is insufficient and that limited coverage of generative variations during training is a key factor. Studies on generative model editing show that small changes in internal representations can produce diverse and meaningful image variations, many of which are not explored under standard sampling. Leveraging this insight, we propose PROBE (Probing Robustness via Boundary Exploration), a framework that improves detector generalization by actively exploring challenging regions of the generative process. Instead of treating the generator as a fixed data source, PROBE uses the detector as a critic to steer the generator through manifold-level modifications, producing realistic samples that are difficult to classify. These samples expose failure cases that are uncommon under standard data sampling strategies and are used to refine the detector. Experimental results across multiple benchmarks indicate that PROBE enhances generalization to unseen generators, resulting in more generalizable AIGI detection performance.

Social Aspects · Fairness

Samuel Gruffaz, Gabriel Singer, Olivier VO VAN, Nicolas Vayatis, Argyris Kalogeratos

In many machine learning applications acquiring reliable ground-truth labels is costly, or unfeasible, leading practitioners to rely on crowdsourcing and aggregation of noisy human annotations. When labels are subjective, however, aggregation may amplify individual biases, particularly with respect to sensitive attributes, raising fairness concerns. Despite this, fairness in crowdsourced aggregation remains largely unexplored, with no existing convergence guarantees and only limited post-processing approaches for enforcing $\varepsilon$-fairness under demographic parity. We address this gap by analyzing fairness properties of crowdsourcing aggregation methods within the $\varepsilon$-fairness framework, focusing on Majority Voting and Optimal Bayesian aggregation. In the small-crowd regime, we derive an upper bound on the fairness gap of Majority Voting in terms of the individual annotators’ fairness gaps. We further show that the fairness gap of the aggregated consensus converges exponentially fast to that of the ground truth under interpretable conditions. Since the ground truth itself may still be unfair, we generalize a state-of-the-art multiclass fairness post-processing algorithm from the continuous to the discrete setting, enabling the enforcement of strict demographic parity constraints for any aggregation rule. Experiments on both synthetic and real-world datasets demonstrate the effectiveness of our approach and corroborate the theoretical insights.

Yijun Pan, Taiwei Shi, Jieyu Zhao, Jiaqi Ma

Large language models (LLMs) are highly sensitive to even small amounts of unsafe training data, making effective detection and filtering essential for trustworthy model development. Current state-of-the-art (SOTA) detection approaches primarily rely on moderation classifiers, which require significant computation overhead for training and are limited to predefined taxonomies. In this work, we explore data attribution approaches that measure the similarity between individual training samples and a small set of unsafe target examples, based on data representations such as hidden states or gradients. We identify a key limitation in existing methods: unsafe target texts contain both critical tokens that make them unsafe and neutral tokens (e.g., stop words or benign facts) that are necessary to form fluent language, and the latter of which makes the overall representations noisy for the purpose of detecting unsafe training data. To address this challenge, we propose Denoised Representation Attribution (DRA), a novel representation-based data attribution approach that denoises training and target representations for unsafe data detection. Across tasks of filtering jailbreaks and detecting gender bias, the proposed approach leads to significant improvement for data attribution methods, outperforming SOTA methods that are mostly based on moderation classifiers.

Applications · Computer Vision

Jun Li, Peifeng Lai, Xuhang Lou, Jinpeng Wang, Yuting Wang, Ke Chen, Yaowei Wang, Shutao Xia

Partially relevant video retrieval aims to retrieve untrimmed videos using text queries that describe only partial content. However, the inherent asymmetry between brief queries and rich video content inevitably introduces uncertainty into the retrieval process. In this setting, vague queries often induce semantic ambiguity across videos, a challenge that is further exacerbated by the sparse temporal supervision within videos, which fails to provide sufficient matching evidence. To address this, we propose Holmes, a hierarchical evidential learning framework that aggregates multi-granular cross-modal evidence to quantify and model uncertainty explicitly. At the inter-video level, similarity scores are interpreted as evidential support and modeled via a Dirichlet distribution. Based on the proposed three-fold principle, we perform fine-grained query identification, which then guides query-adaptive calibrated learning. At the intra-video level, to accumulate denser evidence, we formulate a soft query-clip alignment via flexible optimal transport with an adaptive dustbin, which alleviates sparse temporal supervision while suppressing spurious local responses. Extensive experiments demonstrate that Holmes outperforms state-of-the-art methods. Code is released at https://anonymous.4open.science/r/ICML26-Holmes-E77B/

Deep Learning · Large Language Models

Bharat Runwal, Ashish Agrawal, Anurag Roy, Rameswar Panda

Mid-training is increasingly used to improve the reasoning capabilities of large language models (LLMs), yet its design choices and interaction with evaluation and reinforcement learning (RL) remain poorly understood. Prior work often focuses on narrow domain gains, overlooking retention of general abilities, long-context performance, and RL compatibility. We present $\textbf{PRISM}$ (Demystifying Retention and Interaction in Mid-Training), a holistic empirical study that analyzes mid-training design choices, what to evaluate, and how domain mixtures and training stages interact across model families. Experiments on Granite-3.3 8B, LLaMA-3.1 8B, and Mistral-7B/24B base models show that a relatively small, high-quality mid-training phase of $\textbf{$\sim$27B}$ tokens acts as a critical stabilizing stage for reasoning. Across models, PRISM yields consistent gains of $\textbf{$\sim$6–10}$ points on coding benchmarks and $\textbf{$\sim$17–30}$ points on mathematical reasoning benchmarks while preserving general performance. RL applied on top of PRISM-mid-trained models produces stable, monotonic improvements, adding a further $\textbf{$\sim$3–8}$ points across coding and math tasks such as LiveCodeBench, Codeforces, AIME and MATH500, and $\textbf{$\sim$17–20}$ points on science (GPQA-Diamond), whereas RL applied directly to base models is substantially less effective. Our results demonstrate that retention-aware mid-training is a necessary intermediate step for reliable reasoning enhancement and RL scaling, and provide practical guidance for designing robust mid-training pipelines for modern LLMs.

Deep Learning · Generative Models and Autoencoders

Andrei Atanov, Jesse Allardice, Roman Bachmann, Oğuzhan Fatih Kar, R Devon Hjelm, David Griffiths, Peter Fu, Amir Zamir, Afshin Dehghan

Visual tokenizers map high-dimensional raw pixels into a compressed representation for downstream modeling, e.g., conditional video generation. Beyond compression, tokenizers define what information is preserved and how it is organized. A _de facto_ standard approach is to represent a video with a spatiotemporal 3D grid of tokens, each corresponding to a local patch in the original signal. This requires a downstream model, e.g., a text-to-video model, to learn to predict all low-level details ``pixel-by-pixel'' irrespective of the video's inherent complexity, resulting in high computational cost during training. We present VideoFlexTok, a tokenizer that represents videos with a _variable-length sequence of tokens structured in a coarse-to-fine manner_, where the first tokens capture abstract information like semantics and motion and later tokens provide fine-grained details. The generative flow decoder enables realistic video reconstructions from any token count. This representation structure allows adapting the tokens count to particular downstream needs and encode videos longer than the 3D grid approach under the same budget. We evaluate VideoFlexTok on class-to-video and text-to-video generative tasks and show that it leads to more efficient training compared to 3D grid tokens, e.g., achieving comparable generation quality (gFVD and ViCLIP Score) with a 10x smaller model (0.4B vs 3.6B). Finally, we demonstrate how VideoFlexTok can enable long video generation without prohibitive computational cost by training a text-to-video model on 10-second 81-frame videos with only 672 tokens, 8x fewer than a comparable 3D grid tokenizer.

Deep Learning · Foundation Models

Xiangyu Liu, Di Wang, Zhe Feng, Aranyak Mehta

While large language models (LLMs) have emerged as powerful decision-makers across a wide range of single-agent and stationary environments, fewer efforts have been devoted to settings where LLMs must engage in \emph{repeated} and \emph{strategic} interactions with unknown or dynamic opponents. In such settings, recipes built upon \emph{offline} pre-training or fine-tuning, though robust against worst-case adversaries, do not fully exploit the capability of LLMs to adapt \emph{online} based on interaction feedback. Instead, we explore the more natural perspective of \emph{scaling inference-time computation} as a mechanism for adaptation, embedding the principles of a classical game-theoretical learning dynamic, \emph{smooth Fictitious Play (sFP)}, into LLM inference: (i) for belief formation, we employ an auxiliary opponent model that in-context learns to imitate the \emph{time-averaged} behavior of the opponent; (ii) for best response, we advance best-of-$N$ (BoN) sampling by simulating against the opponent model. Empirical evaluations on two distinct forms of repeated negotiation games demonstrate that our method enables significant performance improvement over online interaction compared to various baselines, offering a scalable and principled approach to strategic decision-making without any parameter updates.

General Machine Learning · Data

Charlotte Högberg, Ericka Johnson, Kiri Wagstaff

Ground truth datasets play a fundamental role as reference values in the training and evaluation of machine learning models. This position paper argues that ground truths are not neutral objective measurements that are naturally given, but instead that they are constructed by arrangements of humans and technologies. We argue that the ML community will benefit by articulating and discussing these often invisible or unreported choices and by acknowledging that reference data sets are contingent, not universal. Focusing on the situated and context-dependent nature of ground truths can improve reliability by enabling a better informed perspective on where, when, and how the datasets, and the models they have shaped, can best be used. We argue for increasing `situated reliability' which includes articulating the limits and strengths of models and their truth claims. Finally, paying more attention to the construction of ground truths can help achieve transparency and accountability and support interdisciplinary work in ML development.