论文检索

Deep Learning · Large Language Models

Boyuan Xiao, Bohong Chen, Yumeng Li, Ji Feng, Yao-Xiang Ding, Kun Zhou

In embodied vision-language decision making tasks such as robotic manipulation and navigation, Vision-Language and Vision-Language-Action Models (VLMs \& VLAs) are powerful tools with different benefits: VLMs are better at long-term planning, while VLAs are better at reactive control. However, their performance is limited by the same perceptual bottleneck: visual hallucinations arise due to the models’ inability to distinguish task-relevant objects from distractors. In principle, accurate identification and focus on critical objects while filtering out irrelevant ones is the key to break this limitation. A straightforward solution is one-step focus: directly attending to essential objects. However, this approach proves ineffective because effective focus inherently requires deep scene understanding. To this end, we propose ${\it SceneDiver}$, a coarse-to-fine focus plan generation method for VLMs leveraging their long-term planning abilities, that first constructs a holistic scene graph to establish initial comprehension, then progressively decomposes the task into simpler sub-problems through an iterative cycle of recognition, understanding, and analysis. To enable reactive control, we also design a lightweight adapter for distilling the deliberate focus ability into VLAs. Evaluations on standard embodied AI benchmarks confirm that our method substantially reduces visual hallucinations for both VLMs and VLAs, while preserving computational efficiency in tasks requiring fast execution.

Probabilistic Methods · Monte Carlo and Sampling Methods

James Cuin, Davide Carbone, Yanbo Tang, O. Akyildiz

The problem of optimising functions with intractable gradients frequently arise in machine learning and statistics, ranging from maximum marginal likelihood estimation procedures to fine-tuning of generative models. Stochastic approximation methods for this class of problems typically require inner sampling loops to obtain (biased) stochastic gradient estimates, which rapidly becomes computationally expensive. In this work, we develop sequential Monte Carlo (SMC) samplers for optimisation of functions with intractable gradients. Our approach replaces expensive inner sampling methods with efficient SMC approximations, which can result in significant computational gains. We establish convergence results for the basic recursions defined by our methodology which SMC samplers approximate. We demonstrate the effectiveness of our approach on the reward-tuning of energy-based models within various settings.

Applications · Everything Else

Jiyi Li

Crowdsourcing has been widely adopted for large-scale data collection and problem solving, yet its outcomes are often noisy and inconsistent, making quality control and aggregation central concerns. Meanwhile, Large Language Models (LLMs) have shown strong capabilities in generation, annotation, evaluation, and reasoning. These developments give rise to a new paradigm at the intersection of crowdsourcing and LLMs, which we term Crowd-LLM-Sourcing, encompassing two directions: (1) Crowd-LLM Collaboration, where humans and LLMs jointly participate in workflows, and (2) LLM-Sourcing Inspired by Crowdsourcing, where crowdsourcing principles guide LLM-driven generation, annotation, evaluation, and inference. Many existing studies on LLMs overlook decades of prior work in crowdsourcing, even though the two domains are grounded in closely related principles on some topics. Our central position is that, in scenarios where an LLM can be regarded as an LLM worker, LLM research should draw upon the rich body of crowdsourcing literature. At the same time, LLM workers differ fundamentally from human workers. Identifying how crowdsourcing mechanisms should be adapted, opens a new research agenda for collective intelligence with model-based agents.

Deep Learning · Graph Neural Networks

PARTH VERMA, Parv P Singh, Vipul Garg, Ishita Thakre, N M Anoop Krishnan, Sayan Ranu

Graph Neural Networks (GNNs) have revolutionized Neural Force Fields for atomistic simulations, achieving near-quantum accuracy at reduced cost, yet adapting these models to new chemical systems requires expensive retraining of foundation models. Inspired by model merging in vision and language processing, we introduce GFFMERGE, the first principled framework for closed-form model merging in GNNs. We exploit the linear structure of message-passing layers and formulate merging as a convex embedding-alignment problem with an analytical solution. Through the first systematic benchmarking of model merging for GNNs, we show that existing methods designed for vision and language catastrophically fail on force field regression, while GFFMERGE recovers performance approaching gold standard joint training. Across molecular (MD17, MD22), solid-state (LiPS20), and large-scale graph benchmarks, GFFMERGE and GNNMERGE (its generic GNN counterpart) achieve 5-27$\times$ speedups while enabling modular composition of specialized models. Remarkably, our closed-form solution alone outperforms all baseline methods before fine-tuning and provides superior initialization for faster, data-efficient convergence.

Applications · Time Series

Xin Qiu, Junlong Tong, Yirong Sun, Yunpu Ma, Xiaoyu Shen

Large-scale models are at the forefront of time series (TS) forecasting, dominated by two paradigms: fine-tuning text-based Large Language Models for TS (LLM4TS) and training Time Series Foundation Models (TSFMs) from scratch. Both approaches share a foundational assumption that scaling up model capacity and data volume leads to improved performance. However, we observe a scaling paradox in TS models, revealing a puzzling phenomenon that larger models do NOT always achieve better performance. Through extensive experiments on two model families across four scales (100M to 1.7B parameters) and diverse data (up to 6B observations), we rigorously confirm that the scaling paradox is a pervasive issue. We then diagnose its root cause by analyzing internal representations, identifying a phenomenon we call few-layer dominance: only a small subset of layers are functionally important, while the majority are redundant, under-utilized, and can even distract training. Based on this discovery, we propose a practical method to automatically identify and retain only these dominant layers. In our models, retaining only 21% of the parameters achieves up to a 12% accuracy improvement and a 2.7x inference speedup. We validate the universality of our method on 8 prominent SOTA models (LLM4TS and TSFMs, 90M to 6B), showing that retaining less than 30% layers achieves superior accuracy in over 95% tasks.

Applications · Computer Vision

Qishen Yin, Tanghui Jia, Peng Jin, Hao Li, Juntong Wu, Guanlin Lu, Li Yuan

Does \emph{Comprehending the main idea of a 2-hour movie} and \emph{Counting the birds appearing in a 15-second clip} really warrant the same video processing pipeline? We present Task-Aware Mechanism (TAM), a hybrid-gated Mixture-of-Experts (MoE) vision tower that adapts frame count and resolution to the user query and video length. TAM introduces a compact 0.1B text-only \emph{Inductor} trained on our TA-116K dataset to infer task types, enabling task-aware visual budgeting and routing: a soft-gated MoE vision encoder for stability, and hard-gated resolution-specific projectors/pipelines for efficient specialization. Built on Qwen2-7B, TallVA-8B-A7B achieves state-of-the-art performance among models with comparable LLMs across diverse video benchmarks and remains competitive against stronger-LLM baselines, showing that task-aware visual budgeting makes video understanding more holistic. The code is included in the supplementary material.

Junlong Tong, Yao Zhang, Anhao Zhao, Yingqi Fan, Yunpu Ma, Xiaoyu Shen

Standard Large Language Models (LLMs) operate on a ''read-then-generate'' paradigm, incurring avoidable latency and computational redundancy. Recently, streaming LLMs have attempted to overcome these bottlenecks by allowing input and output to unfold synchronously. However, this introduces a critical challenge: how should the LLM determine the optimal timing to interact with the input and output stream? Existing approaches remain confined to *passive adaptation*, relying on static or content-irrelevant interaction rules. In this paper, we propose *ProactiveLLM*, which achieves active interaction by treating ''when to generate" and ''what to generate" as decoupled objectives. Through masked streaming modeling and self-distillation, the model actively learns to perceive semantic sufficiency from partial inputs. This yields endogenous cues serving as a versatile interface for the plug-and-play integration of diverse decision heads customized for specific latency-accuracy trade-offs. Extensive evaluation across text and speech streaming tasks confirms that ProactiveLLM significantly reduces interaction latency while maintaining quality, validating its capacity for dynamic and active interaction.

Reinforcement Learning · Batch/Offline

Kai Fukazawa, Kunal Mundada, Iman Soltani

In safety-critical domains where online data collection is infeasible, offline reinforcement learning (RL) is attractive only if policies achieve high returns without catastrophic lower-tail risk. Prior work on risk-averse offline RL achieves safety at the cost of value- or model-based pessimism, and restricted policy classes that limit policy expressiveness, whereas diffusion/flow-based expressive generative policies have largely been used in risk-neutral settings. We introduce **Risk-Aware Multimodal Actor-Critic (RAMAC)**, a simple, modular, model-free framework that couples an expressive generative actor (e.g., diffusion/flow) with a distributional critic and optimizes a composite objective that combines Conditional Value-at-Risk (CVaR) with behavioral cloning (BC), enabling risk-sensitive learning in complex multimodal scenarios. Since out-of-distribution (OOD) actions are a major driver of catastrophic failures in offline RL, we further provide an objective-level analysis showing that controlling behavior divergence via BC suppresses OOD actions and stabilizes CVaR. Instantiating RAMAC with a diffusion actor, we illustrate these insights on a 2-D risky bandit and evaluate on Stochastic-D4RL, observing consistent gains in CVaR$_{0.1}$ while maintaining strong returns.

Deep Learning · Large Language Models

Hantao Hua, Jiming Su, hao tang, Yiping Yao, Feng Zhu

Grammar-constrained decoding is essential for enabling large language models (LLMs) to efficiently generate structured outputs in applications, such as JSON objects for parameter passing. Existing approaches typically execute grammar constraint masking on the CPU, while LLM inference is performed on the GPU. This execution mismatch introduces frequent grammar-induced CPU $\rightarrow$ GPU control and data synchronization, leading to substantial overhead in large-batch inference. In contrast, we propose Gram2Token, which preprocesses grammar constraints into token-level representations that can be executed natively on GPUs at run time, thereby reducing decoding overhead. Specifically, Gram2Token first converts the input grammar into a pushdown automaton and aligns the automaton with tokenizer outputs via a trie. Through this alignment, pushdown stack configurations are encoded into a finite set of augmented grammar states, and tokens are categorized according to the grammar states in which they are valid. We further design a GPU-native grammar-constrained decoding pipeline that replaces complex run-time grammar parsing with $O(1)$ table lookups and eliminates run-time grammar-induced CPU $\rightarrow$ GPU control dependencies. Experimental results on large-batch JSON and SQL generation tasks show that, compared to state-of-the-art implementations, **Gram2Token improves decoding throughput by 1.5×–2.3×.** These results demonstrate that GPU-native grammar-constrained decoding is an effective and scalable approach for structured LLM generation.

General Machine Learning · Representation Learning

Lingjie Yi, Raphael Douady, Chao Chen

Multimodal contrastive learning (MCL) aims to embed data from two modalities in a shared embedding space. However, in practice, representations of images and text occupy completely separate regions of embedding space, a phenomenon called the modality gap. Moreover, experimental findings on how the size of the modality gap affects downstream performance are inconsistent. These observations raise two key questions: (1) What causes the modality gap? (2) What affects downstream performance? To address these questions, we introduce the first theoretical framework for analyzing the convergent optimal representations (COR) of MCL when training is optimized. We prove that, when representations of image and text collapse into different subspaces, a phenomenon called \emph{dimension collapse}, the modality gap occurs. Our theorem also reveals that while the modality gap prevents representations of image and text from aligning directly, their projections onto the shared space can be aligned. And share space alignment plays dominate role in determining downstream performance. Inspired by these findings, we first propose Shared Space Alignment (SSA) to improve MCL pretraining by enhancing alignment within the shared space. Extensive experiments validate our theoretical analysis and proposed methods.

Deep Learning · Other Representation Learning

Wei Fang, Priyadarshini Panda

Neuromorphic event cameras possess superior temporal resolution, power efficiency, and dynamic range compared to traditional cameras. However, their asynchronous and sparse data format poses a significant challenge for conventional deep learning methods. Existing methods either convert the events into dense synchronous frame representations for processing by powerful CNNs or Transformers, but lose the asynchronous, sparse and high temporal resolution characteristics of events during the conversion process; or adopt irregular models such as sparse convolution, spiking neural networks, or graph neural networks to process the irregular event representations but fail to take full advantage of GPU acceleration. Inspired by word-to-vector models, we draw an analogy between words and events to introduce event2vec, a novel representation that allows neural networks to process events directly. This approach is fully compatible with the parallel processing capabilities of Transformers. We demonstrate the effectiveness of event2vec on the DVS Gesture, ASL-DVS, and DVS-Lip benchmarks, showing that event2vec is remarkably parameter-efficient, features high throughput and low latency, and achieves high accuracy even with an extremely low number of events or low spatial resolutions. Event2vec introduces a novel paradigm by demonstrating for the first time that sparse, irregular event data can be directly integrated into high-throughput Transformer architectures. This breakthrough resolves the long-standing conflict between maintaining data sparsity and maximizing GPU efficiency, offering a promising balance for real-time, low-latency neuromorphic vision tasks. The code is provided.

Probabilistic Methods · Everything Else

Amirmohammad Farzaneh, Salvatore D'oro, Osvaldo Simeone

Large language model (LLM)-powered agents can translate high-level user intents into plans and actions in an environment. Yet after observing an outcome, users may wonder: What if I had phrased my intent differently? We introduce a framework that enables such counterfactual reasoning in agentic LLM-driven control scenarios, while providing formal reliability guarantees. Our approach models the closed-loop interaction between a user, an LLM-based agent, and an environment as a structural causal model (SCM), and leverages test-time scaling to generate multiple candidate counterfactual outcomes via probabilistic abduction. Through an offline calibration phase, the proposed conformal counterfactual generation (CCG) yields sets of counterfactual outcomes that are guaranteed to contain the true counterfactual outcome with high probability. We showcase the performance of CCG on a wireless network control use case, demonstrating significant advantages compared to naive re-execution baselines.

Applications · Robotics

Yuchun Feng, Jinliang Zheng, Tianyi Tan, Dongxiu Liu, Yinan Zheng, Jiangmiao Pang, Tai Wang, Xianyuan Zhan

The specification of the action space plays a pivotal role in imitation-based robotic manipulation policy learning, fundamentally shaping the optimization landscape of policy learning. While recent advances have focused heavily on scaling training data and model capacity, the choice of action space remains guided by ad-hoc heuristics or legacy designs, leading to an ambiguous understanding of robotic policy design philosophies. To address this ambiguity, we conducted a large-scale and systematic empirical study, confirming that the action space does have significant and complex impacts on robotic policy learning. We dissect the action design space along temporal and spatial axes, facilitating a structured analysis of how these choices govern both policy learnability and control stability. Based on 13,000+ real-world rollouts on a bimanual robot and evaluation on 500+ trained models over four scenarios, we examine the trade-offs between absolute vs. delta representations, and joint-space vs. task-space parameterizations. Our large-scale results suggest that properly designing the policy to predict delta actions consistently improves performance, while joint-space and task-space representations offer complementary strengths, favoring control stability and generalization, respectively.

Applications · Language, Speech and Dialog

Zeliang Zong, Kai Zhang, Yarong Wang, wenming tan, Ye Ren, Jilin Hu

Mixture-of-Experts (MoE) architectures provide strong scalability for large language models, but their large expert parameter footprint poses challenges for efficient deployment. Expert pruning is widely used to reduce model size and inference cost; however, existing approaches are token-agnostic, treating all tokens equally when estimating expert importance. This uniform treatment dilutes the contributions of informative tokens and leads to suboptimal pruning decisions. To address this fundamental limitation, we propose **Step** (**S**elective **T**oken-guided **E**xpert **P**runing), a token-aware framework that rethinks expert pruning from the perspective of selective token guidance. By incorporating loss-aware expert evaluation and a lightweight knowledge-preserving mechanism, **Step** reduces information loss while removing redundant experts. Extensive experiments across different MoE architectures and model scales demonstrate the effectiveness of **Step**. On the 30B Qwen3 MoE model with 50\% expert sparsity, **Step** achieves nearly a 50\% reduction in memory usage with minimal performance degradation, delivers a 1.5$\times$ throughput improvement, and completes the entire pruning process within 10 minutes.

Yan Zhang, Zhijin Qin, Feng Xu, guiguang ding, Jungong Han

Despite rapid progress in Grounded Visual Reasoning (GVR) with MLLMs and RL-style fine-tuning, existing approaches often lack effective learning signals for intermediate grounding decisions and are prone to shortcut solutions. In this work, we explicitly decompose GVR into Evidence Generation followed by Counterfactual Answer Reasoning, and formalize this structure as a Causal Grounding Graph (CGG) in which the generated evidence acts as a causal mediator. Building on this formulation, we propose CofactGVR, which estimates the mediator’s utility via a matched counterfactual intervention that perturbs the predicted region while keeping the original image–question context fixed. The factual–counterfactual reward gap yields a principled intermediate bonus, selectively assigned to high-quality factual rollouts to promote evidence-faithful reasoning. To further stabilize and efficiently exploit this causal training signal, we incorporate a Quantile-filtered Prioritized Advantage Sampling (Q-PAS) strategy that preferentially updates on trajectories with high-magnitude advantages while filtering low-signal samples. Extensive experiments across GVR benchmarks show consistent improvements, indicating that CofactGVR strengthens reliance on informative visual evidence under controlled interventions. The source codes will be publicly available.

Theory · Learning Theory

Jiaolong Wang, Fode Zhang, Lingrui Wang

Reliable estimation of Amari $\alpha$-divergences underpins variational inference, yet unconstrained neural critics are notoriously prone to instability. We propose a scalable estimator by constraining the critic to a Reproducing Kernel Hilbert Space (RKHS) ball and approximating the kernel via band-limited Random Fourier Features (RFF). This formulation yields a linear-time objective amenable to mini-batch stochastic optimization while avoiding the cubic complexity of Gram-matrix methods. We present a unified analysis based on a four-term error decomposition—comprising RKHS approximation, feature discretization, statistical deviation, and optimization residual. Under a spectral source condition, we derive non-asymptotic bounds establishing that the RKHS approximation bias scales as $\mathcal{O}(R^{-\gamma})$, the RFF discretization error as $\mathcal{O}(R D^{-1/2})$, and the statistical error as $\mathcal{O}(R n^{-1/2})$. We further show that statistical non-degeneracy induces intrinsic local curvature, enabling our proposed Armijo-SGD to achieve local linear convergence. Empirical evaluations demonstrate that the RFF-RKHS estimator outperforms varying-representation baselines in stability, and applying this spectral regularization to GAN critics significantly enhances the capture of high-frequency data components.

Wenhao Li, Yue Lin, Yun Hua, Xiangfeng Wang, Bo Jin, Hongyuan Zha, Baoxiang Wang

Information design (ID) explores how a sender influence the optimal behavior of receivers to achieve specific objectives. While ID originates from everyday human communication, existing game-theoretic and learning methods often model information structures as numbers, which limits many applications to toy games. This work leverages LLMs and proposes a verbalized framework in Bayesian persuasion (BP), which extends classic BP to real-world games involving human dialogues for the first time. We map the BP to a verbalized mediator-augmented game, where LLMs instantiate the sender and receiver. To efficiently solve the verbalized game, we propose a generalized equilibrium-finding algorithm combining LLM and game solver. The algorithm is reinforced with techniques including verbalized commitment assumptions, verbalized obedience constraints, and information obfuscation. Experiments in dialogue scenarios, such as recommendation letters, law enforcement, diplomacy with press, validate that our framework can reproduce theoretical results in classic BP and discover effective persuasion strategies in more complex natural language and multi-stage scenarios.

Deep Learning · Other Representation Learning

Hyeonbin Hwang, Yeachan Park

Grokking, a sudden transition from memorization to generalization, has been closely linked to the emergence of low-dimensional representations; yet the mechanism driving this organization remains elusive. Here, we propose that intrinsic task symmetries are the key drivers of grokking, inducing structured geometries in representation space. Our analysis reveals a consistent three-stage training dynamic: (i) data memorization, (ii) intrinsic symmetry acquisition, and (iii) geometric organization. We show that generalization emerges during the symmetry acquisition phase, and subsequently the embedding space organizes into a low-dimensional structured geometry. We validate this mechanism across diverse algorithmic domains, spanning algebraic (modular arithmetic), structural (graph metric completion), and relational (comparison) reasoning tasks. Leveraging these insights, we formulate a symmetry-based criterion for generalization and propose symmetry- and geometry-prompting training strategies that can accelerate generalization. Together, our results establish intrinsic symmetry as a central mechanism enabling neural networks to move beyond memorization and achieve robust algorithmic reasoning.

Theory · Everything Else

Alexander Munteanu, Matteo Russo, David Saulpic, Chris Schwiegelshohn

Terminal embeddings have emerged as a powerful tool for dimension reduction. Given a set of points $P\subset \mathbb{R}^d$, a terminal embedding is a mapping $f:\mathbb{R}^d\rightarrow \mathbb{R}^t$ that preserves the pairwise distance between any pair of points $p\in P$ and $q\in \mathbb{R}^d$ up to small distortion under this mapping. Terminal embeddings have been particularly fruitful for constructing $k$-means and $k$-median coresets, where the objective is to find a typically weighted subset $\Omega$ of $P$ such that for any candidate solution, the cost of the clustering objective on $\Omega$ approximates the cost of the clustering objective on $P$ up to small distortion. Unfortunately, these techniques have not been extended to more complicated structures such as clustering time-series data under common straight-line interpolation between measurements. The main issue is that terminal embeddings, arguably the central technique in this line of research, cannot be linear and are thus not immediately suitable to preserve linear structures. In this work, we develop a generalization of terminal embeddings to affine line-segments that overcomes this issue. We showcase their applicability by using our lines-preserving terminal embeddings to obtain the first dimension-free coresets for clustering time-series under the Fréchet distance. The underlying dimension reduction uses Johnson-Lindenstrauss embeddings, and our experiments indicate that they compare favorably against PCA for real-world time-series.

Huang Huang, Sriram Yenamandra, Arjun Majumdar, Elie Aljalbout, Tushar Nagarajan, Jimmy Yang, Akshara Rai, Michael Rabbat, Li Fei-Fei, Jiajun Wu 等

The diversity of robot embodiments and action spaces makes it challenging to build robot world models that generalize across different embodiments. We introduce a Latent Action Conditioned Robot World Model (LAC-WM), which operates within a learned unified latent action space shared across diverse embodiments. We show how this unified action space improves the world model’s performance when adapted to previously unseen robot embodiments. We compare LAC-WM to a baseline model, Explicit Action Conditioned World Model (EAC-WM) conditioned on explicit motion labels. Our results show that conditioning on explicit labels creates disjoint action spaces across embodiments, limiting downstream task performance when adapting to new robots. We evaluate both models on a dexterous manipulation task. The latent action-conditioned model LAC-WM achieves up to a 46.7% improvement in performance over EAC-WM. Crucially, the unified latent action space allows LAC-WM’s downstream performance to scale positively with the number of embodiments used during pretraining. In contrast, the disjoint action space in EAC-WM leads to decreased performance as the number of pretraining embodiments increases. These results highlights the importance of a unified action space for efficient cross-embodiment learning, addressing a key challenge in robotics.