论文检索

Deep Learning · Generative Models and Autoencoders

Marco Pegoraro, Evan Atherton, Bruno Roy, Aliasghar Khani, Arianna Rampini

Learning natural body motion remains challenging due to the strong coupling between spatial geometry and temporal dynamics. Embedding motion in phase manifolds, latent spaces that capture local periodicity, has proven effective for motion prediction; however, existing approaches lack scalability and remain confined to specific settings. We introduce FunPhase, a functional periodic autoencoder that learns a phase manifold for motion and replaces discrete temporal decoding with a function-space formulation, enabling smooth trajectories that can be sampled at arbitrary temporal resolutions. FunPhase unifies motion prediction and generation within a single interpretable phase manifold, enabling motion generation via latent diffusion, generalizes across skeletons and datasets, and supports downstream tasks such as motion super-resolution and partial-body completion. Our model achieves substantially lower reconstruction error than prior periodic autoencoder baselines, achieving uniform improvements of at least 45% across all metrics, while enabling a broader range of applications and performing on par with state-of-the-art motion generation methods.

Optimization · Large Scale, Parallel and Distributed

Benjamin Thérien, Xiaolong Huang, Aaron Defazio, Irina Rish, Eugene Belilovsky

DiLoCo is a powerful framework for training large language models (LLMs) under networking constraints, allowing for increased parallelism and accelerator utilization in data center settings. A critical but often overlooked factor in DiLoCo’s behavior is the choice of inner optimizer, which shapes the pseudogradient used by the outer optimizer. Given the recent success of Muon relative to AdamW for data parallel (DP) training, in this work, we examine how Muon's normalized optimizer steps can affect the pseudogradient's quality. Empirically, we find that, relative to AdamW, Muon yields more \emph{directionally correct} pseudogradients as the number of workers ($K$) is increased. In our experiments pre-training language models, we conduct extensive hyperparameter tuning across 150M, 416M, 914M, 1.76B, and 3.1B models for DiLoCo, MuLoCo, AdamW DP, and Muon DP. Consistently across all scales, we find that with $K\geq1$ workers, MuLoCo (Muon inner optimizer DiLoCo) achieves superior performance to DiLoCo in absolute terms and for $K>2$ it outperforms DiLoCo relative to their data parallel baselines, while being compatible with quantization, streaming, and long synchronization intervals. At $K=1$, we find that MuLoCo can even outperform the data-parallel gold standard while having larger optimal and critical batch sizes.

Deep Learning · Other Representation Learning

David Méndez, Roberto Confalonieri, Natalia Díaz-Rodríguez

Vision-Language Models (VLMs) excel at tasks like zero-shot classification and cross-modal retrieval by mapping images and text to a shared space, but this requires expensive end-to-end training with massive paired datasets. Current post-hoc alignment methods reduce computational costs by connecting pretrained encoders through lightweight mappings, yet still demand substantial paired data. In this work, we investigate the potential of repurposing the classification heads of pretrained vision models as semantic prototypes. The recycling of these weights, typically discarded after pretraining, unlocks two distinct capabilities: it enables zero-shot alignment by using weights as semantic anchors, and serves as a robust data augmentation strategy by mixing these prototypes with real image-text pairs. We demonstrate that integrating our approach with several state-of-the-art post-hoc alignment techniques consistently boosts accuracy in cross-modal retrieval, zero- and few-shot classification tasks.

Deep Learning · Generative Models and Autoencoders

Yida Xiong, Jiameng Chen, Xiuwen Gong, Jia Wu, Shirui Pan, wenbin Hu

Graph generation aims to sample discrete node and edge attributes while satisfying coupled structural constraints. Diffusion models for graphs often adopt largely factorized forward-noising, and many flow-matching methods start from factorized reference noise and coordinate-wise interpolation, so node–edge coupling is not encoded by the generative geometry and must be recovered implicitly by the core network, which can be brittle after discrete decoding. Bayesian Flow Networks (BFNs) evolve distribution parameters and naturally support discrete generation. But classical BFNs typically rely on factorized beliefs and independent channels, which limit geometric evidence fusion. We propose Variational Bayesian Flow Network (VBFN), which performs a variational lifting to a tractable joint Gaussian variational belief family governed by structured precisions. Each Bayesian update reduces to solving a symmetric positive definite linear system, enabling coupled node and edge updates within a single fusion step. We construct sample-agnostic sparse precisions from a representation-induced dependency graph, thereby avoiding label leakage while enforcing node–edge consistency. On synthetic and molecular graph datasets, VBFN improves fidelity and diversity, and surpasses baseline methods. The reproducible code is available at https://anonymous.4open.science/r/VBFN-D6B0/.

Deep Learning · Generative Models and Autoencoders

Vamshi Sunku Mohan, Kaustubh Gupta, Aneesha Das, Chandan Singh

State-space models (SSMs) have emerged as an efficient strategy for building powerful language models, avoiding the quadratic complexity of computing attention in transformers. Despite their promise, the interpretability and steerability of modern SSMs remain relatively underexplored. We take a major step in this direction by identifying *activation subspace bottlenecks* in the Mamba family of SSM models using tools from mechanistic interpretability. We then introduce a test-time steering intervention that simply multiplies the activations of the identified bottlenecks by a scalar. Across 5 SSMs and 6 diverse benchmarks, this intervention improves performance by an average of 8.27%, without requiring any task-specific tuning. Finally, we validate that the identified bottlenecks are indeed hindering performance by modifying them to yield an architecture we call Stable-Mamba, which achieves long-context performance gains when retrained from scratch.

Deep Learning · Generative Models and Autoencoders

Dengsheng Chen, Yangming Shi, Enhua Wu

Conventional autoregressive models achieve causality through next-token prediction, but suffer from prohibitive latency and typically under-perform non-causal alternatives such as masked prediction and diffusion. We propose self-token prediction, which enables predicting a flexible number of tokens per step, and introduce AGARIC, the first spatially causal image generator built on this paradigm. AGARIC delivers markedly faster inference and consistently outperforms prior autoregressive baselines, matching the performance of state-of-the-art non-causal models. Our findings point to self-token prediction as a key step toward unified and efficient multimodal autoregressive modeling.

Deep Learning · Generative Models and Autoencoders

Bin Zang, Rengan Xie, Xiaoliang Luo, Zhiyuan Fang, Shi Li, Lvchun Wang, Wei Yu, Yi Zhao, Yuchi Huo, Tian Xie 等

Recently, a line of works can generate impressive 3D objects from a single image, but they are limited by restricted representation resolution, making them unsuitable for 3D scene generation. In this work, we introduce \name, a novel method for high-quality 3D scene generation based on hierarchical voxel enhancement framework. Specifically, given a single scene image as input, we first produce a coarse initial scene, then introduce image segmentation and attention-based retrieval to align 2D image components with 3D scene components. Subsequently, we organize these scene relations into a hierarchical component tree, where nodes closer to the leaves denote finer-grained components. Finally, we propose a voxel super-resolution model that generates refined voxels for the target instance while maintaining strong consistency with the coarse voxels. Equipped with this model, we perform coarse-to-fine hierarchical super-resolution on images and voxels for each component, producing a high-resolution and high-quality 3D scene. Extensive experiments demonstrate that our method significantly outperforms previous approaches, achieving state-of-the-art performance.

Deep Learning · Generative Models and Autoencoders

Henry Smith, Nathaniel Diamant, Brian Trippe

Generative models frequently suffer miscalibration, wherein statistics of the sampling distribution such as class probabilities deviate from desired values. We frame calibration as a constrained optimization problem and seek the closest model in Kullback-Leibler divergence satisfying calibration constraints. To address the intractability of imposing these constraints exactly, we introduce two surrogate objectives for fine-tuning: (1) the relax loss, which replaces the constraint with a miscalibration penalty, and (2) the reward loss, which converts calibration into a reward fine-tuning problem. We demonstrate that these approaches substantially reduce calibration error across hundreds of simultaneous constraints and models with up to nine billion parameters, spanning applications in protein design, image generation, and language modeling.

Reinforcement Learning · Inverse

Kathryn Wantlin, Chongyi Zheng, Benjamin Eysenbach

In the same way that generative models today conduct most of their training in a self-supervised fashion, how can agentic models conduct their training in a self-supervised fashion, interactively exploring, learning, and preparing to quickly adapt to new tasks? The problem of reward-free exploration is well studied in the unsupervised reinforcement learning (URL) literature but fails to prepare agents for rapid adaptation to new demonstrations. Today's language and vision models are trained on data provided by humans, which provides a strong inductive bias for the sorts of tasks that the model will have to solve. However, when prompted to imitate a new task, some methods perform distribution matching against the demonstration data without properly accounting for the difficulty of various tasks. The key contribution of our paper is a method for pre-training interactive agents in a self-supervised fashion, so that they can instantly mimic expert demonstrations. Our method treats goals (i.e., observations) as the atomic construct. During training, our method automatically proposes goals and practices reaching them, building off prior work in reinforcement learning exploration. During evaluation, our method solves an (amortized) inverse reinforcement learning problem to explain demonstrations as optimal goal-reaching behavior. Experiments on standard benchmarks (not designed for goal-reaching) show that our approach outperforms prior methods for zero-shot imitation.

Deep Learning · Foundation Models

Seulbi Lee, Sangheum Hwang

Large Vision Language Models (LVLMs) have achieved remarkable progress, yet they often suffer from language bias, producing answers without relying on visual evidence. While prior work attempts to mitigate this issue through decoding strategies, architectural modifications, or curated instruction data, they typically lack a quantitative measure of how much individual training samples or tokens actually benefit from the image. In this work, we introduce Visual Information Gain (VIG), a perplexity-based metric that measures the reduction in prediction uncertainty provided by visual input. VIG enables fine-grained analysis at both sample and token levels, effectively highlighting visually grounded elements such as colors, spatial relations, and attributes. Leveraging this, we propose a VIG-guided selective training scheme that prioritizes high-VIG samples and tokens. This approach improves visual grounding and mitigates language bias, achieving superior performance with significantly reduced supervision by focusing exclusively on visually informative samples and tokens.

Deep Learning · Foundation Models

Jungin Park, Jiyoung Lee, Kwanghoon Sohn

This study introduces an intriguing phenomenon in Video LLMs: rather than merely translating frames into textual embeddings, Video LLMs establish a continuous manifold, token interface, allowing visual tokens to operate as standalone entities within the architecture. Exploiting this discovery, we propose LynX, a scalable framework that integrates novel modalities by repurposing the internalized interface. Departing from conventional paradigms that necessitate heavy modality-specific encoders or paired supervision, LynX employs a lightweight auxiliary pathway in parallel with the frozen vision encoder. By aligning both the attention responses and the statistical distributions using unimodal data alone, our method synchronizes new sensory inputs with intrinsic video priors. Crucially, our distributional alignment ensures manifold compatibility while preserving the integrity of the Video LLMs. Extensive benchmarks demonstrate that LynX achieves state-of-the-art performance and efficiency across audio-visual QA, 3D reasoning, high-frame-rate, and multi-view video understanding. The code is available at https://anonymous.4open.science/r/lynx-DDC8/.

Theory · Learning Theory

Dechen Zhang, Xuan Tang, Yingyu Liang, Difan Zou

Low-precision training is critical for optimizing the trade-off between model quality and training costs, necessitating the joint allocation of model size, dataset size, and numerical precision. While empirical scaling laws suggest that quantization impacts effective model and data capacities or acts as an additive error, the theoretical mechanisms governing these effects remain largely unexplored. In this work, we initiate a theoretical study of scaling laws for low-precision training within a high-dimensional sketched linear regression framework. We demonstrate that the impact of quantization is twofold: it introduces an additive error and, simultaneously, shrinks the effective model and data sizes. Crucially, our analysis reveals distinct behaviors across quantization regimes: multiplicative quantization (where error variance scales with signal magnitude) primarily reduces the effective data size, whereas additive quantization (where error variance is independent of the signal) diminishes both the effective model size and data size. Numerical experiments validate our theoretical findings. By rigorously characterizing the complex interplay among model scale, dataset size, and quantization error, our work provides a principled theoretical basis for optimizing training protocols under practical hardware constraints.

Theory · Learning Theory

Jichu Li, Xuan Tang, Difan Zou

A variety of widely used optimization methods like SignSGD and Muon can be interpreted as instances of steepest descent under different norm-induced geometries. In this work, we study the implicit bias of mini-batch stochastic steepest descent in multi-class classification, characterizing how batch size, momentum, and variance reduction shape the limiting max-margin behavior and convergence rates under general entry-wise and Schatten-$p$ norms. We show that without momentum, convergence only occurs with large batches, yielding a batch-dependent margin gap but the full-batch convergence rate. In contrast, momentum enables small-batch convergence through a batch–momentum trade-off, though it slows convergence. This approach provides fully explicit, dimension-free rates that improve upon prior results. Moreover, we prove that variance reduction can recover the exact full-batch implicit bias for any batch size, albeit at a slower convergence rate. Finally, we further investigate the batch-size-one steepest descent without momentum, and reveal its convergence to a fundamentally different bias via a concrete data example, which reveals a key limitation of purely stochastic updates. Overall, our unified analysis clarifies when stochastic optimization aligns with full‑batch behavior, and paves the way for perform deeper explorations of the training behavior of stochastic gradient steepest descent algorithms.

Deep Learning · Attention Mechanisms

Sumin Park, Seojin Kim, Noseong Park

Linear attention reformulates sequence modeling as recurrent state evolution, enabling efficient linear-time inference. Under the key–value associative paradigm, existing approaches restrict the role of the query to the readout operation, decoupling it from state evolution. We show that query-conditioned state readout induces a structured value prediction over accumulated memory that complements key-based retrieval. Based on this insight, we propose Q-Delta, a query-aware delta rule that integrates mixed key--query prediction errors into state evolution, enabling jointly corrective dynamics while preserving delta-rule efficiency. We establish stability guarantees for the resulting dynamics and derive a hardware-efficient chunkwise-parallel formulation with a custom Triton implementation. Empirical results demonstrate stable optimization, competitive throughput, and consistent improvements over strong baselines on language modeling and long-context retrieval tasks.

Deep Learning · Large Language Models

Sumin Park, Noseong Park

Mixture-of-Experts (MoE) scales model capacity efficiently by selectively routing inputs to a specialized subset of experts. However, input-expert specialization, the core motivation of MoE, critically depends on whether the router is actually aware of input structure. In practice, MoE routing is typically implemented as a shallow linear projection with limited awareness of input representation, which often leads to unstable and imbalanced routing. We propose STAR, a STructure-Aware Routing that rethinks MoE routing as a subspace learning problem by augmenting standard learnable routing with an evolving principal subspace that tracks dominant input structure via Generalized Hebbian Algorithm (GHA). By aligning routing decisions directly with input structure along with the task-supervision from learnable gate, STAR enables stable and balanced expert specialization without relying on auxiliary load-balancing losses. We evaluate STAR on controlled synthetic setup and large-scale language and vision tasks, where it consistently improves routing quality and downstream performance over strong MoE baselines. Moreover, optional test-time subspace updates further enhance routing robustness under distribution shifts.

Deep Learning · Foundation Models

Jianxiong Zhang, Bing Guo, Yuming Jiang, Haobo Wang, Bo An, Sean Du

Large reasoning models (LRMs) often generate long, seemingly coherent reasoning traces yet still produce incorrect answers, making hallucination detection challenging. Although trajectories contain useful signals, directly using trace text or vanilla hidden states for detection is brittle: traces vary in form and detectors can overfit to superficial patterns rather than answer validity. We introduce Answer-agreement Representation Shaping (ARS), which learns detection-friendly trace-conditioned representations by explicitly encoding answer stability. ARS generates counterfactual answers through small latent interventions, specifically, perturbing the trace-boundary embedding, and labels each perturbation by whether the resulting answer agrees with the original. It then learns representations that bring answer-agreeing states together and separate answer-disagreeing ones, exposing latent instability indicative of hallucination risk. The shaped embeddings are plug-and-play with existing embedding-based detectors and require no human annotations during training. Experiments demonstrate that ARS consistently improves detection and achieves substantial gains over strong baselines.

Theory · Everything Else

Naïl B. Khelifa, Richard E Turner, Ramji Venkataramanan

Machine learning models are increasingly trained or fine-tuned on synthetic data. Recursively training on such data has been observed to significantly degrade performance in a wide range of tasks, often characterized by a progressive drift away from the target distribution. In this work, we theoretically analyze this phenomenon in the setting of score-based diffusion models. For a realistic pipeline where each training round uses a combination of synthetic data and fresh samples from the target distribution, we obtain upper and lower bounds on the accumulated divergence between the generated and target distributions. This allows us to characterize different regimes of drift, depending on the score estimation error and the proportion of fresh data used in each generation. We also provide empirical results on synthetic data and images to illustrate the theory.

Applications · Computer Vision

Zhuang Xiong, Chen Zhang, Qingshan Xu, Wenbing Tao

Despite recent progress in calibration-free monocular SLAM via 3D vision foundation models, scale drift remains severe on long sequences. Motion-agnostic partitioning breaks contextual coherence and causes zero-motion drift, while conventional geometric alignment is computationally expensive. To address these issues, we propose VGGT-Motion, a calibration-free SLAM system for efficient and robust global consistency over kilometer-scale trajectories. Specifically, we first propose a motion-aware submap construction mechanism that uses optical flow to guide adaptive partitioning, prune static redundancy, and encapsulate turns for stable local geometry. We then design an anchor-driven direct Sim(3) registration strategy. By exploiting context-balanced anchors, it achieves search-free, pixel-wise dense alignment and efficient loop closure without costly feature matching. Finally, a lightweight submap-level pose graph optimization enforces global consistency with linear complexity, enabling scalable long-range operation. Experiments show that VGGT-Motion markedly improves trajectory accuracy and efficiency, achieving state-of-the-art performance in zero-shot, long-range calibration-free monocular SLAM.

Optimization · Large Scale, Parallel and Distributed

Insu Jang, Runyu Lu, Nikhil Bansal, Ang Chen, Mosharaf Chowdhury

Multimodal large language models (MLLMs) extend the capabilities of large language models (LLMs) by combining heterogeneous model architectures to handle diverse modalities like images and audio. However, this inherent heterogeneity in MLLM model structure and data types makes makeshift extensions to existing LLM training frameworks unsuitable for efficient MLLM training, especially in distributed training. In this paper, we present ModalGlue, an efficient distributed MLLM training framework that contemplates MLLM's unique characteristics in both model and data parallelization. ModalGlue introduces frozen-aware pipeline parallelism and workload-balanced context parallelism to improve MLLM training throughput. Our extensive evaluation shows that \name outperforms state-of-the-art solutions by $2.26\times$ on average in terms of MLLM training throughput.

Deep Learning · Foundation Models

Xinchen Zhang, Bowei Liu, Jiale Liu, Chufan Shi, Yizhen Zhang, Junhong Liu, Youliang Zhang, Zhiheng Li, Yujiu Yang, Ling Yang

Visual outcomes are increasingly central to multimodal large language models, making reliable and fine-grained verification essential for scaling generalist foundation models. In this work, we investigate ***multimodal meta-verification***, which leverages verifier-generated rationales rather than decision-only signals, and explore how to effectively incorporate meta-verification feedback into multimodal verifier training. We identify two key findings. First, symbolic verifier outputs (e.g., bounding boxes) outperform textual explanations as meta-verification rationales, enabling efficient rule-based reinforcement learning rewards while avoiding reliance on model-based rewards from auxiliary judge models. Second, decoupling reinforcement learning objectives for binary judgment and meta-verification substantially outperforms joint reward optimization, due to intrinsic differences in output structure and learning dynamics. Based on these insights, we train **OmniVerifier-M1**, a generalist visual verifier leveraging symbolic meta-verification and decoupled reinforcement learning. OmniVerifier-M1 provides robust verification and fine-grained error localization, and further enables **M1-TTS**, a verifier-driven agentic generation system achieving dynamic region-level self-correction. This approach paves the way for more reliable, interpretable, and fine-grained multimodal verification, supporting safer and more controllable foundation model deployment.