arXiv subjects

Yujie Feng

Publications and source records attributed to Yujie Feng.

At least 19 recordsLinked to original sources

arXiv2026

ROSD: Reflective On-Policy Self-Distillation for Language Model Reasoning across Domains

On-policy self-distillation (OPSD) improves the reasoning performance of large language models (LLMs) by providing dense token-level supervision for on-policy rollouts. However, existing OPSD methods often yield limited gains on in-domain reasoning and generalize poorly to out-of-domain problems. We identify two key causes: conditioning the self-teacher on a verified solution encourages imitation of training-domain reference trajectories rather than error-specific correction, and applying distillation to the full response can overwrite valid reasoning prefixes and reinforce overfitting. We propose Reflective On-policy Self-Distillation (ROSD), a framework that turns reference-solution imitation into targeted reasoning correction through reflection-guided, error-localized distillation. For each rollout, ROSD uses a self-reflector to extract a corrective idea and locate the first erroneous span. The corrective idea guides the self-teacher toward targeted supervision, while the localized error span restricts distillation to where correction is needed. This design corrects flawed reasoning while preserving valid prefixes. Experiments on multiple in-domain and out-of-domain reasoning benchmarks show that ROSD yields stronger in-domain reasoning performance overall and substantially better out-of-domain generalization than standard OPSD. Code is available at https://github.com/ZiqiZhao1/ROSD.

cs.CL↗

arXiv2026

FOREVER: Forgetting Curve-Inspired Memory Replay for Language Model Continual Learning

Continual learning (CL) for large language models (LLMs) aims to enable sequential knowledge acquisition without catastrophic forgetting. Memory replay methods are widely used for their practicality and effectiveness, but most rely on fixed, step-based heuristics that often misalign with the model's actual learning progress, since identical training steps can result in varying degrees of parameter change. Motivated by recent findings that LLM forgetting mirrors the Ebbinghaus human forgetting curve, we propose FOREVER (FORgEtting curVe-inspired mEmory Replay), a novel CL framework that aligns replay schedules with a model-centric notion of time. FOREVER defines model time using the magnitude of optimizer updates, allowing forgetting curve-inspired replay intervals to align with the model's internal evolution rather than raw training steps. Building on this approach, FOREVER incorporates a forgetting curve-based replay scheduler to determine when to replay and an intensity-aware regularization mechanism to adaptively control how to replay. Extensive experiments on three CL benchmarks and models ranging from 0.6B to 13B parameters demonstrate that FOREVER consistently mitigates catastrophic forgetting.

cs.LG↗

arXiv2026

Micro-Macro Retrieval: Reducing Long-Form Hallucination in Large Language Models

Large Language Models (LLMs) achieve impressive performance across many tasks but remain prone to hallucination, especially in long-form generation where redundant retrieved contexts and lengthy reasoning chains amplify factual errors. Recent studies highlight a critical phenomenon: the closer key information appears to the model outputs, the higher the factual accuracy. However, existing retrieval-augmented language models (RALMs) lack effective mechanisms to ensure this proximity - external evidence is injected into reasoning via multi-turn retrieval, but this cannot ensure key information stays close to the outputs. We propose Micro-Macro Retrieval (M2R), a novel retrieve-while-generate framework to fill this gap. At the macro level, M2R retrieves coarse-grained evidence from external sources; at the micro level, it extracts essential results from a key information repository built during reasoning and reuses them while generating answers. This design directly addresses the key-information-to-output proximity bottleneck, effectively reducing hallucination in long-form tasks. M2R is trained with a curriculum learning-based reinforcement learning strategy using customized rule-based rewards, enabling stable acquisition of retrieval and grounding skills. Extensive experiments across different benchmarks demonstrate the effectiveness of M2R, especially in lengthy-context settings.

cs.CL↗

arXiv2026

How Order-Sensitive Are LLMs? OrderProbe for Deterministic Structural Reconstruction

Large language models (LLMs) excel at semantic understanding, yet their ability to reconstruct internal structure from scrambled inputs remains underexplored. Sentence-level restoration is difficult to evaluate automatically because scrambled sentences often admit multiple valid reorderings. We introduce OrderProbe, a deterministic benchmark for structural reconstruction using fixed four-character expressions in Chinese, Japanese, and Korean, which have a unique canonical order and thus support exact-match scoring. We further propose a diagnostic framework that evaluates models beyond recovery accuracy, including Semantic Accuracy, Logical Validity, Structural Consistency, Robustness, and Information Density. Experiments on twelve widely used LLMs show that structural reconstruction remains difficult even for frontier systems: zero-shot recovery frequently falls below 35%. We also observe a consistent gap between meaning-oriented generation and exact structural reconstruction, suggesting that structural robustness is not an automatic byproduct of semantic competence.

cs.CL↗

arXiv2025

AIMMerging: Adaptive Iterative Model Merging Using Training Trajectories for Language Model Continual Learning

Continual learning (CL) is essential for deploying large language models (LLMs) in dynamic real-world environments without the need for costly retraining. Recent model merging-based methods have attracted significant attention, but they still struggle to effectively manage the trade-off between learning new knowledge and preventing forgetting, a challenge largely stemming from suboptimal number of merges and merging frequency. In this paper, we introduce Adaptive Iterative Model Merging (AimMerging), a novel CL framework that utilizes learning and forgetting signals from the training trajectory to dynamically monitor the model's training status. Guided by dynamic monitoring, the training trajectory-guided merge controller adaptively determines the timing and frequency of iterative fusion, while the rehearsal-based knowledge fusion module computes the merging weights and executes the fusion. Comprehensive experiments on three CL benchmarks with various model sizes (from 770M to 13B) demonstrate that AimMerging achieves significant performance improvements over existing state-of-the-art methods, with an average relative improvement of 80% and 59% on FWT and BWT, respectively. The source code is provided for reproducibility.

cs.CL↗

arXiv2025

A Progressive Image Restoration Network for High-order Degradation Imaging in Remote Sensing

Recently, deep learning methods have gained remarkable achievements in the field of image restoration for remote sensing (RS). However, most existing RS image restoration methods focus mainly on conventional first-order degradation models, which may not effectively capture the imaging mechanisms of remote sensing images. Furthermore, many RS image restoration approaches that use deep learning are often criticized for their lacks of architecture transparency and model interpretability. To address these problems, we propose a novel progressive restoration network for high-order degradation imaging (HDI-PRNet), to progressively restore different image degradation. HDI-PRNet is developed based on the theoretical framework of degradation imaging, also Markov properties of the high-order degradation process and Maximum a posteriori (MAP) estimation, offering the benefit of mathematical interpretability within the unfolding network. The framework is composed of three main components: a module for image denoising that relies on proximal mapping prior learning, a module for image deblurring that integrates Neumann series expansion with dual-domain degradation learning, and a module for super-resolution. Extensive experiments demonstrate that our method achieves superior performance on both synthetic and real remote sensing images.

cs.CV↗

arXiv2025

Parenting: Optimizing Knowledge Selection of Retrieval-Augmented Language Models with Parameter Decoupling and Tailored Tuning

Retrieval-Augmented Generation (RAG) offers an effective solution to the issues faced by Large Language Models (LLMs) in hallucination generation and knowledge obsolescence by incorporating externally retrieved knowledge. However, existing methods lack effective control mechanisms for integrating internal and external knowledge. Inspired by human cognitive processes, we propose Parenting, a novel framework that decouples, identifies, and purposefully optimizes parameter subspaces related to adherence and robustness. Specifically, Parenting utilizes a key parameter mining method that combines forward and backward propagation signals to localize subspaces representing different capabilities. Then, Parenting employs a type-tailored tuning strategy, applying specific and appropriate optimizations to different subspaces, aiming to achieve a balanced enhancement of both adherence and robustness. Extensive experiments on various datasets and models validate the effectiveness and generalizability of our method.

cs.CL↗

arXiv2025

Influence Mechanism of Truncation on Low-Frequency Phase Measurement

Driven by advances in electronic technology, modern digital phasemeters have significantly improved in integration and functionality, enabling real-time measurement and analysis of dynamic signals. High-precision phase measurement is closely associated with the quantization process. This paper specifically analyzes the white and non-white noise characteristics associated with the quantization errors of phase truncation in digital phasemeters. The error can be considered white noise under specific conditions, which power correlates with the resolution of quantizer and is uniformly distributed within the Nyquist frequency. However, when the signal frequency and sampling frequency are close to an integer multiple, the non-white noise caused by truncation can result in low-frequency phase noise. Additionally, artifacts may induce nonlinear phase errors. Introducing Gaussian dither synthesized by LFSRs can smooth the truncation process, thereby mitigating its impacts on phase measurement. The results indicate that for a 10 MHz signal under test, the noise floor of the phasemeter exceeds the requirement from 2 mHz to 0.1 Hz due to the integer multiple. After adding dither, the phase noise was optimized by 9.5 dB at 10 mHz, achieving the requirement of 1.3 $\rm{\upmu rad/Hz^{1/2}} \cdot \rm{NSF}$ from 0.1 mHz to 1 Hz in space gravitational wave detection. This demonstrates that adding dither can effectively suppress the low-frequency phase noise caused by truncation.

physics.ins-det↗

arXiv2025

Recurrent Knowledge Identification and Fusion for Language Model Continual Learning

Continual learning (CL) is crucial for deploying large language models (LLMs) in dynamic real-world environments without costly retraining. While recent model ensemble and model merging methods guided by parameter importance have gained popularity, they often struggle to balance knowledge transfer and forgetting, mainly due to the reliance on static importance estimates during sequential training. In this paper, we present Recurrent-KIF, a novel CL framework for Recurrent Knowledge Identification and Fusion, which enables dynamic estimation of parameter importance distributions to enhance knowledge transfer. Inspired by human continual learning, Recurrent-KIF employs an inner loop that rapidly adapts to new tasks while identifying important parameters, coupled with an outer loop that globally manages the fusion of new and historical knowledge through redundant knowledge pruning and key knowledge merging. These inner-outer loops iteratively perform multiple rounds of fusion, allowing Recurrent-KIF to leverage intermediate training information and adaptively adjust fusion strategies based on evolving importance distributions. Extensive experiments on two CL benchmarks with various model sizes (from 770M to 13B) demonstrate that Recurrent-KIF effectively mitigates catastrophic forgetting and enhances knowledge transfer.

cs.LG↗

arXiv2025

GeoEdit: Geometric Knowledge Editing for Large Language Models

Regular updates are essential for maintaining up-to-date knowledge in large language models (LLMs). Consequently, various model editing methods have been developed to update specific knowledge within LLMs. However, training-based approaches often struggle to effectively incorporate new knowledge while preserving unrelated general knowledge. To address this challenge, we propose a novel framework called Geometric Knowledge Editing (GeoEdit). GeoEdit utilizes the geometric relationships of parameter updates from fine-tuning to differentiate between neurons associated with new knowledge updates and those related to general knowledge perturbations. By employing a direction-aware knowledge identification method, we avoid updating neurons with directions approximately orthogonal to existing knowledge, thus preserving the model's generalization ability. For the remaining neurons, we integrate both old and new knowledge for aligned directions and apply a "forget-then-learn" editing strategy for opposite directions. Additionally, we introduce an importance-guided task vector fusion technique that filters out redundant information and provides adaptive neuron-level weighting, further enhancing model editing performance. Extensive experiments on two publicly available datasets demonstrate the superiority of GeoEdit over existing state-of-the-art methods.

cs.CL↗

arXiv2025

Understanding Layer Significance in LLM Alignment

Aligning large language models (LLMs) through supervised fine-tuning is essential for tailoring them to specific applications. Recent studies suggest that alignment primarily adjusts a model's presentation style rather than its foundational knowledge, indicating that only certain components of the model are significantly impacted. To uncover how alignment affects model behavior at a granular level, we propose identifying which layers within LLMs are most critical to the alignment process. Our approach, named ILA, involves learning a binary mask for the parameter changes in each layer during alignment, as an indicator of layer significance. Experimental results reveal that, despite substantial differences in alignment datasets, the important layers of a model identified by ILA exhibit nearly 90\% overlap, highlighting fundamental patterns in LLM alignment. The results also indicate that freezing non-essential layers improves overall model performance, while selectively tuning the most critical layers significantly enhances fine-tuning efficiency with minimal performance loss. Finally, we discuss how these findings extend from LLM alignment to reasoning.

cs.CL↗

arXiv2025

KIF: Knowledge Identification and Fusion for Language Model Continual Learning

Language model continual learning (CL) has recently attracted significant interest for its ability to adapt large language models (LLMs) to dynamic real-world scenarios without retraining. A major challenge in this domain is catastrophic forgetting, where models lose previously acquired knowledge upon learning new tasks. Existing approaches commonly utilize multiple parameter-efficient fine-tuning (PEFT) blocks to acquire task-specific knowledge, yet these methods are inefficient and fail to leverage potential knowledge transfer across tasks. In this paper, we introduce a novel CL framework for language models, named Knowledge Identification and Fusion (KIF), which boosts knowledge transfer without depending on memory replay. KIF initially segregates the model into 'skill units' based on parameter dependencies, allowing for more precise control. Subsequently, it employs a novel group-wise knowledge identification technique to ascertain the importance distribution of skill units for a new task. By comparing this importance distribution with those from previous tasks, we implement a fine-grained knowledge fusion strategy that retains task-specific knowledge, thereby preventing forgetting, and updates task-shared knowledge, which facilitates bi-directional knowledge transfer. As a result, KIF achieves an optimal balance between retaining prior knowledge and excelling in new tasks. KIF also demonstrates strong generalizability, making it suitable for various base models and adaptable to PEFT methods like LoRA. Furthermore, it offers notable extensibility, supporting enhancements through integration with memory replay techniques. Comprehensive experiments conducted on two CL benchmarks, involving models ranging from 220M to 7B parameters, affirm the effectiveness of KIF and its variants across different settings.

cs.CL↗

arXiv2024

Medium recoil mode of $Δ$ production in single isobaric charge-exchange reactions

The dynamic mechanisms underlying single charge-exchange reactions have been investigated using a theoretical framework that combines the Isospin-dependent Quantum Molecular Dynamics (IQMD) model with the statistical decay model GEMINI++. Two distinct channels contribute to the single isobaric charge-exchange reaction: quasi-elastic channel, where neutron-proton scattering drives the charge-exchange, and inelastic channel, where the $Δ$ particle is produced during the process. In a referenced study [Phys.RevC 106.014618(2022)], experimental data have revealed that the inelastic channel accounts for approximately 50 percent of the single isobaric charge-exchange reaction. However, our current model fails in reproducing the significant contribution of inelastic channel unless the novel medium recoil mode associated with $Δ$ production is considered in the calculations. Notably, this in-medium effect arising from inelastic nucleon-nucleon collisions is not yet incorporated into mainstream microscopic transport models. The dynamical properties of protons and pions emitting in the single isobaric charge-exchange reactions are predicted. This exploration of in-medium effects adds a valuable dimension to our understanding of the intricate dynamics involved in single charge-exchange reactions.

nucl-th↗

arXiv2024

Continual Dialogue State Tracking via Reason-of-Select Distillation

An ideal dialogue system requires continuous skill acquisition and adaptation to new tasks while retaining prior knowledge. Dialogue State Tracking (DST), vital in these systems, often involves learning new services and confronting catastrophic forgetting, along with a critical capability loss termed the "Value Selection Quandary." To address these challenges, we introduce the Reason-of-Select (RoS) distillation method by enhancing smaller models with a novel 'meta-reasoning' capability. Meta-reasoning employs an enhanced multi-domain perspective, combining fragments of meta-knowledge from domain-specific dialogues during continual learning. This transcends traditional single-perspective reasoning. The domain bootstrapping process enhances the model's ability to dissect intricate dialogues from multiple possible values. Its domain-agnostic property aligns data distribution across different domains, effectively mitigating forgetting. Additionally, two novel improvements, "multi-value resolution" strategy and Semantic Contrastive Reasoning Selection method, significantly enhance RoS by generating DST-specific selection chains and mitigating hallucinations in teachers' reasoning, ensuring effective and reliable knowledge transfer. Extensive experiments validate the exceptional performance and robust generalization capabilities of our method. The source code is provided for reproducibility.

cs.CL↗

arXiv2024

Diversity-grounded Channel Prototypical Learning for Out-of-Distribution Intent Detection

In the realm of task-oriented dialogue systems, a robust intent detection mechanism must effectively handle malformed utterances encountered in real-world scenarios. This study presents a novel fine-tuning framework for large language models (LLMs) aimed at enhancing in-distribution (ID) intent classification and out-of-distribution (OOD) intent detection, which utilizes semantic matching with prototypes derived from ID class names. By harnessing the highly distinguishable representations of LLMs, we construct semantic prototypes for each ID class using a diversity-grounded prompt tuning approach. We rigorously test our framework in a challenging OOD context, where ID and OOD classes are semantically close yet distinct, referred to as \emph{near} OOD detection. For a thorough assessment, we benchmark our method against the prevalent fine-tuning approaches. The experimental findings reveal that our method demonstrates superior performance in both few-shot ID intent classification and near-OOD intent detection tasks.

cs.CL↗

arXiv2024

TaSL: Continual Dialog State Tracking via Task Skill Localization and Consolidation

A practical dialogue system requires the capacity for ongoing skill acquisition and adaptability to new tasks while preserving prior knowledge. However, current methods for Continual Dialogue State Tracking (DST), a crucial function of dialogue systems, struggle with the catastrophic forgetting issue and knowledge transfer between tasks. We present TaSL, a novel framework for task skill localization and consolidation that enables effective knowledge transfer without relying on memory replay. TaSL uses a novel group-wise technique to pinpoint task-specific and task-shared areas. Additionally, a fine-grained skill consolidation strategy protects task-specific knowledge from being forgotten while updating shared knowledge for bi-directional knowledge transfer. As a result, TaSL strikes a balance between preserving previous knowledge and excelling at new tasks. Comprehensive experiments on various backbones highlight the significant performance improvements of TaSL over existing state-of-the-art methods. The source code is provided for reproducibility.

cs.CL↗

arXiv2024

How Good Are LLMs at Out-of-Distribution Detection?

Out-of-distribution (OOD) detection plays a vital role in enhancing the reliability of machine learning (ML) models. The emergence of large language models (LLMs) has catalyzed a paradigm shift within the ML community, showcasing their exceptional capabilities across diverse natural language processing tasks. While existing research has probed OOD detection with relative small-scale Transformers like BERT, RoBERTa and GPT-2, the stark differences in scales, pre-training objectives, and inference paradigms call into question the applicability of these findings to LLMs. This paper embarks on a pioneering empirical investigation of OOD detection in the domain of LLMs, focusing on LLaMA series ranging from 7B to 65B in size. We thoroughly evaluate commonly-used OOD detectors, scrutinizing their performance in both zero-grad and fine-tuning scenarios. Notably, we alter previous discriminative in-distribution fine-tuning into generative fine-tuning, aligning the pre-training objective of LLMs with downstream tasks. Our findings unveil that a simple cosine distance OOD detector demonstrates superior efficacy, outperforming other OOD detectors. We provide an intriguing explanation for this phenomenon by highlighting the isotropic nature of the embedding spaces of LLMs, which distinctly contrasts with the anisotropic property observed in smaller BERT family models. The new insight enhances our understanding of how LLMs detect OOD data, thereby enhancing their adaptability and reliability in dynamic environments. We have released the source code at \url{https://github.com/Awenbocc/LLM-OOD} for other researchers to reproduce our results.

cs.CL↗

arXiv2024

Microscopic study of deformation and orientation effects in heavy-ion reactions above Coulomb barrier using the Boltzmann-Uehling-Uhlenbeck model

Background: The understanding of the impact of initial deformation and collision orientation on quasi-fission and fusion-fission reactions remains incomplete. Purpose: This article aims to explore how the orientation of deformed nuclei influences quasi-fission and fusion-fission around 1.2 VB, employing a micro dynamical method in systems with diverse shapes, namely 24Mg + 178Hf, 34S + 168Er, and 48Ti + 154Sm. Method: Utilizing the Boltzmann-Uehling-Uhlenbeck model, this study investigates quasi-fission and fusion fission reactions. The model elucidates micro-dynamic processes and microscopic observables through the definition of the window and event-by-event simulations. Results: The findings reveal that the orientation of deformed nuclei significantly influences the nucleus-nucleus interaction potential, thereby impacting the competition between quasi-fission and fusion-fission. Particularly, the orientation of the deformed target nucleus emerges as the primary factor affecting this competition. Notably, a higher proportion of fusion-fission events is observed when the target nucleus is in the belly orientation compared to the tip. The study also observes that the configuration of the dinuclear system contributes to fluctuations and dissipation. Collisions with different orientations result in distinct dinuclear system configurations, with belly-oriented collisions leading to larger fluctuations between events, while tip-oriented collisions exhibit smaller fluctuations. Conclusions: Considering diverse orientations of nuclei with distinct initial deformations, this study concludes that the orientation of the target nucleus is the key factor influencing quasi-fission and fusion-fission reactions around 1.2 VB.

nucl-th↗