Reinforcement learning for deep-search agents has largely focused on trajectory-level scoring—outcome correctness, citation-aware rewards, and evidence coverage. Yet the actions that expose supporting documents receive no targeted credit, a gap we call the reward-credit mismatch. We propose STAMP, in which a reference-based verifier judges whether each cited document supports an entity or relation in a training-time evidence graph, and first-exposure attribution traces each supported citation back to the action that first surfaced it. This step credit is injected through sign-preserving advantage modulation, which redistributes advantage across steps without changing the trajectory-level reward or the relative ranking of trajectories within each group. On BrowseComp, BrowseComp-ZH, and xbench-DS, STAMP improves the strongest GRPO baseline by +2.0/+5.5/+3.0 points under matched SFT initialization, training data, and search tools, and composes with both outcome-only and citation-rubric base rewards. Component ablations confirm that the provenance-based credit signal and the sign-preserving advantage modulation each contribute to the gains.

The rising demand for long-video understanding presents significant challenges due to massive information and complex temporal dependencies. Direct processing with Large Visual Language Models (VLMs) incurs high computational costs and risks information loss due to frame sampling. Meanwhile, existing agent-based methods often struggle with fragmented contexts and the high overhead of iterative processing. Therefore, we propose VideoNav, which introduces a new paradigm of index-centric navigation, reformulating video understanding as cost-aware structured query processing. VideoNav decouples heavy perception from online reasoning, including offline index construction and online adaptive navigation. In the offline phase, we transform unstructured video data into a structured, multi-granular index tree. During the online phase, the framework executes a top-down, coarse-to-fine navigating strategy. Leveraging the LLM agents, it efficiently locates relevant information and answers the query with navigating guidance, multi-level navigation, dual-path routing and visual inspection. Crucially, this adaptive mechanism enables early termination when high-level information suffices, minimizing redundant computation. Extensive experiments across different benchmarks demonstrate the effectiveness of our framework. VideoNav achieves decent accuracy while delivering 2.5x-5.7x cost savings and a 2.9x-5.8x speedup.

Reinforcement learning with verifiable rewards (RLVR) optimizes LLMs using sparse verifiable final-answer rewards. This sparse anchor reliably verifies whether a trajectory succeeds but provides no direct feedback on the reasoning path that produced it. Before success, prerequisite progress on hard problems receives no reward signal; after success, outcome rewards cannot distinguish well-organized correct trajectories from redundant or locally flawed ones. We introduce SCOPE-RL (Scaffolded Chain Optimization with Process Efficiency), a two-stage framework that densifies this anchor while retaining the GRPO update: Adaptive Scaffolded RL adds prefix-decomposed verifiable rewards on answer-hidden sub-question chains before success, and Quality-Aware Process RL applies correctness-gated process-shape rewards to refine correct trajectories after success. An expert-validated Step-Quality Evaluation Protocol evaluates useful-step density, error localization, and token efficiency beyond final-answer accuracy. On Qwen3-8B-Instruct trained on DAPO-Math and Big-Math, SCOPE-RL improves average accuracy by up to 11.2 pp and reduces reasoning tokens by up to 27.1% over outcome-only GRPO; the gains hold under GSPO and on Qwen3-0.6B-Instruct, indicating that reward-signal densification is complementary to policy-update-level RLVR advances.

Low-Rank Adaptation (LoRA) is widely used for parameter-efficient fine-tuning, yet merging multiple task-specific LoRA updates without additional training remains challenging. Most existing LoRA merging methods rely on SVD-based alignment, which emphasizes globally shared structure across tasks. In this work, we show that LoRA merging performance can be further improved by combining SVD with CUR decomposition. Through a representation-level analysis, we find that SVD-based decompositions primarily model shared components across tasks, while CUR-based decompositions better preserve task-specific and localized updates. These two perspectives are geometrically misaligned and exhibit complementary advantages, revealing an inherent trade-off between capturing shared structure and preserving task-specific information in LoRA model merging. Guided by this analysis, we propose a training-free merging procedure that explicitly combines the shared structure captured by SVD with the task-specific components preserved by CUR. Experiments on both vision and language benchmarks demonstrate consistent improvements over existing gradient-free LoRA merging methods.

Offline preference optimization aligns reasoning models from fixed chosen-rejected pairs, yet standard methods apply gradient updates from every pair regardless of its training value under the current policy. We argue that this uniform treatment is wasteful and potentially harmful. From the perspective of gradient utility, we show that a pair’s contribution depends jointly on informativeness and stability. Pair utility drifts as the policy evolves, high-gradient samples can coincide with high-curvature regions, leading to noisy and destabilizing updates, and the most effective supervision comes from stable confident errors where the model is reliably wrong yet curvature remains low. These findings motivate **SAGE** (Stability-Aware Gradient Efficiency), which maintains difficulty-stratified candidate pools refreshed during training and selects pairs within each pool by a forward-pass signal-to-curvature score. Only pairs with high current utility receive gradient computation; the rest are excluded from backpropagation. On mathematical reasoning benchmarks across multiple model scales, SAGE outperforms full-data and size-matched baselines while producing substantially smoother optimization trajectories.

On-policy distillation reduces train-inference mismatch by supervising student-generated trajectories, but full-vocabulary token-level distribution matching remains expensive for large language models. A common approximation truncates reverse-KL distillation to the student’s top-k tokens, which preserves efficiency but drops tail-token supervision and ignores support-level mass mismatch between the student and teacher. We analyze these approximation gaps and propose Stochastic Support Sampling with Teacher Mass Reweighting, which samples fixed-size supports from the student distribution and uses teacher support mass as a reliability weight for the conditional distillation loss. Under the same support budget, our method outperforms the strongest baseline by +1.4 hard-math average points on Qwen3-8B→Qwen3-1.7B-Base, and by +1.1 math average points and +2.8 code average points on DeepSeek-R1-Distill-Llama-8B→Meta-Llama-3.1-8B-Base.

The effectiveness of open-domain question answering (ODQA), particularly those employing a retriever-reader architecture, depends on the ability to recall relevant documents - a critical step that enables the reader to accurately extract answers. To enhance this retrieval phase, current query expansion (QE) techniques leverage pre-trained language models (PLM) to mitigate word mismatches and improve the recall of relevant documents. Despite their advancements, these techniques often treat all expanded terms uniformly, which can lead to less-than-optimal retrieval outcomes. In response, we propose a novel Recall-oriented Adaptive Learning (ReAL) method, which iteratively adjusts the importance weights of QE terms based on their relevance, thereby refining term distinction and enhancing the separation of relevant terms. Specifically, ReAL employs a similarity-based model to classify documents into pseudo-relevant and pseudo-irrelevant sets, and then optimizes term weights via two tailored loss functions to maximize the scoring gap between them. Experiments on four ODQA datasets and five QE methods show that ReAL consistently enhances retrieval accuracy and overall end-to-end QA performance, providing a robust and efficient solution for improving QE strategies in ODQA scenarios.

Query expansion (QE) is a critical component in the open-domain question answering (OpenQA) pipeline, enhancing the retrieval performance by broadening the scope of queries with additional relevant texts. However, existing methods like GAR and EAR rely heavily on supervised training and often struggle to maintain effectiveness across domains and datasets. Meanwhile, although large language models (LLMs) have demonstrated QE capability for information retrieval (IR) tasks, their application in OpenQA is hindered by the inadequate analysis of query’s informational needs and the lack of quality control for generated QEs, failing to meet the unique requirements of OpenQA. To bridge this gap, we propose a novel LLM-based QE approach named AGR for the OpenQA task, leveraging a three-step prompting strategy. AGR begins with an analysis of the query, followed by the generation of answer-oriented expansions, and culminates with a refinement process for better query formulation. Extensive experiments on four OpenQA datasets reveal that AGR not only rivals in-domain supervised methods in retrieval accuracy, but also outperforms state-of-the-art baselines in out-domain zero-shot scenarios. Moreover, it exhibits enhanced performance in end-to-end QA evaluations, underscoring the superiority of AGR for OpenQA.