Lang Feng, Fuchao Yang, Feng Chen, Xin Cheng, Haiyang Xu, Zhenglin Wan, Ming Yan, Bo An. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Online fine-tuning vision-language model (VLM) agents with reinforcement learning (RL) has shown promise for equipping agents with multi-step, goal-oriented capabilities in dynamic environments. However, their open-ended textual action space and non-end-to-end nature of action generation present significant challenges to effective online exploration in RL, e.g., explosion of the exploration space. We propose a novel online fine-tuning method, Counterfactual Soft Reinforcement Learning (CoSo), better suited to the textual output space of VLM agents. Compared to prior methods that assign uniform uncertainty to all tokens, CoSo leverages counterfactual reasoning to dynamically assess the causal influence of individual tokens on post-processed actions. By prioritizing the exploration of action-critical tokens while reducing the impact of semantically redundant or low-impact tokens, CoSo enables a more targeted and efficient online rollout process. We provide theoretical analysis proving CoSo's convergence and policy improvement guarantees, and extensive empirical evaluations supporting CoSo's effectiveness. Our results across a diverse set of agent tasks, including Android device control, card gaming, and embodied AI, highlight its remarkable ability to enhance exploration efficiency and deliver consistent performance gains. The code is available at https://github.com/langfengQ/CoSo.

To learn from data collected in diverse dynamics, Imitation from Observation (IfO) methods leverage expert state trajectories based on the premise that recovering expert state distributions in other dynamics facilitates policy learning in the current one. However, Imitation Learning inherently imposes a performance upper bound of learned policies. Additionally, as the environment dynamics change, certain expert states may become inaccessible, rendering their distributions less valuable for imitation. To address this, we propose a novel framework that integrates reward maximization with IfO, employing F-distance regularized policy optimization. This framework enforces constraints on globally accessible states—those with nonzero visitation frequency across all considered dynamics—mitigating the challenge posed by inaccessible states. By instantiating F-distance in different ways, we derive two theoretical analysis and develop a practical algorithm called Accessible State Oriented Policy Regularization (ASOR). ASOR serves as a general-purpose module that can be incorporated into various RL approaches, including offline RL and off-policy RL. Extensive experiments across multiple benchmarks demonstrate ASOR’s effectiveness in enhancing state-of-the-art cross-domain policy transfer algorithms, significantly improving their performance.

Single-domain generalization (SDG) in time series classification (TSC) poses significant challenges for current time-series domain generalization methods due to the extremely limited data available from only one source domain. In this study, we propose Segment-dErived Expansion of Domains (SEED), a diffusion-based method that effectively expands domain diversity for SDG. We reveal that individual instances exhibit intrinsic temporal shifts over time, which provides a principled foundation for creating multiple pseudo domains by segmenting each instance into distinct parts. To do so, SEED extracts two complementary representations from each time-series segment: 1) a segment-specific representation that captures diverse distributional variations, and 2) a segment-invariant representation that preserves class semantics. SEED formulates these representations as pseudo-domain prompts to guide a diffusion model in generating diverse yet semantically consistent time-series data. Additionally, SEED introduces a novel prompt-fused sampling for diffusion, enabling flexible recombination of segment-specific features to continuously expand the pseudo-domain space. We provide both theoretical analysis and extensive empirical evaluations on four widely used TSC benchmarks to validate its ability in reducing generalization error and improving model's performances in SDG. In our experiments, SEED significantly improves classification accuracy by 7.68% on average compared to the strong baselines.

Recent advances in group-based reinforcement learning (RL) have driven frontier large language models (LLMs) in single-turn tasks like mathematical reasoning. However, their scalability to multi-turn LLM agent training remains limited. Unlike static tasks, agent-environment interactions unfold over many steps and often yield sparse or delayed rewards, making credit assignment across individual steps significantly more challenging. In this work, we propose Group-in-Group Policy Optimization (GiGPO), a novel RL algorithm that achieves fine-grained credit assignment for LLM agents while preserving the appealing properties of group-based RL: critic-free, low memory, and stable convergence. GiGPO introduces a two-level structure for estimating relative advantage: (i) At the episode-level, GiGPO computes macro relative advantages based on groups of complete trajectories; (ii) At the step-level, GiGPO introduces an anchor state grouping mechanism that retroactively constructs step-level groups by identifying repeated environment states across trajectories. Actions stemming from the same state are grouped together, enabling micro relative advantage estimation. This hierarchical structure effectively captures both global trajectory quality and local step effectiveness without relying on auxiliary models or additional rollouts. We evaluate GiGPO on challenging agent benchmarks, including ALFWorld and WebShop, as well as tool-integrated reasoning on search-augmented QA tasks, using Qwen2.5-1.5B/3B/7B-Instruct. Crucially, GiGPO delivers fine-grained per-step credit signals, achieves performance gains of > 12\% on ALFWorld and > 9\% on WebShop over GRPO, and obtains superior performance on QA tasks (42.1\% on 3B and 47.2\% on 7B): all while maintaining the same GPU memory overhead, identical LLM rollout, and incurring little to no additional time cost.

Existing domain generalization (DG) methods for cross-person sensor-based activity recognition tasks often struggle to capture both intra- and inter-domain style diversity, leading to significant domain gaps with the target domain. In this study, we explore a novel perspective to tackle this problem, a process conceptualized as domain padding. This proposal aims to enrich the domain diversity by synthesizing intra- and inter-domain style data while maintaining robustness to class labels. We instantiate this concept using a conditional diffusion model and introduce a style-fused sampling strategy to enhance data generation diversity, termed Diversified Intra- and Inter-domain distributions via activity Style-fused Diffusion modeling (DI2SDiff). In contrast to traditional condition-guided sampling, our style-fused sampling strategy allows for the flexible use of one or more random style representations from the same class to guide data synthesis. This feature presents a notable advancement: it allows for the maximum utilization of possible combinations among existing styles to generate a broad spectrum of new style instances. We further extend DI2SDiff into DI2SDiff++ by enhancing the diversity of style guidance. Specifically, DI2SDiff++ integrates a multi-head style conditioner to provide multiple distinct, decomposed substyles and introduces a substyle-fused sampling strategy that allows cross-class substyle fusion for broader guidance. Empirical evaluations on a wide range of datasets demonstrate that our generated data achieves remarkable diversity within the domain space. Both intra- and inter-domain generated data have been proven significant and valuable, enabling DI2SDiff and DI2SDiff++ to surpass state-of-the-art DG methods in various cross-person activity recognition tasks.

Diffusion planners have shown promise in handling long-horizon and sparse-reward tasks due to the non-autoregressive plan generation. However, their inherent stochastic risk of generating infeasible trajectories presents significant challenges to their reliability and stability. We introduce a novel approach, the Trajectory Aggregation Tree (TAT), to address this issue in diffusion planners. Compared to prior methods that rely solely on raw trajectory predictions, TAT aggregates information from both historical and current trajectories, forming a dynamic tree-like structure. Each trajectory is conceptualized as a branch and individual states as nodes. As the structure evolves with the integration of new trajectories, unreliable states are marginalized, and the most impactful nodes are prioritized for decision-making. TAT can be deployed without modifying the original training and sampling pipelines of diffusion planners, making it a training-free, ready-to-deploy solution. We provide both theoretical analysis and empirical evidence to support TAT’s effectiveness. Our results highlight its remarkable ability to resist the risk from unreliable trajectories, guarantee the performance boosting of diffusion planners in 100% of tasks, and exhibit an appreciable tolerance margin for sample quality, thereby enabling planning with a more than $3\times$ acceleration.

Existing domain generalization (DG) methods for cross-person generalization tasks often face challenges in capturing intra- and inter-domain style diversity, resulting in domain gaps with the target domain. In this study, we explore a novel perspective to tackle this problem, a process conceptualized as domain padding. This proposal aims to enrich the domain diversity by synthesizing intra- and inter-domain style data while maintaining robustness to class labels. We instantiate this concept using a conditional diffusion model and introduce a style-fused sampling strategy to enhance data generation diversity. In contrast to traditional condition-guided sampling, our style-fused sampling strategy allows for the flexible use of one or more random styles to guide data synthesis. This feature presents a notable advancement: it allows for the maximum utilization of possible permutations and combinations among existing styles to generate a broad spectrum of new style instances. Empirical evaluations on a broad range of datasets demonstrate that our generated data achieves remarkable diversity within the domain space. Both intra- and inter-domain generated data have proven to be significant and valuable, contributing to varying degrees of performance enhancements. Notably, our approach outperforms state-of-the-art DG methods in all human activity recognition tasks.

While one-dimensional convolutional neural networks (1D-CNNs) have been empirically proven effective in time series classification tasks, we find that there remain undesirable outcomes that could arise in their application, motivating us to further investigate and understand their underlying mechanisms. In this work, we propose a Temporal Convolutional Explorer (TCE) to empirically explore the learning behavior of 1D-CNNs from the perspective of the frequency domain. Our TCE analysis highlights that deeper 1D-CNNs tend to distract the focus from the low-frequency components leading to the accuracy degradation phenomenon, and the disturbing convolution is the driving factor. Then, we leverage our findings to the practical application and propose a regulatory framework, which can easily be integrated into existing 1D-CNNs. It aims to rectify the suboptimal learning behavior by enabling the network to selectively bypass the specified disturbing convolutions. Finally, through comprehensive experiments on widely-used UCR, UEA, and UCI benchmarks, we demonstrate that 1) TCE's insight into 1D-CNN's learning behavior; 2) our regulatory framework enables state-of-the-art 1D-CNNs to get improved performances with less consumption of memory and computational overhead.

Human brain can effectively integrate visual and auditory information. Dynamic Vision Sensor (DVS) and Dynamic Audio Sensor (DAS) are event-based sensors imitating the mechanism of human retina and cochlea. Since the sensors record the visual and auditory input as asynchronous discrete events, they are inherently suitable to cooperate with the spiking neural network (SNN). Existing works of SNNs for processing events mainly focus on unimodality, however, audiovisual multimodal SNNs are still limited. In this paper, we propose an end-to-end event-based multimodal spiking neural network. The network consists of visual and auditory unimodal subnetworks and a novel attention-based cross-modal subnetwork for fusion. The attention mechanism measures the significance of each modality and allocates the weights to two modalities. We evaluate our proposed multimodal network on an event-based audiovisual joint dataset (MNIST-DVS and N-TIDIGITS datasets). Experimental results show the performance improvement of this multimodal network and the effectiveness of our proposed attention mechanism.

Spiking neural networks (SNNs) are bio-inspired neural networks with asynchronous discrete and sparse characteristics, which have increasingly manifested their superiority in low energy consumption. Recent research is devoted to utilizing spatio-temporal information to directly train SNNs by backpropagation. However, the binary and non-differentiable properties of spike activities force directly trained SNNs to suffer from serious gradient vanishing and network degradation, which greatly limits the performance of directly trained SNNs and prevents them from going deeper. In this paper, we propose a multi-level firing (MLF) method based on the existing spatio-temporal back propagation (STBP) method, and spiking dormant-suppressed residual network (spiking DS-ResNet). MLF enables more efficient gradient propagation and the incremental expression ability of the neurons. Spiking DS-ResNet can efficiently perform identity mapping of discrete spikes, as well as provide a more suitable connection for gradient propagation in deep SNNs. With the proposed method, our model achieves superior performances on a non-neuromorphic dataset and two neuromorphic datasets with much fewer trainable parameters and demonstrates the great ability to combat the gradient vanishing and degradation problem in deep SNNs.

Abstract Low earth orbit constellation satellite communication has the characteristics of low propagation delay, low path loss, low launch cost and wide range of applications. Due to the low orbital altitude and the short orbital period of low earth orbit satellites, the relative position between satellites and gateway stations changes fast. As a result, the links between gateway stations and satellites need to be switched continuously. Based on the minimum handover frequency algorithm, this paper considers the balance of satellite workloads, and proposes a load balanced satellite handover strategy. In the proposed handover strategy, a joint optimization algorithm is employed, and the power allocation of the satellite is optimized to improve the system capacity. In the multi‐satellite connection model, an adaptive power allocation algorithm is proposed to guarantee the service quality of the system. Simulation results demonstrate the efficiency of the proposed satellite handover strategy.