Model merging aims to achieve multitask performance by merging multiple expert models without the need to access the raw training data. Recent research identified the representation bias of model merging, characterized by a discrepancy in the representation distribution between the merged and individual models, hindering the performance of model merging methods. To mitigate the representation bias, a task-specific MLP, Surgery, was built to model the bias that is subsequently decreased on the merged representation. However, this strategy is still suboptimal due to the limited modeling capability within the deterministic manner. To address this issue, we present ProbSurgery, a probabilistic module specifically designed to accurately model the representation bias. This module generates an embedding distribution for each sample and outputs the representation bias through a sampling process. ProbSurgery offers superior representational capacity by naturally handling the uncertainty resulting from parameter interference of merging multiple models. Besides, we provide a theoretical analysis to reveal the advance of the probabilistic manner and propose an extension of ProSurgery for adapting to the task-sharing setting. Extensive experiments verify the effectiveness of ProbSurgery for representation surgery while maintaining generalization capabilities in real-world scenarios, including out-of-distribution and domain shift challenges.

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.

Spatial regularization has been proved as an effective method for alleviating the boundary effect and boosting the performance of a discriminative correlation filter (DCF) in aerial visual object tracking. However, existing spatial regularization methods usually treat the regularizer as a supplementary term apart from the main regression and neglect to regularize the filter involved in the correlation operation. To address the aforementioned issue, this article introduces a novel object saliency-aware dual regularized correlation filter, i.e., DRCF. Specifically, the proposed DRCF tracker suggests a dual regularization strategy to directly regularize the filter involved with the correlation operation inside the core of the filter generating ridge regression. This allows the DRCF tracker to suppress the boundary effect and consequently enhance the performance of the tracker. Furthermore, an efficient method based on a saliency detection algorithm is employed to generate the dual regularizers dynamically and provide the regularizers with online adjusting ability. This enables the generated dynamic regularizers to automatically discern the object from the background and actively regularize the filter to accentuate the object during its unpredictable appearance changes. By the merits of the dual regularization strategy and the saliency-aware dynamical regularizers, the proposed DRCF tracker performs favorably in terms of suppressing the boundary effect, penalizing the irrelevant background noise coefficients and boosting the overall performance of the tracker. Exhaustive evaluations on 193 challenging video sequences from multiple well-known challenging aerial object tracking benchmarks validate the accuracy and robustness of the proposed DRCF tracker against 27 other state-of-the-art methods. Meanwhile, the proposed tracker can perform real-time aerial tracking applications on a single CPU with sufficient speed of 38.4 frames/s.