Peilin Wu

Chen Gong, Peilin Wu, Jinho D. Choi

This paper presents a novel evaluation approach to text-based speaker diarization (SD), tackling the limitations of traditional metrics that do not account for any contextual information in text. Two new metrics are proposed, Text-based Diarization Error Rate and Diarization F1, which perform utterance- and word-level evaluations by aligning tokens in reference and hypothesis transcripts. Our metrics encompass more types of errors compared to existing ones, allowing us to make a more comprehensive analysis in SD. To align tokens, a multiple sequence alignment algorithm is introduced that supports multiple sequences in the reference while handling high-dimensional alignment to the hypothesis using dynamic programming. Our work is packaged into two tools, align4d providing an API for our alignment algorithm and TranscribeView for visualizing and evaluating SD errors, which can greatly aid in the creation of high-quality data, fostering the advancement of dialogue systems.

Kaiyu He, Mian Zhang, Shuo Yan et al.

While large language models (LLMs) have been thoroughly evaluated for deductive and inductive reasoning, their proficiency in holistic rule learning in interactive environments remains less explored. We introduce RULEARN, a novel benchmark to assess the rule-learning abilities of LLM agents in interactive settings. In RULEARN, agents strategically interact with simulated environments to gather observations, discern patterns, and solve complex problems. To enhance the rule-learning capabilities for LLM agents, we propose IDEA, a novel reasoning framework that integrates the process of Induction, Deduction, and Abduction. The IDEA agent generates initial hypotheses from limited observations through abduction, devises plans to validate these hypotheses or leverages them to solve problems via deduction, and refines previous hypotheses through induction, dynamically establishing and applying rules that mimic human rule-learning behaviors. Our evaluation of the IDEA framework, which involves five representative LLMs, demonstrates significant improvements over the baseline. Furthermore, our study with human participants reveals notable discrepancies in rule-learning behaviors between humans and LLMs. We believe our benchmark will serve as a valuable and challenging resource, and IDEA will provide crucial insights for the development of LLM agents capable of human-like rule learning in real-world scenarios. Our code and data is publicly available.

Peilin Wu, Xinlu Zhang, Kun Wan et al.

Rubric-based reward shaping is an effective method for fine-tuning LLMs via RL, where structured rubrics decompose standard outcome rewards into multiple dimensions to provide richer reward signals. Recent works make the rubrics adaptive based on local signals such as the rollouts from the current step or pairwise comparisons. However, these methods discard the diagnostics produced during evaluation after immediate use and prevent the long-term accumulation and strategic reuse of evaluation knowledge. This forces the system to re-derive evaluation principles from scratch, limits its ability to detect recurring suboptimal behaviors, and forfeits the curriculum-like progression that a persistent training history would naturally support. To address these limitations, we introduce AMARIS, which grounds rubric modifications in long-term training history. At each training step, AMARIS analyzes individual rollouts, aggregates findings into step-level summaries, retrieves relevant historical context from a persistent evaluation memory through both static (recent steps) and dynamic (semantically matched) retrieval, and updates rubrics based on these accumulated analyses. This procedure runs asynchronously alongside the normal RL loop with minimal overhead. Experiments across both closed and open-ended domains show that AMARIS consistently outperforms the baselines. Ablation studies show that static and dynamic memory retrieval contributes to the performance gain and their combination provides the strongest results with moderate retrieval budgets sufficient to provide most of the gain, and that the entire pipeline adds only ~5\% time overhead through asynchronous execution. These results show that persistent evaluation memory can transform rubric-based reward shaping from a stateless, per-step heuristic into an evidence-driven loop for RL training.

Peilin Wu, Weiji Xie, Jiahang Cao et al.

Reinforcement Learning (RL) has shown its remarkable and generalizable capability in legged locomotion through sim-to-real transfer. However, while adaptive methods like domain randomization are expected to enhance policy robustness across diverse environments, they potentially compromise the policy's performance in any specific environment, leading to suboptimal real-world deployment due to the No Free Lunch theorem. To address this, we propose LoopSR, a lifelong policy adaptation framework that continuously refines RL policies in the post-deployment stage. LoopSR employs a transformer-based encoder to map real-world trajectories into a latent space and reconstruct a digital twin of the real world for further improvement. Autoencoder architecture and contrastive learning methods are adopted to enhance feature extraction of real-world dynamics. Simulation parameters for continual training are derived by combining predicted values from the decoder with retrieved parameters from a pre-collected simulation trajectory dataset. By leveraging simulated continual training, LoopSR achieves superior data efficiency compared with strong baselines, yielding eminent performance with limited data in both sim-to-sim and sim-to-real experiments. Please refer to https://peilinwu.site/looping-sim-and-real.github.io/ for videos and code.

Peilin Wu, Xinlu Zhang, Wenhao Yu et al.

Recent advancements in Retrieval-Augmented Language Models (RALMs) have demonstrated their efficacy in knowledge-intensive tasks. However, existing evaluation benchmarks often assume a single optimal approach to leveraging retrieved information, failing to account for varying user needs. This paper introduces a novel evaluation framework that systematically assesses RALMs under three user need cases-Context-Exclusive, Context-First, and Memory-First-across three distinct context settings: Context Matching, Knowledge Conflict, and Information Irrelevant. By varying both user instructions and the nature of retrieved information, our approach captures the complexities of real-world applications where models must adapt to diverse user requirements. Through extensive experiments on multiple QA datasets, including HotpotQA, DisentQA, and our newly constructed synthetic URAQ dataset, we find that restricting memory usage improves robustness in adversarial retrieval conditions but decreases peak performance with ideal retrieval results and model family dominates behavioral differences. Our findings highlight the necessity of user-centric evaluations in the development of retrieval-augmented systems and provide insights into optimizing model performance across varied retrieval contexts. We will release our code and URAQ dataset upon acceptance of the paper.

Kaiyu He, Zhang Mian, Peilin Wu et al.

While Large Language Models (LLMs) excel at factual retrieval, they often struggle with the "curse of two-hop reasoning" in compositional tasks. Recent research suggests that parameter-sharing transformers can bridge this gap by forming a "Generalization Circuit" during a prolonged "grokking" phase. A fundamental question arises: Is a grokked model superior to its non-grokked counterparts on downstream tasks? Furthermore, is the extensive computational cost of waiting for the grokking phase worthwhile? In this work, we conduct a mechanistic study to evaluate the Generalization Circuit's role in knowledge assimilation and transfer. We demonstrate that: (i) The inference paths established by non-grokked and grokked models for in-distribution compositional queries are identical. This suggests that the "Generalization Circuit" does not represent the sudden acquisition of a new reasoning paradigm. Instead, we argue that grokking is the process of integrating memorized atomic facts into an naturally established reasoning path. (ii) Achieving high accuracy on unseen cases after prolonged training and the formation of a certain reasoning path are not bound; they can occur independently under specific data regimes. (iii) Even a mature circuit exhibits limited transferability when integrating new knowledge, suggesting that "grokked" Transformers do not achieve a full mastery of compositional logic.

Peilin Wu, Mian Zhang, Xinlu Zhang et al.

Agentic Retrieval-Augmented Generation (RAG) systems enhance Large Language Models (LLMs) by enabling dynamic, multi-step reasoning and information retrieval. However, these systems often exhibit sub-optimal search behaviors like over-search (retrieving redundant information) and under-search (failing to retrieve necessary information), which hinder efficiency and reliability. This work formally defines and quantifies these behaviors, revealing their prevalence across multiple QA datasets and agentic RAG systems (e.g., one model could have avoided searching in 27.7% of its search steps). Furthermore, we demonstrate a crucial link between these inefficiencies and the models' uncertainty regarding their own knowledge boundaries, where response accuracy correlates with model's uncertainty in its search decisions. To address this, we propose $β$-GRPO, a reinforcement learning-based training method that incorporates confidence threshold to reward high-certainty search decisions. Experiments on seven QA benchmarks show that $β$-GRPO enable a 3B model with better agentic RAG ability, outperforming other strong baselines with a 4% higher average exact match score.

Shuo Yan, Ruochen Li, Ziming Luo et al. · microsoft-research

Large language model (LLM) agents have demonstrated remarkable potential in advancing scientific discovery. However, their capability in the fundamental yet crucial task of reproducing code from research papers, especially in the NLP domain, remains underexplored. This task includes unique complex reasoning challenges in the intellectual synthesis of abstract concepts and the comprehension of code repositories with interdependent files. Motivated by this gap, we present LMR-BENCH, a benchmark designed to systematically evaluate the capability of LLM agents on code reproduction from Language Modeling Research. It consists of 28 code reproduction tasks derived from 23 research papers published in top-tier NLP venues over the past five years, spanning nine fundamental categories. Models are provided with a research paper, a code repository containing one or more masked functions, and instructions for implementing these functions. We conduct extensive experiments in standard prompting and LLM agent settings with state-of-the-art LLMs, evaluating the accuracy of unit tests and performing LLM-based evaluation of code correctness. Experimental results reveal that even the most advanced models still exhibit persistent limitations in scientific reasoning and code synthesis, highlighting critical gaps in LLM agents' ability to autonomously reproduce scientific research

Peilin Wu, Mian Zhang, Kun Wan et al.

Agentic RAG is a powerful technique for incorporating external information that LLMs lack, enabling better problem solving and question answering. However, suboptimal search behaviors exist widely, such as over-search (retrieving information already known) and under-search (failing to search when necessary), which leads to unnecessary overhead and unreliable outputs. Current training methods, which typically rely on outcome-based rewards in a RL framework, lack the fine-grained control needed to address these inefficiencies. To overcome this, we introduce Hierarchical Process Rewards for Efficient agentic RAG (HiPRAG), a training methodology that incorporates a fine-grained, knowledge-grounded process reward into the RL training. Our approach evaluates the necessity of each search decision on-the-fly by decomposing the agent's reasoning trajectory into discrete, parsable steps. We then apply a hierarchical reward function that provides an additional bonus based on the proportion of optimal search and non-search steps, on top of commonly used outcome and format rewards. Experiments on the Qwen2.5 and Llama-3.2 models across seven diverse QA benchmarks show that our method achieves average accuracies of 65.4% (3B) and 67.2% (7B). This is accomplished while improving search efficiency, reducing the over-search rate to just 2.3% and concurrently lowering the under-search rate. These results demonstrate the efficacy of optimizing the reasoning process itself, not just the final outcome. Further experiments and analysis demonstrate that HiPRAG shows good generalizability across a wide range of RL algorithms, model families, sizes, and types. This work demonstrates the importance and potential of fine-grained control through RL, for improving the efficiency and optimality of reasoning for search agents.

Kaiyu He, Peilin Wu, Mian Zhang et al.

Since the advent of large language models (LLMs), research has focused on instruction following and deductive reasoning. A central question remains: can these models discover new knowledge, and how can we evaluate this ability? We address this by studying abductive reasoning-the generation of plausible hypotheses to explain observations-and introduce GEAR (General Evaluation for Abductive Reasoning), a general-purpose, fully automated, transparent, and label-free evaluation paradigm. GEAR scores hypothesis sets by three metrics: consistency (each hypothesis explains the observations), generalizability (consistent hypotheses make meaningful predictions on unseen inputs), and diversity (the set covers distinct predictions and patterns). Built this way, GEAR is scalable (no human gold answers), reliable (deterministic scoring aligned with classical abduction), and open-ended (scores improve only when models produce new plausible hypotheses, unlike static benchmarks that saturate once accuracy is high). Using GEAR, we conduct a fine-grained study of nine LLMs on four abduction benchmarks with 1,500 problems, generating over 50,000 candidate hypotheses and revealing model differences obscured by gold-answer or purely human evaluations. We further propose a momentum-based curriculum that adjusts GEAR-derived training data by learning velocity: it starts with what the model learns quickly and shifts toward harder objectives such as generating diverse hypotheses once the model is confident on foundational objectives. Without gold-label supervision, this strategy improves all GEAR objectives and these gains transfer to established abductive reasoning benchmarks. Taken together, GEAR provides a principled framework that evaluates abduction and supplies label-free, scalable training signals that help LLMs produce more diverse and reliable hypotheses.

Peilin Wu, Jinho D. Choi

We present a novel approach to Speaker Diarization (SD) by leveraging text-based methods focused on Sentence-level Speaker Change Detection within dialogues. Unlike audio-based SD systems, which are often challenged by audio quality and speaker similarity, our approach utilizes the dialogue transcript alone. Two models are developed: the Single Prediction Model (SPM) and the Multiple Prediction Model (MPM), both of which demonstrate significant improvements in identifying speaker changes, particularly in short conversations. Our findings, based on a curated dataset encompassing diverse conversational scenarios, reveal that the text-based SD approach, especially the MPM, performs competitively against state-of-the-art audio-based SD systems, with superior performance in short conversational contexts. This paper not only showcases the potential of leveraging linguistic features for SD but also highlights the importance of integrating semantic understanding into SD systems, opening avenues for future research in multimodal and semantic feature-based diarization.

Haoran He, Peilin Wu, Chenjia Bai et al.

Reinforcement Learning (RL) has recently achieved remarkable success in robotic control. However, most works in RL operate in simulated environments where privileged knowledge (e.g., dynamics, surroundings, terrains) is readily available. Conversely, in real-world scenarios, robot agents usually rely solely on local states (e.g., proprioceptive feedback of robot joints) to select actions, leading to a significant sim-to-real gap. Existing methods address this gap by either gradually reducing the reliance on privileged knowledge or performing a two-stage policy imitation. However, we argue that these methods are limited in their ability to fully leverage the available privileged knowledge, resulting in suboptimal performance. In this paper, we formulate the sim-to-real gap as an information bottleneck problem and therefore propose a novel privileged knowledge distillation method called the Historical Information Bottleneck (HIB). In particular, HIB learns a privileged knowledge representation from historical trajectories by capturing the underlying changeable dynamic information. Theoretical analysis shows that the learned privileged knowledge representation helps reduce the value discrepancy between the oracle and learned policies. Empirical experiments on both simulated and real-world tasks demonstrate that HIB yields improved generalizability compared to previous methods. Videos of real-world experiments are available at https://sites.google.com/view/history-ib .