arXiv subjects

Wenfeng Feng

Publications and source records attributed to Wenfeng Feng.

13 recordsLinked to original sources

arXiv2026

Knowing When Not to Reuse: Conditional Experience Transfer in Autonomous LLM Post-Training

Large language models offer broad capabilities, but adapting them to evolving domains, tools, and requirements often entails repeated post-training. Autonomous systems automate parts of this process by proposing updates, training candidates, and using evaluation feedback to select subsequent proposals. As evidence accumulates, a central problem emerges: which past update evidence remains actionable after subsequent training has changed the parent model? An update's effect depends on its parent, data, and training stage. Treating past success as context-free permission can waste compute. If the resulting child is promoted, it can also degrade the subsequent training trajectory. We formulate this problem as conditional experience transfer and introduce Boundary-Calibrated Intervention Transfer (BCIT), a method that authorizes experience reuse before weight-changing training. BCIT binds an observed effect to its source context, checks applicability conditions, vetoes candidates with named hard conflicts, and obtains current-state evidence through a bounded training trial when needed. Fully trained candidates still face a shared adoption rule, and only observed events extend memory. On one 4B model adapted across finance reasoning, text-to-SQL, and function calling, candidate updates exhibit heterogeneous target and retention effects across the evaluated contexts. Under matched candidates, evidence, and compute, BCIT authorizes fewer harmful updates and attains higher equal-budget final-model quality than the evaluated alternatives. These results support treating experience authorization as a distinct problem in autonomous post-training.

cs.AI↗

arXiv2026

Beyond Stochastic Exploration: What Makes Training Data Valuable for Agentic Search

Reinforcement learning (RL) has become an effective approach for advancing the reasoning capabilities of large language models (LLMs) through the strategic integration of external search engines. However, current RL-based search agents often rely on a process of stochastic exploration guided by carefully crafted outcome rewards, leading to inefficient reasoning trajectories and unstable training. To address these issues, we propose a novel framework, Hierarchical Experience (HiExp), to enhance the performance and training stability of search agents. Specifically, we extract empirical knowledge through contrastive analysis and a multi-level clustering mechanism, transforming raw reasoning trajectories into hierarchical experience knowledge. By leveraging experience-aligned training, we effectively regularize stochastic exploration, evolving it into a strategic and experience-driven search process. Extensive evaluations on multiple complex agentic search and mathematical reasoning benchmarks demonstrate that our approach not only achieves substantial performance gains but also exhibits strong cross-task and cross-algorithm generalization.

cs.AI↗

arXiv2025

EDIT: Enhancing Vision Transformers by Mitigating Attention Sink through an Encoder-Decoder Architecture

In this paper, we propose EDIT (Encoder-Decoder Image Transformer), a novel architecture designed to mitigate the attention sink phenomenon observed in Vision Transformer models. Attention sink occurs when an excessive amount of attention is allocated to the [CLS] token, distorting the model's ability to effectively process image patches. To address this, we introduce a layer-aligned encoder-decoder architecture, where the encoder utilizes self-attention to process image patches, while the decoder uses cross-attention to focus on the [CLS] token. Unlike traditional encoder-decoder framework, where the decoder depends solely on high-level encoder representations, EDIT allows the decoder to extract information starting from low-level features, progressively refining the representation layer by layer. EDIT is naturally interpretable demonstrated through sequential attention maps, illustrating the refined, layer-by-layer focus on key image features. Experiments on ImageNet-1k and ImageNet-21k, along with transfer learning tasks, show that EDIT achieves consistent performance improvements over DeiT3 models. These results highlight the effectiveness of EDIT's design in addressing attention sink and improving visual feature extraction.

cs.CV↗

arXiv2025

DynaSearcher: Dynamic Knowledge Graph Augmented Search Agent via Multi-Reward Reinforcement Learning

Multi-step agentic retrieval systems based on large language models (LLMs) have demonstrated remarkable performance in complex information search tasks. However, these systems still face significant challenges in practical applications, particularly in generating factually inconsistent intermediate queries and inefficient search trajectories, which can lead to reasoning deviations or redundant computations. To address these issues, we propose DynaSearcher, an innovative search agent enhanced by dynamic knowledge graphs and multi-reward reinforcement learning (RL). Specifically, our system leverages knowledge graphs as external structured knowledge to guide the search process by explicitly modeling entity relationships, thereby ensuring factual consistency in intermediate queries and mitigating biases from irrelevant information. Furthermore, we employ a multi-reward RL framework for fine-grained control over training objectives such as retrieval accuracy, efficiency, and response quality. This framework promotes the generation of high-quality intermediate queries and comprehensive final answers, while discouraging unnecessary exploration and minimizing information omissions or redundancy. Experimental results demonstrate that our approach achieves state-of-the-art answer accuracy on six multi-hop question answering datasets, matching frontier LLMs while using only small-scale models and limited computational resources. Furthermore, our approach demonstrates strong generalization and robustness across diverse retrieval environments and larger-scale models, highlighting its broad applicability.

cs.LG↗

arXiv2025

VCRL: Variance-based Curriculum Reinforcement Learning for Large Language Models

Policy-based reinforcement learning currently plays an important role in improving LLMs on mathematical reasoning tasks. However, existing rollout-based reinforcement learning methods (GRPO, DAPO, GSPO, etc.) fail to explicitly consider LLMs' learning ability for samples of different difficulty levels, which is contrary to the human cognitive process of mathematical reasoning tasks from easy to difficult. Intuitively, we find that the variance of the rollout group's reward in RLVR partly reflects the difficulty of the current sample for LLMs. Samples that are too easy or too difficult have a lower variance, while samples with moderate difficulty have a higher variance. Based on this, we propose VCRL, a curriculum reinforcement learning framework that dynamically controls the difficulty of training samples based on the variance of group rewards. Experiments on five mathematical benchmarks and two models reveal the advantages of VCRL over the current LLM RL baselines.

cs.LG↗

arXiv2025

PVPO: Pre-Estimated Value-Based Policy Optimization for Agentic Reasoning

Critic-free reinforcement learning methods, particularly group policies, have attracted considerable attention for their efficiency in complex tasks. However, these methods rely heavily on multiple sampling and comparisons within the policy to estimate advantage, which may cause the policy to fall into local optimum and increase computational cost. To address these issues, we propose PVPO, an efficient reinforcement learning method enhanced by an advantage reference anchor and data pre-sampling. Specifically, we use the reference model to rollout in advance and employ the calculated reward score as a reference anchor. Our approach effectively corrects the cumulative bias introduced by intra-group comparisons and significantly reduces reliance on the number of rollouts during training. Meanwhile, the reference model can assess sample difficulty during data pre-sampling, enabling effective selection of high-gain data to improve training efficiency. Moreover, PVPO is orthogonal to other advanced critic-free RL algorithms, making it compatible with and complementary to these methods. Experiments conducted on nine datasets across two domains demonstrate that PVPO achieves State-Of-The-Art (SOTA) performance. Our approach not only demonstrates robust generalization across multiple tasks, but also exhibits scalable performance across models of varying scales.

cs.LG↗

arXiv2025

AirRAG: Autonomous Strategic Planning and Reasoning Steer Retrieval Augmented Generation

Leveraging the autonomous decision-making capabilities of large language models (LLMs) has demonstrated superior performance in reasoning tasks. However, despite the success of iterative or agentic retrieval-augmented generation (RAG) techniques, these methods are often constrained to a single solution space when confronted with complex problems. In this paper, we propose a novel thinking pattern in RAG that integrates autonomous strategic planning with efficient reasoning actions, significantly activating intrinsic reasoning capabilities and expanding the solution space of specific tasks via Monte Carlo Tree Search (MCTS), which we refer to as AirRAG. Specifically, our approach designs five fundamental reasoning actions, which are expanded to a broad tree-based reasoning space using MCTS. The approach also incorporates self-consistency verification to explore potential reasoning paths and inference scaling law. Additionally, computationally optimal strategies are employed to allocate more inference resources to key actions, thereby enhancing overall performance. Experimental results demonstrate the effectiveness of AirRAG, showing significant performance gains on complex question-answering datasets. Furthermore, AirRAG is flexible and lightweight, making it easy to integrate with other advanced technologies and models.

cs.AI↗

arXiv2025

RASD: Retrieval-Augmented Speculative Decoding

Speculative decoding accelerates inference in large language models (LLMs) by generating draft tokens for target model verification. Current approaches for obtaining draft tokens rely on lightweight draft models or additional model structures to generate draft tokens and retrieve context from databases. Due to the draft model's small size and limited training data, model-based speculative decoding frequently becomes less effective in out-of-domain scenarios. Additionally, the time cost of the drafting phase results in a low upper limit on acceptance length during the verification step, limiting overall efficiency. This paper proposes RASD (Retrieval-Augmented Speculative Decoding), which adopts retrieval methods to enhance model-based speculative decoding. We introduce tree pruning and tree fusion to achieve this. Specifically, we develop a pruning method based on the draft model's probability distribution to construct the optimal retrieval tree. Second, we employ the longest prefix matching algorithm to merge the tree generated by the draft model with the retrieval tree, resulting in a unified tree for verification. Experimental results demonstrate that RASD achieves state-of-the-art inference acceleration across tasks such as DocQA, Summary, Code, and In-Domain QA. Moreover, RASD exhibits strong scalability, seamlessly integrating with various speculative decoding approaches, including both generation-based and retrieval-based methods.

cs.CL↗

arXiv2024

Mixture-of-LoRAs: An Efficient Multitask Tuning for Large Language Models

Instruction Tuning has the potential to stimulate or enhance specific capabilities of large language models (LLMs). However, achieving the right balance of data is crucial to prevent catastrophic forgetting and interference between tasks. To address these limitations and enhance training flexibility, we propose the Mixture-of-LoRAs (MoA) architecture which is a novel and parameter-efficient tuning method designed for multi-task learning with LLMs. In this paper, we start by individually training multiple domain-specific LoRA modules using corresponding supervised corpus data. These LoRA modules can be aligned with the expert design principles observed in Mixture-of-Experts (MoE). Subsequently, we combine the multiple LoRAs using an explicit routing strategy and introduce domain labels to facilitate multi-task learning, which help prevent interference between tasks and ultimately enhances the performance of each individual task. Furthermore, each LoRA model can be iteratively adapted to a new domain, allowing for quick domain-specific adaptation. Experiments on diverse tasks demonstrate superior and robust performance, which can further promote the wide application of domain-specific LLMs.

cs.CL↗

arXiv2020

Federated Deep Reinforcement Learning

In deep reinforcement learning, building policies of high-quality is challenging when the feature space of states is small and the training data is limited. Despite the success of previous transfer learning approaches in deep reinforcement learning, directly transferring data or models from an agent to another agent is often not allowed due to the privacy of data and/or models in many privacy-aware applications. In this paper, we propose a novel deep reinforcement learning framework to federatively build models of high-quality for agents with consideration of their privacies, namely Federated deep Reinforcement Learning (FedRL). To protect the privacy of data and models, we exploit Gausian differentials on the information shared with each other when updating their local models. In the experiment, we evaluate our FedRL framework in two diverse domains, Grid-world and Text2Action domains, by comparing to various baselines.

cs.LG↗

arXiv2019

ResNetX: a more disordered and deeper network architecture

Designing efficient network structures has always been the core content of neural network research. ResNet and its variants have proved to be efficient in architecture. However, how to theoretically character the influence of network structure on performance is still vague. With the help of techniques in complex networks, We here provide a natural yet efficient extension to ResNet by folding its backbone chain. Our architecture has two structural features when being mapped to directed acyclic graphs: First is a higher degree of the disorder compared with ResNet, which let ResNetX explore a larger number of feature maps with different sizes of receptive fields. Second is a larger proportion of shorter paths compared to ResNet, which improves the direct flow of information through the entire network. Our architecture exposes a new dimension, namely "fold depth", in addition to existing dimensions of depth, width, and cardinality. Our architecture is a natural extension to ResNet, and can be integrated with existing state-of-the-art methods with little effort. Image classification results on CIFAR-10 and CIFAR-100 benchmarks suggested that our new network architecture performs better than ResNet.

cs.CV↗

arXiv2018

Extracting Action Sequences from Texts Based on Deep Reinforcement Learning

Extracting action sequences from natural language texts is challenging, as it requires commonsense inferences based on world knowledge. Although there has been work on extracting action scripts, instructions, navigation actions, etc., they require that either the set of candidate actions be provided in advance, or that action descriptions are restricted to a specific form, e.g., description templates. In this paper, we aim to extract action sequences from texts in free natural language, i.e., without any restricted templates, provided the candidate set of actions is unknown. We propose to extract action sequences from texts based on the deep reinforcement learning framework. Specifically, we view "selecting" or "eliminating" words from texts as "actions", and the texts associated with actions as "states". We then build Q-networks to learn the policy of extracting actions and extract plans from the labeled texts. We demonstrate the effectiveness of our approach on several datasets with comparison to state-of-the-art approaches, including online experiments interacting with humans.

cs.AI↗

arXiv2014

Climatic seasonality may affect ecological network structure: Food webs and mutualistic networks

Ecological networks exhibit non-random structural patterns, such as modularity and nestedness, which indicate ecosystem stability, species diversity, and connectance. Such structure-stability relationships are well known. However, another important perspective is less well understood: the relationship between the environment and structure. Inspired by theoretical studies that suggest that network structure can change due to environmental variability, we collected data on a number of empirical food webs and mutualistic networks and evaluated the effect of climatic seasonality on ecological network structure. As expected, we found that climatic seasonality affects ecological network structure. In particular, an increase in modularity due to climatic seasonality was observed in food webs; however, it is debatable whether this occurs in mutualistic networks. Interestingly, the type of climatic seasonality that affects network structure differs with ecosystem type. Rainfall and temperature seasonality influence freshwater food webs and mutualistic networks, respectively; food webs are smaller, and more modular, with increasing rainfall seasonality. Mutualistic networks exhibit a higher diversity (particularly of animals) with increasing temperature seasonality. These results confirm the theoretical prediction that stability increases with greater perturbation. Although these results are still debatable because of several limitations in the data analysis, they may enhance our understanding of environment-structure relationships.

q-bio.PE↗