Researcher profile

Liang Pang

Liang Pang contributes to research discovery and scholarly infrastructure.

ResearcherAffiliation not importedOpen to collaborate

Computation and Language Artificial Intelligence Machine Learning Information Retrieval

Trust snapshot

Quick read

Trust 21 - EmergingVerification L1Unclaimed author

10works

0followers

4topics

4close collaborators

Actions

Decide how to stay connected

Follow researcher0

Claiming links this public author record to a researcher profile and unlocks direct collaboration workflows.

Direct collaboration

Open a focused conversation when the fit is right

Claim this author entity first to unlock direct invitations.

Inspect adjacent work, topics, institutions and collaborators without jumping out to a separate graph page.

Building this graph slice

BZPEER is loading the nearby papers, people, topics and institutions for this page.

preprint2026arXiv

Circular Reasoning: Understanding Self-Reinforcing Loops in Large Reasoning Models

Despite the success of test-time scaling, Large Reasoning Models (LRMs) frequently encounter repetitive loops that lead to computational waste and inference failure. In this paper, we identify a distinct failure mode termed Circular Reasoning. Unlike traditional model degeneration, this phenomenon manifests as a self-reinforcing trap where generated content acts as a logical premise for its own recurrence, compelling the reiteration of preceding text. To systematically analyze this phenomenon, we introduce LoopBench, a dataset designed to capture two distinct loop typologies: numerical loops and statement loops. Mechanistically, we characterize circular reasoning as a state collapse exhibiting distinct boundaries, where semantic repetition precedes textual repetition. We reveal that reasoning impasses trigger the loop onset, which subsequently persists as an inescapable cycle driven by a self-reinforcing V-shaped attention mechanism. Guided by these findings, we employ the Cumulative Sum (CUSUM) algorithm to capture these precursors for early loop prediction. Experiments across diverse LRMs validate its accuracy and elucidate the stability of long-chain reasoning.

preprint2026arXiv

Latent-GRPO: Group Relative Policy Optimization for Latent Reasoning

Latent reasoning offers a more efficient alternative to explicit reasoning by compressing intermediate reasoning into continuous representations and substantially shortening reasoning chains. However, existing latent reasoning methods mainly focus on supervised learning, and reinforcement learning in latent space remains highly unstable. We study this problem through the lens of Group Relative Policy Optimization (GRPO), and show that directly adapting GRPO to latent reasoning is fundamentally non-trivial: latent reasoning changes both the probability density and the sampling mechanism, causing three coupled bottlenecks: absence of intrinsic latent manifolds, where unconstrained exploration pushes rollouts off the valid latent manifold; exploration-optimization misalignment, where trajectory-level rewards can induce incorrect token-level updates; and latent mixture non-closure, where jointly reinforcing multiple correct latent paths can produce an invalid averaged state. To address them, we propose \textbf{Latent-GRPO}, which combines invalid-sample advantage masking, one-sided noise sampling, and optimal correct-path first-token selection. Across four low-difficulty benchmarks (e.g., GSM8K-Aug) and four high-difficulty benchmarks (e.g., AIME), Latent-GRPO improves over its latent initialization by 7.86 Pass@1 points on low-difficulty tasks and surpasses explicit GRPO by 4.27 points on high-difficulty tasks while using 3--4$\times$ shorter reasoning chains. It also achieves stronger pass@$k$ performance under Gumbel sampling. These results establish Latent-GRPO as an effective approach for stable and efficient latent reasoning.

preprint2026arXiv

Projecting Out the Malice: A Global Subspace Approach to LLM Detoxification

Large language models (LLMs) exhibit exceptional performance but pose inherent risks of generating toxic content, restricting their safe deployment. While traditional methods (e.g., alignment) adjust output preferences, they fail to eliminate underlying toxic regions in parameters, leaving models vulnerable to adversarial attacks. Prior mechanistic studies characterize toxic regions as "toxic vectors" or "layer-wise subspaces", yet our analysis identifies critical limitations: i) Removed toxic vectors can be reconstructed via linear combinations of non-toxic vectors, demanding targeting of entire toxic subspace; ii) Contrastive objective over limited samples inject noise into layer-wise subspaces, hindering stable extraction. These highlight the challenge of identifying robust toxic subspace and removing them. Therefore, we propose GLOSS (GLobal tOxic Subspace Suppression), a lightweight method that mitigates toxicity by identifying and eliminating this global subspace from FFN parameters. Experiments on LLMs (e.g., Qwen3) show GLOSS achieves SOTA detoxification while preserving general capabilities without requiring large-scale retraining. WARNING: This paper contains context which is toxic in nature.

preprint2026arXiv

The Evolution of Thought: Tracking LLM Overthinking via Reasoning Dynamics Analysis

Test-time scaling via explicit reasoning trajectories significantly boosts large language model (LLM) performance but often triggers overthinking. To explore this, we analyze reasoning through two lenses: Reasoning Length Dynamics, which reveals a compensatory trade-off between thinking and answer content length that eventually leads to thinking redundancy, and Reasoning Semantic Dynamics, which identifies semantic convergence and repetitive oscillations. These dynamics uncover an instance-specific Reasoning Completion Point (RCP), beyond which computation continues without further performance gain. Since the RCP varies across instances, we propose a Reasoning Completion Point Detector (RCPD), an inference-time early-exit method that identifies the RCP by monitoring the rank dynamics of termination tokens (e.g., </think>). Across AIME and GPQA benchmarks using Qwen3 and DeepSeek-R1, RCPD reduces token usage by up to 44% while preserving accuracy, offering a principled approach to efficient test-time scaling.

preprint2025arXiv

Large Language Model Sourcing: A Survey

Due to the black-box nature of large language models (LLMs) and the realism of their generated content, issues such as hallucinations, bias, unfairness, and copyright infringement have become significant. In this context, sourcing information from multiple perspectives is essential. This survey presents a systematic investigation organized around four interrelated dimensions: Model Sourcing, Model Structure Sourcing, Training Data Sourcing, and External Data Sourcing. Moreover, a unified dual-paradigm taxonomy is proposed that classifies existing sourcing methods into prior-based (proactive traceability embedding) and posterior-based (retrospective inference) approaches. Traceability across these dimensions enhances the transparency, accountability, and trustworthiness of LLMs deployment in real-world applications.

preprint2022arXiv

LoL: A Comparative Regularization Loss over Query Reformulation Losses for Pseudo-Relevance Feedback

Pseudo-relevance feedback (PRF) has proven to be an effective query reformulation technique to improve retrieval accuracy. It aims to alleviate the mismatch of linguistic expressions between a query and its potential relevant documents. Existing PRF methods independently treat revised queries originating from the same query but using different numbers of feedback documents, resulting in severe query drift. Without comparing the effects of two different revisions from the same query, a PRF model may incorrectly focus on the additional irrelevant information increased in the more feedback, and thus reformulate a query that is less effective than the revision using the less feedback. Ideally, if a PRF model can distinguish between irrelevant and relevant information in the feedback, the more feedback documents there are, the better the revised query will be. To bridge this gap, we propose the Loss-over-Loss (LoL) framework to compare the reformulation losses between different revisions of the same query during training. Concretely, we revise an original query multiple times in parallel using different amounts of feedback and compute their reformulation losses. Then, we introduce an additional regularization loss on these reformulation losses to penalize revisions that use more feedback but gain larger losses. With such comparative regularization, the PRF model is expected to learn to suppress the extra increased irrelevant information by comparing the effects of different revised queries. Further, we present a differentiable query reformulation method to implement this framework. This method revises queries in the vector space and directly optimizes the retrieval performance of query vectors, applicable for both sparse and dense retrieval models. Empirical evaluation demonstrates the effectiveness and robustness of our method for two typical sparse and dense retrieval models.

preprint2022arXiv

Match-Prompt: Improving Multi-task Generalization Ability for Neural Text Matching via Prompt Learning

Text matching is a fundamental technique in both information retrieval and natural language processing. Text matching tasks share the same paradigm that determines the relationship between two given texts. The relationships vary from task to task, e.g.~relevance in document retrieval, semantic alignment in paraphrase identification and answerable judgment in question answering. However, the essential signals for text matching remain in a finite scope, i.e.~exact matching, semantic matching, and inference matching. Ideally, a good text matching model can learn to capture and aggregate these signals for different matching tasks to achieve competitive performance, while recent state-of-the-art text matching models, e.g.~Pre-trained Language Models (PLMs), are hard to generalize. It is because the end-to-end supervised learning on task-specific dataset makes model overemphasize the data sample bias and task-specific signals instead of the essential matching signals. To overcome this problem, we adopt a specialization-generalization training strategy and refer to it as Match-Prompt. In specialization stage, descriptions of different matching tasks are mapped to a few prompt tokens. In generalization stage, matching model explores the essential matching signals by being trained on diverse matching tasks. High diverse matching tasks avoid model fitting the data bias on a specific task, so that model can focus on learning the essential matching signals. Meanwhile, the prompt tokens obtained in the first step help the model distinguish different task-specific matching signals. Experimental results on public datasets show that Match-Prompt can improve multi-task generalization capability of PLMs in text matching and yield better in-domain multi-task, out-of-domain multi-task and new task adaptation performance than multi-task and task-specific models trained by previous fine-tuning paradigm.

preprint2020arXiv

Ranking Enhanced Dialogue Generation

How to effectively utilize the dialogue history is a crucial problem in multi-turn dialogue generation. Previous works usually employ various neural network architectures (e.g., recurrent neural networks, attention mechanisms, and hierarchical structures) to model the history. However, a recent empirical study by Sankar et al. has shown that these architectures lack the ability of understanding and modeling the dynamics of the dialogue history. For example, the widely used architectures are insensitive to perturbations of the dialogue history, such as words shuffling, utterances missing, and utterances reordering. To tackle this problem, we propose a Ranking Enhanced Dialogue generation framework in this paper. Despite the traditional representation encoder and response generation modules, an additional ranking module is introduced to model the ranking relation between the former utterance and consecutive utterances. Specifically, the former utterance and consecutive utterances are treated as query and corresponding documents, and both local and global ranking losses are designed in the learning process. In this way, the dynamics in the dialogue history can be explicitly captured. To evaluate our proposed models, we conduct extensive experiments on three public datasets, i.e., bAbI, PersonaChat, and JDC. Experimental results show that our models produce better responses in terms of both quantitative measures and human judgments, as compared with the state-of-the-art dialogue generation models. Furthermore, we give some detailed experimental analysis to show where and how the improvements come from.

preprint2020arXiv

Robust Reinforcement Learning with Wasserstein Constraint

Robust Reinforcement Learning aims to find the optimal policy with some extent of robustness to environmental dynamics. Existing learning algorithms usually enable the robustness through disturbing the current state or simulating environmental parameters in a heuristic way, which lack quantified robustness to the system dynamics (i.e. transition probability). To overcome this issue, we leverage Wasserstein distance to measure the disturbance to the reference transition kernel. With Wasserstein distance, we are able to connect transition kernel disturbance to the state disturbance, i.e. reduce an infinite-dimensional optimization problem to a finite-dimensional risk-aware problem. Through the derived risk-aware optimal Bellman equation, we show the existence of optimal robust policies, provide a sensitivity analysis for the perturbations, and then design a novel robust learning algorithm--Wasserstein Robust Advantage Actor-Critic algorithm (WRAAC). The effectiveness of the proposed algorithm is verified in the Cart-Pole environment.

preprint2020arXiv

SetRank: Learning a Permutation-Invariant Ranking Model for Information Retrieval

In learning-to-rank for information retrieval, a ranking model is automatically learned from the data and then utilized to rank the sets of retrieved documents. Therefore, an ideal ranking model would be a mapping from a document set to a permutation on the set, and should satisfy two critical requirements: (1)~it should have the ability to model cross-document interactions so as to capture local context information in a query; (2)~it should be permutation-invariant, which means that any permutation of the inputted documents would not change the output ranking. Previous studies on learning-to-rank either design uni-variate scoring functions that score each document separately, and thus failed to model the cross-document interactions; or construct multivariate scoring functions that score documents sequentially, which inevitably sacrifice the permutation invariance requirement. In this paper, we propose a neural learning-to-rank model called SetRank which directly learns a permutation-invariant ranking model defined on document sets of any size. SetRank employs a stack of (induced) multi-head self attention blocks as its key component for learning the embeddings for all of the retrieved documents jointly. The self-attention mechanism not only helps SetRank to capture the local context information from cross-document interactions, but also to learn permutation-equivariant representations for the inputted documents, which therefore achieving a permutation-invariant ranking model. Experimental results on three large scale benchmarks showed that the SetRank significantly outperformed the baselines include the traditional learning-to-rank models and state-of-the-art Neural IR models.

Institution

Affiliation not imported yet

This author record came from a source that does not expose affiliation metadata. Once the author claims the profile or we enrich the record from another provider, this section will link to the concrete institution.

Topic footprint