论文检索

Deep Learning · Generative Models and Autoencoders

Luca Saglietti, Luca Biggio, Jerome Garnier-Brun, Davide Beltrame, Marc Mezard

Generalization in generative modelling is defined as the ability to learn an underlying distribution from a finite dataset and produce novel samples, with evaluation largely driven by held-out performance and perceived sample quality. In practice, training is often stopped at the minimum of the test loss, taken as an operational indicator of generalization. We challenge this viewpoint by identifying a phase of \emph{biased generalization} during training, in which the model continues to decrease the test loss while favoring samples with anomalously high proximity to training data. By training the same network on two disjoint datasets and comparing the mutual distances of generated samples and their similarity to training data, we introduce a quantitative measure of bias and demonstrate its presence on real images. We then study the mechanism of bias, using a controlled hierarchical data model where access to exact scores and ground-truth statistics allows us to precisely characterize its onset. We attribute this phenomenon to the sequential nature of feature learning in deep networks, where coarse structure is learned early in a data-independent manner, while finer features are resolved later in a way that increasingly depends on individual training samples. Our results show that early stopping at the test loss minimum, while optimal under standard generalization criteria, may be insufficient for privacy-critical applications.

Applications · Chemistry, Physics, and Earth Sciences

Minjun Park, Joseph Stitt, Robert Clapp, Ilan Naiman, Artem Goncharuk, Kevin Smith

We introduce SeisMark, a large-scale open benchmark designed to bridge the gap between verifiable ground truth and realistic texture in 3D seismic fault detection. Using a novel pipeline merging procedural geology with diffusion-based synthesis, we produce domain-realistic (survey-specific) textured volumes that expose significant brittleness in existing models masked by simplified physics data. Experiments demonstrate that SeisMark acts as a rigorous discriminator, distinguishing robust modern architecture from legacy model that suffers performance collapse under realistic domain shifts. We release this benchmark to the community to serve as a verifiable standard for developing trustworthy, deployment-ready AI for safety-critical subsurface applications.

Deep Learning · Foundation Models

Yiming Zhong, Yaoyu He, Zemin Yang, Pengfei Tian, Yifan Huang, Qingqiu Huang, Xinge Zhu, Yuexin Ma

Bridging high-level semantic understanding with low-level physical control remains a persistent challenge in embodied intelligence, stemming from the fundamental spatiotemporal scale mismatch between cognition and action. Existing generative policies typically adopt a "Generation-from-Noise" paradigm, which disregards this disparity, leading to representation inefficiency and optimization "Loss Collapse". In this work, we propose ResVLA, a novel architecture that shifts the paradigm to "Refinement-from-Intent". Recognizing that robotic motion naturally decomposes into global intent and local dynamics, ResVLA utilizes spectral analysis to decouple control into a deterministic low-frequency anchor and a stochastic high-frequency residual. By anchoring the generative process on the predicted intent, our model focuses strictly on refining local dynamics via a residual diffusion bridge. Extensive evaluations on LIBERO and the challenging LIBERO-Plus benchmarks demonstrate that ResVLA achieves state-of-the-art performance. Notably, our approach exhibits exceptional robustness against semantic drift and kinematic perturbations while achieving significantly faster convergence than standard generative baselines.

Deep Learning · Large Language Models

Romeo Valentin, Sydney Katz, Vincent Vanhoucke, Mykel Kochenderfer

Dictionary learning has recently emerged as a promising approach for mechanistic interpretability of large transformer models. Disentangling high-dimensional transformer embeddings requires algorithms that scale to high-dimensional data with large sample sizes. Recent work has explored sparse autoencoders (SAEs) for this problem. However, SAEs use a simple linear encoder to solve the sparse encoding subproblem, which is known to be NP-hard. It is therefore interesting to understand whether this approach is sufficient to find good solutions to the dictionary learning problem or if a more sophisticated algorithm could find better solutions. In this work, we propose Double-Batch KSVD (DB-KSVD), a scalable dictionary learning algorithm that adapts the classic KSVD algorithm. DB-KSVD is informed by the rich theoretical foundations of KSVD but scales to datasets with millions of samples and thousands of dimensions. We demonstrate the efficacy of DB-KSVD by disentangling text embeddings of the Gemma-2-2B and Pythia-160M models and evaluating on six metrics from the SAEBench benchmark, where we achieve competitive results when compared to established approaches based on SAEs. We further show similar results when disentangling image embeddings obtained from the DINOv2-S and DINOv2-B models, solidifying our findings. By matching SAE performance with an entirely different optimization approach, our results suggest that (i) SAEs do find strong solutions to the dictionary learning problem and (ii) traditional optimization approaches can be scaled to the required problem sizes, offering a promising avenue for further research. We make an implementation of DB-KSVD available.

Deep Learning · Generative Models and Autoencoders

Jerry Huang, Justin Lin, Sheel Shah, Kartik Nair, Nicholas Boffi

In generative modeling, we often wish to produce samples that satisfy a user-specified reward such as measurement consistency, aesthetic quality, or alignment with human intent, a problem known as inference-time guidance. While flow-based models enable high-quality generation, existing guidance methods either require expensive multi-particle, many-step schemes to sample from a reward-tilted distribution or rely on heuristic approximations. To design efficient algorithms, we instead reformulate guidance as a deterministic optimal control problem, rather than the stochastic control problem from which the reward tilt emerges. We find that the flow map arises naturally in the optimal solution, and that many existing flow-based guidance methods are best understood as coarse approximations that replace it with a single Euler step. Rather than relying on these approximations, we propose Flow Map Trajectory Guidance (FMTG): a principled framework that uses the flow map to both integrate and guide, enabling training-free alignment in just a few network evaluations. We demonstrate FMTG at text-to-image scale using a FLUX.1-distilled flow map, showing that it achieves comparable performance to baselines across inverse problems and image editing tasks with up to 10 times fewer function evaluations.

Applications · Health / Medicine

Yinhua Piao, Hyomin Kim, SEONGHWAN KIM, Yunhak Oh, Junhyeok Jeon, Sangyeon Hwang, Jaechang Lim, Woo Youn Kim, Chanyoung Park, Sungsoo Ahn

Predicting high-dimensional transcriptional responses to genetic perturbations is challenging due to severe experimental noise and sparse gene-level effects. Existing methods often suffer from mean collapse, where high correlation is achieved by predicting global average expression rather than perturbation-specific responses, leading to many false positives and limited biological interpretability. Recent approaches incorporate biological knowledge graphs into perturbation models, but these graphs are typically treated as dense and static, which can propagate noise and obscure true perturbation signals. We propose AdaPert, a perturbation-conditioned framework that addresses mean collapse by explicitly modeling sparsity and biological structure. AdaPert learns perturbation-specific subgraphs from biological knowledge graphs and applies adaptive learning to separate true signals from noise. Across multiple genetic perturbation benchmarks, AdaPert consistently outperforms existing baselines and achieves substantial improvements on DEG-aware evaluation metrics, indicating more accurate recovery of perturbation-specific transcriptional changes.

Deep Learning · Large Language Models

Kanghoon Yoon, Minsub Kim, Sungjae Lee, Joonhyung Lee, Sunghyeon Woo, Yeonjun In, Se Jung Kwon, Chanyoung Park, Dongsoo Lee

Speculative decoding accelerates LLM inference by verifying candidate tokens from a draft model against a larger target model. Recent "judge'' decoding boosts this process by relaxing verification criteria by accepting draft tokens that may exhibit minor discrepancies from target model output, but existing methods are restricted by their reliance on human annotations or tasks with verifiable ground truths, limiting generalizability across diverse NLP tasks. We propose SelfJudge, which trains judge verifiers via self-supervision of the target model. Our method measures semantic preservation by assessing whether token-substituted responses preserve the meaning of original responses, enabling automatic verifier training across diverse NLP tasks. Our experiments show SelfJudge achieves superior inference-accuracy trade-offs than judge decoding baselines, offering a broadly applicable solution for faster LLM inference.

Reinforcement Learning · Deep RL

Ali Saheb pasand, Johan Obando-Ceron, Aaron Courville, Pouya Bashivan, Pablo Samuel Castro

Deep reinforcement learning systems often suffer from unstable training dynamics due to non-stationarity, where learning objectives and data distributions evolve over time. We show that under non-stationary targets, isotropic Gaussian embeddings are provably advantageous. In particular, they induce stable tracking of time-varying targets for linear readouts, achieve maximal entropy under a fixed variance budget, and encourage a balanced use of all representational dimensions. Building on this insight, we propose the use of Sketched Isotropic Gaussian Regularization for shaping representations toward an isotropic Gaussian during training. We demonstrate empirically, over a variety of domains, that this simple and computationally inexpensive method improves performance under non-stationarity while reducing representation collapse, neuron dormancy, and training instability.

Reinforcement Learning · Deep RL

Muxi Tao, jiangtao wen, Yuxing Han

Model-based reinforcement learning (MBRL) improves sample efficiency by leveraging learned dynamics models, but often suffers from unstable training due to dynamics model learning mismatch: models are trained on data from historical policies while being queried under the continually updated current policy. This mismatch can cause policy-relevant local model error to remain large even as global prediction error decreases, leading to oscillatory updates. We present a finite-horizon performance analysis that decomposes the policy performance gap into global model error, policy-induced distribution shift, and historical policy mixture effects, showing that minimizing global error alone is insufficient for stable optimization. Motivated by this analysis, we propose Prioritized Model Experience Replay (PMER), a lightweight replay mechanism that prioritizes high-error transitions during dynamics model training. PMER implicitly emphasizes policy-relevant regions without explicit policy distance estimation and integrates seamlessly into Dyna-style MBRL frameworks. Experiments on MuJoCo benchmarks demonstrate improved stability, faster convergence, and higher sample efficiency.

Applications · Chemistry, Physics, and Earth Sciences

Sungwon Kim, Juho Song, Seungmin Shin, Guimok Cho, Sangkook Kim, Chanyoung Park

Deep learning surrogates for 3D Partial Differential Equations (PDEs) often fail to generalize across geometric transformations because they depend heavily on specific coordinate systems. While equivariant networks offer a solution, they typically rely on local operations in the spatial domain, making the global receptive field—essential for PDE dynamics—computationally expensive. Conversely, Fourier Neural Operators (FNOs) efficiently capture global interactions, yet establishing 3D equivariance within them remains impractical due to the prohibitive cost of spectral group convolutions. To bridge this gap, we introduce EqGINO, a geometrically robust framework that enforces isotropy in the spectral domain. By design, EqGINO guarantees exact equivariance to the discrete symmetries inherent to the discretized computational domain. Beyond this discrete guarantee, our structural prior enables effective generalization to arbitrary continuous orientations even with a limited number of SE(3)-transformed training samples. Consequently, our method robustly models coordinate-invariant physical laws on complex irregular 3D geometries. Our code is available at https://anonymous.4open.science/r/EqGINO/

Applications · Computer Vision

Seojeong Park, Jiho Choi, Junyong Kang, Seonho Lee, Jaeyo Shin, Hyunjung Shim

Recent multimodal large language models have demonstrated strong reasoning ability, yet their reliability as automated evaluators remains limited by a critical weakness: when visual evidence conflicts with textual cues, MLLM judges tend to reward plausible narratives over perceptually correct answers. We identify and systematically analyze this phenomenon, which we term \textit{Perceptual Judgment Bias}. Through controlled visual perturbations, existing multimodal judges frequently anchor on the response text instead of their own visual perception, leading to inconsistent and non-verifiable evaluations. To address this issue, we introduce the \textit{Perceptually Perturbed Judgment Dataset}, which constructs minimally edited counterfactual responses that isolate perceptual errors and enable verifiable supervision. Building on this dataset, we develop a unified training framework that combines a structured GRPO-based reward with a batch-ranking objective, achieving coherent global ordering without explicit pairwise labels. Experiments across diverse MLLM-as-a-Judge benchmarks show that our approach substantially improves perceptual fidelity, ranking coherence, and alignment with human evaluation. Our results establish a scalable and generalizable pathway for training multimodal judges that are perceptually grounded, interpretable, and robust to visual–reasoning conflicts.

General Machine Learning · Evaluation

Wonjoong Kim, Sangwu Park, Yeonjun In, Sein Kim, Dongha Lee, Chanyoung Park

Driven by recent advancements in tool-augmented Large Language Model (LLM) agents, comprehensive benchmark datasets for evaluating these tool-augmented agents are being actively developed. Although these benchmarks incorporate increasingly complex user requests and a diverse array of tools, the evaluation methods for most of them remain limited to answer matching. However, as the number of steps required to resolve a user request increases, a proper evaluation of an agent's performance must go beyond the final answer to also assess the problem-solving trajectory, including previously ignored aspects such as efficiency, hallucinations, and adaptivity. The most straightforward method for evaluating these aspects is to compare the trajectory of the agent with a ground-truth trajectory, but this approach is fundamentally limited since annotating all possible ground-truth trajectories is prohibitively expensive. To address these significant gaps, we introduce TRACE, a framework for the multi-dimensional evaluation of tool-augmented LLM agent performance. By incorporating evidence store, TRACE enables a multi-faceted analysis and evaluation of an agent's reasoning trajectory, eliminating the need for a predefined ground-truth trajectory. To validate our framework, we develop a new meta-evaluation dataset by augmenting existing benchmarks with diverse and flawed trajectories, each labeled with multi-faceted performance scores. Our results confirm that TRACE accurately evaluates these complex behaviors in a scalable and cost-effective manner, even with small open-source LLMs.

Reinforcement Learning · Deep RL

Prabhat Nagarajan, Brett Daley, Martha White, Marlos C. Machado

Learning action-values efficiently is central to reinforcement learning (RL), as they underpin many control algorithms such as Q-learning. However, action-value learning can be slow, requiring many updates to move values from their initialization, typically near zero, to their true values, which may be far from zero. Moreover, action-value learning algorithms typically update each state–action pair independently, without learning shared value structure across actions within a state. In this paper, we address these inefficiencies by introducing the mean-expansion transformation, which accelerates action-value learning by sharing values across actions within a state and by changing the problem from directly learning potentially large action-values to learning a lower-norm representation of them. In deep RL, this transformation can be applied as a parameter-free modification to Q-network architectures without altering the underlying algorithm. Empirically, we show that it improves DQN's performance in aggregate across 57 Atari games while increasing action gaps and dramatically reducing value overestimation.

Qihang Yu, Qihao Liu, Ju He, Xinyang Zhang, Yang Liu, Liang-Chieh Chen, Peter Chen

This paper challenges the dominance of continuous pipelines in visual generation. We systematically investigate the performance gap between discrete and continuous methods. Contrary to the belief that discrete tokenizers are intrinsically inferior, we demonstrate that the disparity arises primarily from the total number of bits allocated in the latent space (i.e., the compression ratio). We show that scaling up the codebook size effectively bridges this gap, allowing discrete tokenizers to match or surpass their continuous counterparts. However, existing discrete generation methods struggle to capitalize on this insight, suffering from performance degradation or prohibitive training costs with scaled codebook. To address this, we propose masked **B**it **A**uto**R**egressive modeling (**BAR**), a scalable framework that supports arbitrary codebook sizes. By equipping an autoregressive transformer with a masked bit modeling head, BAR predicts discrete tokens through progressively generating their constituent bits. BAR achieves a new state-of-the-art gFID of **0.99** on ImageNet-256, outperforming leading methods across both continuous and discrete paradigms, while significantly reducing sampling costs and converging faster than prior continuous approaches.

Deep Learning · Large Language Models

Yuezhou Hu, Harman Singh, Monishwaran Maheswaran, Haocheng Xi, Coleman Hooper, Jintao Zhang, Aditya Tomar, Michael Mahoney, Sewon Min, Mehrdad Farajtabar 等

Diffusion Large Language Models (dLLMs) have emerged as a promising alternative to purely autoregressive language models because they can decode multiple tokens in parallel. However, state-of-the-art block-wise dLLMs rely on a ``remasking" mechanism that decodes only the most confident tokens and discards the rest, effectively wasting computation. We demonstrate that recycling computation from the discarded tokens is beneficial, as these tokens retain contextual information useful for subsequent decoding iterations. In light of this, we propose Residual Context Diffusion (RCD), a module that converts these discarded token representations into contextual residuals and injects them back for the next denoising step. RCD uses a decoupled two-stage training pipeline to bypass the memory bottlenecks associated with backpropagation. We validate our method on both long CoT reasoning (SDAR) and short CoT instruction following (LLaDA) models. We demonstrate that a standard dLLM can be efficiently converted to the RCD paradigm with merely $\sim$1 billion tokens. RCD consistently improves frontier dLLMs by 5--10 points in accuracy with minimal extra computation overhead across a wide range of benchmarks. Notably, on the most challenging AIME tasks, RCD nearly doubles baseline accuracy and attains up to 4--5x fewer denoising steps at equivalent accuracy levels.

Deep Learning · Large Language Models

Haocheng Xi, Harman Singh, Yuezhou Hu, Coleman Hooper, Rishabh Tiwari, Aditya Tomar, Wonjun Kang, Minjae Lee, Michael Mahoney, Chenfeng Xu 等

Block-wise diffusion language models (DLMs) generate multiple tokens in parallel, offering a promising alternative to autoregressive decoding. However, their inference efficiency remains bottlenecked by memory-bound attention in long-context scenarios. Naïve sparse attention is ineffective for DLMs due to the KV inflation problem: different queries select different prefix positions, causing the union of accessed KV pages to remain large. To address this challenge, we observe that block-wise diffusion exhibits locality of representation changes across denoising steps: only a small fraction of tokens (active tokens) undergo significant hidden-state updates, while most tokens (stable tokens) remain nearly unchanged. Based on this insight, we propose LoSA (Locality-aware Sparse Attention), which reuses cached prefix-attention results for stable tokens and applies sparse attention only to active tokens with large representation changes. This design reduces the number of queries contributing to the union of KV indices, substantially shrinking the KV pages that must be loaded. Across multiple block-wise DLMs and reasoning benchmarks, LoSA preserves near-dense accuracy while significantly improving efficiency, achieving up to 4.14× speedup over dense attention on RTX A6000 GPUs. LoSA also delivers up to 5% average improvement over baselines across all datasets and configurations, demonstrating the effectiveness of the proposed method.

Applications · Computer Vision

Hulingxiao He, Zhi Tan, Yuxin Peng

Taxonomies provide key information about the semantic relationships between concepts and the inherent organization of vision and language. Despite their impressive capabilities, large multimodal models (LMMs) often lack taxonomic knowledge, leading to low hierarchical visual recognition (HVR) consistency. These models typically only rely on language modeling objectives during fine-tuning and lack explicit taxonomy-aware regularization. To address this, we propose Hierarchical Representation Regularization (HiR$^2$), a simple plug-and-play regularizer that improves hierarchical consistency in LMMs. Specifically, we introduce a semantic-aware visual tree construction framework that extracts coarse-to-fine visual features from intermediate LLM layers guided by textual cues. The regularizer combines two complementary objectives: a taxonomic entailment loss that enforces hierarchy via hyperbolic entailment cones in the Lorentz model, and a discriminative dispersive loss that promotes angular separation of semantically similar embeddings on the unit sphere without disturbing the radial hierarchical structure. Extensive experiments demonstrate that HiR$^2$ effectively captures taxonomic structures across diverse LMMs and fine-tuning methods.

Deep Learning · Foundation Models

Houjun Liu, Shikhar Murty, Christopher Manning, Róbert Csordás

Current approaches for scaling inference-time compute in transformers train them to emit explicit chain-of-thought tokens before producing an answer. While these methods are powerful, they are limited because they cannot be applied during pretraining and rely solely on serially-generated, natural-language verbalization. In this work, we propose Thoughtbubbles, a transformer variant that natively performs parallel adaptive computation in latent space by learning to fork or delete residual streams. Thus, tokens requiring more computation can form a "bubble" of cloned residuals in the middle of the network. Crucially, this behavior is learned during pretraining with only language modeling loss. Using half of the training budget, Thoughtbubbles outperforms the perplexity and zero-shot evals of both standard decoder LMs and those using non-adaptive parallel computation approaches. These results hold across model sizes from 150M to 1.9B. Thoughtbubbles achieves competitive GSM8K results using half of the baseline's token budget. The implicit nature of our method enables models to begin learning adaptive computation at pretraining time, paving the way to unified train-time and test-time scaling behaviors.

Jiawei Xu, Minghui Liu, Aakriti Agrawal, Yifan Chen, Furong Huang

Masked diffusion language models decode by iteratively unmasking tokens, where the unmasking order defines an ``order of thought'' that strongly influences generation quality yet is typically chosen heuristically. We derive a tractable upper bound on the sequential decoding mismatch, measured by the Kullback–Leibler divergence and expressed in terms of the model’s pathwise log-likelihood, with tightness under sufficient model expressivity. This bound induces a dense self-aware reward for a target sequence $x$ and unmasking order $\sigma$, over ordered paths, casting order selection as a principled policy optimization problem with a frozen denoiser. We instantiate this idea as **Self-Aware Scheduling (SAS)**, which learns a lightweight order policy using Group Relative Policy Optimization and applies seamlessly to both sequential and semi-autoregressive decoding. On Sudoku with 1B MDM, SAS improves puzzle accuracy from $82.0\%$ (best heuristic schedule) to $91.8\%$, and reaches $97.9\%$ with second-stage fine-tuning along learned trajectories. On LLaDA-8B, SAS improves pass@1 on GSM8K from $64\%$ to $76\%$ (full diffusion) and on MBPP from $39.5\%$ to $41\%$, while consistently matching or exceeding heuristic schedules across generation lengths and block sizes.

Applications · Time Series

Jihye Na, Patara Trirat, Chanyoung Park, Jae-Gil Lee

Recent studies have attempted to fine-tune time-series foundation models to enhance a target dataset's forecasting performance. However, these approaches proceed without a clear criterion for identifying complex datasets that require fine-tuning due to performance degradation in zero-shot forecasting. To distinguish datasets that are more challenging from standard benchmarks, we introduce data-driven temporal complexity and multichannel complexity. *Temporal complexity* captures the difficulty of identifying distinct patterns by quantifying spectral entropy in the frequency domain, while *multichannel complexity* captures inter-channel dependencies by measuring the channel information flow impacting predictive uncertainty. These metrics serve as *effective proxies for performance gains* achievable through fine-tuning. Based on the two metrics, we develop *Time-PEFT*, a parameter-efficient fine-tuning framework that incorporates a frequency adapter for top-$k$ filtering and a channel adapter for multichannel modeling. *Time-PEFT* is shown to significantly improve forecasting performance by up to 2.51 times compared with existing fine-tuning techniques on complex datasets.