arXiv subjects

Zichen Zhu

Publications and source records attributed to Zichen Zhu.

At least 19 recordsLinked to original sources

arXiv2026

LigBench: A Unified and Human-Aligned Benchmark for LLM-based Research Idea Generation

With the rapid advancement of large language models (LLMs), research idea generation has attracted increasing attention. Existing approaches enable LLMs to retrieve relevant literature and propose novel ideas for research areas. However, current evaluation practices for idea generation remain fragmented and lack objective standards, often relying on direct LLM scoring, which limits their ability to provide unified and reliable assessments across a coherent distribution of generated ideas. To address this challenge, we propose LigBench, an automated evaluation benchmark that enables fine-grained and reliable evaluation of AI research ideas, consistently applicable across different generation distributions. In addition, we introduce PAIR-IQ, a dataset tailored for training pairwise idea judgment models and serving as an auxiliary reference to support more objective comparative evaluation. Extensive experiments demonstrate that LigBench achieves stable and interpretable evaluations, significantly improving alignment with expert judgments. Furthermore, models trained on PAIR-IQ exhibit enhanced ranking accuracy and robustness, establishing a principled standard for scalable and objective research idea assessment.

cs.CL↗

arXiv2026

One Student, Many Teachers: Multi-Task On-Policy Distillation via Soft-Prompt Privileged Context

On-policy self-distillation (OPSD) teaches large language models new skills through a teacher that shares the student's backbone and supervises its own rollouts. Existing teachers either inject privileged context at the input -- inducing post-hoc rationalization -- or fine-tune weights, accumulating drift and forgetting across tasks. We propose \method, whose teacher differs from the student only by a learnable soft prompt: trained on $(x, y_\text{gold})$ pairs with the backbone frozen, the prompt yields a task-specific teacher that preserves the student's exact representational geometry. \method\ extends naturally to multi-task settings by routing each example in a merged corpus to its corresponding soft-prompt teacher, allowing a single student to absorb knowledge from $K$ teachers in parallel; at inference, all prompts are discarded. On Qwen3-1.7B-Base and Phi-4-mini-instruct across four tasks (Science, Tool Use, Biology, Math), the single-task variant (OPD with a PT teacher) matches or exceeds full fine-tuning while training orders of magnitude fewer parameters, and the multi-task variant achieves the best overall average ($56.2$ on Qwen3-1.7B-Base) while preserving general-capability benchmarks -- in contrast to sequential SFT, which degrades both.

cs.LG↗

arXiv2026

Externalizing Research Synthesis and Validation in AI Scientists through a Research Harness

AI systems can increasingly automate scientific workflows, but the reasoning that links prior evidence, generated ideas, experiments and final claims often remains implicit inside model inference. Here we introduce Xcientist, a research harness that externalizes research synthesis and experimental validation into inspectable, contract-governed processes. Xcientist organizes literature evidence, idea states, implementation plans, ablation records and repair traces as persistent research artifacts, so that generated mechanisms can be grounded, executed, tested and revised without losing their evidential basis. We identify claim drift as a failure mode of automated research, where runnable artifacts no longer support the mechanism originally claimed. Across training-free memory systems, graph-structured traffic forecasting and multi-scale physics-informed neural networks, Xcientist preserves traceable trajectories from problem formulation to mechanism design, validation and bounded revision. These results suggest that AI scientists should be evaluated not only by their final artifacts, but by whether their synthesis and validation processes remain attributable, inspectable and scientifically accountable.

cs.AI↗

arXiv2026

IEA: Amateur-Friendly Conversational Image Editing Agent via Three Stages of Multitask Alignment

Current image editing software often hinges on fixed filters or expert tuning, leaving a gap between amateur users' intent and outcomes. Creations by generative models may contain artifacts, implausible details, or stylistic drift away from photorealism and offer little insight into why an edit was made. We propose IEA, a conversational Image Editing Agent that learns to operate parameterized tools in an explicit, interpretable action space. IEA is trained via a three-stage multitask pipeline: (1) SFT on distilled expert edits, (2) GRPO with rewards for likeness improvement, tool usefulness, and intent summarization, and (3) large-scale synthetic fine-tuning to jointly master image editing, refinement, and user intent summarization. By manipulating 16 editing tools step by step, IEA produces transparent edit traces that can be inspected and debugged. In quantitative experiments, it attains a lower pixel distance on the edit task and a higher ROUGE-L on the summary task than strong baselines. In user studies, it ranks best among tool-calling methods for instruction following while surpassing generative methods in overall perceptual quality. Our results validate interpretable, tool-centric VLMs as a reliable path to human instruction-guided image retouching.

cs.CV↗

arXiv2026

SSR: Can Simulated Patients Learn to Stigmatize Themselves? Modeling Self-Stigma through Internal Monologue

Simulating patients with large language models (LLMs) is a promising tool for mental health training, but existing approaches fail to capture a key clinical reality: self-stigma. Patients experiencing self-stigma, the internalization of negative stereotypes, often exhibit context-sensitive resistance, such as avoidance, denial, or self-blame, which current models render as static or uniformly compliant behavior. To address this, we introduce a novel simulation framework grounded in the psychological 3A1H model of self-stigmatization. Our core innovation is the creation of a \textbf{Stigmatized Self-Reflection} (\textbf{SSR}) dataset, where we augment mental health dialogues with internal monologues that reflect stigma-aware reasoning. By fine-tuning LLMs with this data using a chain-of-thought approach, we train patient agents to dynamically adjust their level and expression of stigma based on conversational triggers. Evaluations demonstrate that our approach significantly outperforms specialized baselines, generating more authentic and situationally appropriate patient responses. This work provides a crucial step towards realistic stigma simulation for clinical training and empathetic dialogue systems.

cs.CL↗

arXiv2026

CharTool: Tool-Integrated Visual Reasoning for Chart Understanding

Charts are ubiquitous in scientific and financial literature for presenting structured data. However, chart reasoning remains challenging for multimodal large language models (MLLMs) due to the lack of high-quality training data, as well as the need for fine-grained visual grounding and precise numerical computation. To address these challenges, we first propose DuoChart, a scalable dual-source data pipeline that combines synthesized charts with real-world charts to construct diverse, high-quality chart training data. We then introduce CharTool, which equips MLLMs with external tools, including image cropping for localized visual perception and code-based computation for accurate numerical reasoning. Through agentic reinforcement learning on DuoChart, CharTool learns tool-integrated reasoning grounded in chart content. Extensive experiments on six chart benchmarks show that our method consistently improves over strong MLLM baselines across model scales. Notably, CharTool-7B outperforms the base model by +8.0% on CharXiv (Reasoning) and +9.78% on ChartQAPro, while achieving competitive performance with substantially larger or proprietary models. Moreover, CharTool demonstrates positive generalization to out-of-domain visual math reasoning benchmarks.

cs.AI↗

arXiv2026

O^3-LSM: Maximizing Disaggregated LSM Write Performance via Three-Layer Offloading

Log-Structured Merge-tree-based Key-Value Stores (LSM-KVS) have been optimized and redesigned for disaggregated storage via techniques such as compaction offloading to reduce the network I/Os between compute and storage. However, the constrained memory space and slow flush at the compute node severely limit the overall write throughput of existing optimizations. In this paper, we propose O3-LSM, a fundamental new LSM-KVS architecture, that leverages the shared Disaggregated Memory (DM) to support a three-layer offloading, i.e., memtable Offloading, flush Offloading, and the existing compaction Offloading. Compared to the existing disaggregated LSM-KVS with compaction offloading only, O3-LSM maximizes the write performance by addressing the above issues. O3-LSM first leverages a novel DM-Optimized Memtable to achieve dynamic memtable offloading, which extends the write buffer while enabling fast, asynchronous, and parallel memtable transmission. Second, we propose Collaborative Flush Offloading that decouples the flush control plane from execution and supports memtable flush offloading at any node with dedicated scheduling and global optimizations. Third, O3-LSM is further improved with the Shard-Level Optimization, which partitions the memtable into shards based on disjoint key-ranges that can be transferred and flushed independently, unlocking parallelism across shards. Besides, to mitigate slow lookups in the disaggregated setting, O3-LSM also employs an adaptive Cache-Enhanced Read Delegation mechanism to combine a compact local cache with DM-assisted memtable delegated read. Our evaluation shows that O3-LSM achieves up to 4.5X write, 5.2X range query, and 1.8X point lookup throughput improvement, and up to 76% P99 latency reduction compared with Disaggregated-RocksDB, CaaS-LSM, and Nova-LSM.

cs.DB↗

arXiv2026

PACER: Blockwise Pre-verification for Speculative Decoding with Adaptive Length

Speculative decoding (SD) is a powerful technique for accelerating the inference process of large language models (LLMs) without sacrificing accuracy. Typically, SD employs a small draft model to generate a fixed number of draft tokens, which are then verified in parallel by the target model. However, our experiments reveal that the optimal draft length varies significantly across different decoding steps. This variation suggests that using a fixed draft length limits the potential for further improvements in decoding speed. To address this challenge, we propose Pacer, a novel approach that dynamically controls draft length using a lightweight, trainable pre-verification layer. This layer pre-verifies draft tokens blockwise before they are sent to the target model, allowing the draft model to stop token generation if the blockwise pre-verification fails. We implement Pacer on multiple SD model pairs and evaluate its performance across various benchmarks. Our results demonstrate that Pacer achieves up to 2.66x Speedup over autoregressive decoding and consistently outperforms standard speculative decoding. Furthermore, when integrated with Ouroboros, Pacer attains up to 3.09x Speedup.

cs.CL↗

arXiv2025

ChemDFM-R: A Chemical Reasoning LLM Enhanced with Atomized Chemical Knowledge

Atomized chemical knowledge, such as functional group information of molecules and reactions, plays a pivotal intermediate role in the reasoning process that connects molecular structures with their properties and reactivities. While large language models (LLMs) have achieved impressive progress, the absence of atomized chemical knowledge results in their superficial understanding of chemistry and limited chemical reasoning capabilities. In this work, to tackle this problem, we develop a Chemical Reasoning LLM, ChemDFM-R. We first construct a comprehensive dataset of atomized chemical knowledge, ChemFG, annotating the presence of functional groups in molecules and the changes of functional groups during chemical reactions, to enhance the model's understanding of the fundamental principles and internal logic of chemistry. Then, we propose a mixed-source distillation method that initializes the model's reasoning capability with limited distilled data, and develop a four-stage training pipeline to equip the model with atomized chemical knowledge and chemical reasoning logic. Experiments on diverse chemical benchmarks demonstrate that ChemDFM-R achieves cutting-edge performance while providing interpretable, rationale-driven outputs, surpassing both the general-domain LLMs and domain-specific chemical LLMs. Moreover, ChemDFM-R achieves comparable or superior performance compared with cutting-edge commercial LLMs, such as o4-mini. Further case studies illustrate how explicit reasoning chains significantly improve the model's reliability, transparency, and practicality in real-world human-AI collaboration scenarios.

cs.CE↗

arXiv2025

RetroDFM-R: Reasoning-Driven Retrosynthesis Prediction with Large Language Models via Reinforcement Learning

Retrosynthetic planning is a cornerstone of organic synthesis and drug discovery. Yet existing AI methods often rely on pattern matching rather than transferable chemical reasoning, limiting both generalizability and interpretability. Here we introduce RetroDFM-R, a reasoning-driven large language model (LLM) for chemical retrosynthesis. Leveraging large-scale reinforcement learning, RetroDFM-R moves beyond black-box prediction by coupling improved accuracy with transparent, step-by-step rationale. On the USPTO-50K benchmark, RetroDFM-R achieves 60.4% accuracy without augmentation and 66.1% with the full inference setup, outperforming previous state-of-the-art baselines. Beyond standard metrics, double-blind expert evaluation further supports the chemical plausibility and practical utility of its proposed pathways. We also demonstrate that RetroDFM-R can reconstruct complex, multistep synthetic routes for real-world pharmaceuticals and self-assembled monolayer materials. By making its reasoning explicit and human-interpretable, RetroDFM-R addresses a key barrier to trust and supports practical deployment in automated retrosynthetic planning.

cs.CE↗

arXiv2025

ProgRM: Build Better GUI Agents with Progress Rewards

LLM-based (Large Language Model) GUI (Graphical User Interface) agents can potentially reshape our daily lives significantly. However, current LLM-based GUI agents suffer from the scarcity of high-quality training data owing to the difficulties of trajectory collection and reward annotation. Existing works have been exploring LLMs to collect trajectories for imitation learning or to offer reward signals for online RL training. However, the Outcome Reward Model (ORM) used in existing works cannot provide finegrained feedback and can over-penalize the valuable steps in finally failed trajectories. To this end, we propose Progress Reward Model (ProgRM) to provide dense informative intermediate rewards by predicting a task completion progress for each step in online training. To handle the challenge of progress reward label annotation, we further design an efficient LCS-based (Longest Common Subsequence) self-annotation algorithm to discover the key steps in trajectories and assign progress labels accordingly. ProgRM is evaluated with extensive experiments and analyses. Actors trained with ProgRM outperform leading proprietary LLMs and ORM-trained actors, illustrating the effectiveness of ProgRM. The codes for experiments will be made publicly available upon acceptance.

cs.AI↗

arXiv2025

Alignment for Efficient Tool Calling of Large Language Models

Recent advancements in tool learning have enabled large language models (LLMs) to integrate external tools, enhancing their task performance by expanding their knowledge boundaries. However, relying on tools often introduces tradeoffs between performance, speed, and cost, with LLMs sometimes exhibiting overreliance and overconfidence in tool usage. This paper addresses the challenge of aligning LLMs with their knowledge boundaries to make more intelligent decisions about tool invocation. We propose a multi objective alignment framework that combines probabilistic knowledge boundary estimation with dynamic decision making, allowing LLMs to better assess when to invoke tools based on their confidence. Our framework includes two methods for knowledge boundary estimation, consistency based and absolute estimation, and two training strategies for integrating these estimates into the model decision making process. Experimental results on various tool invocation scenarios demonstrate the effectiveness of our framework, showing significant improvements in tool efficiency by reducing unnecessary tool usage.

cs.CL↗

arXiv2025

Delusions of Large Language Models

Large Language Models often generate factually incorrect but plausible outputs, known as hallucinations. We identify a more insidious phenomenon, LLM delusion, defined as high belief hallucinations, incorrect outputs with abnormally high confidence, making them harder to detect and mitigate. Unlike ordinary hallucinations, delusions persist with low uncertainty, posing significant challenges to model reliability. Through empirical analysis across different model families and sizes on several Question Answering tasks, we show that delusions are prevalent and distinct from hallucinations. LLMs exhibit lower honesty with delusions, which are harder to override via finetuning or self reflection. We link delusion formation with training dynamics and dataset noise and explore mitigation strategies such as retrieval augmented generation and multi agent debating to mitigate delusions. By systematically investigating the nature, prevalence, and mitigation of LLM delusions, our study provides insights into the underlying causes of this phenomenon and outlines future directions for improving model reliability.

cs.CL↗

arXiv2024

AdaEAGLE: Optimizing Speculative Decoding via Explicit Modeling of Adaptive Draft Structures

Speculative Decoding (SD) is a popular lossless technique for accelerating the inference of Large Language Models (LLMs). We show that the decoding speed of SD frameworks with static draft structures can be significantly improved by incorporating context-aware adaptive draft structures. However, current studies on adaptive draft structures are limited by their performance, modeling approaches, and applicability. In this paper, we introduce AdaEAGLE, the first SD framework that explicitly models adaptive draft structures. AdaEAGLE leverages the Lightweight Draft Length Predictor (LDLP) module to explicitly predict the optimal number of draft tokens during inference to guide the draft model. It achieves comparable speedup results without manual thresholds and allows for deeper, more specialized optimizations. Moreover, together with threshold-based strategies, AdaEAGLE achieves a $1.62\times$ speedup over the vanilla AR decoding and outperforms fixed-length SotA baseline while maintaining output quality.

cs.AI↗

arXiv2024

Reducing Tool Hallucination via Reliability Alignment

Large Language Models (LLMs) have expanded their capabilities beyond language generation to interact with external tools, enabling automation and real-world applications. However, tool hallucinations, where models either select inappropriate tools or misuse them, pose significant challenges, leading to erroneous task execution, increased computational costs, and reduced system reliability. To systematically address this issue, we define and categorize tool hallucinations into two main types, tool selection hallucination and tool usage hallucination. To evaluate and mitigate these issues, we introduce RelyToolBench, which integrates specialized test cases and novel metrics to assess hallucination-aware task success and efficiency. Finally, we propose Relign, a reliability alignment framework that expands the tool-use action space to include indecisive actions, allowing LLMs to defer tool use, seek clarification, or adjust tool selection dynamically. Through extensive experiments, we demonstrate that Relign significantly reduces tool hallucinations, improves task reliability, and enhances the efficiency of LLM tool interactions.

cs.CL↗

arXiv2024

MobA: Multifaceted Memory-Enhanced Adaptive Planning for Efficient Mobile Task Automation

Existing Multimodal Large Language Model (MLLM)-based agents face significant challenges in handling complex GUI (Graphical User Interface) interactions on devices. These challenges arise from the dynamic and structured nature of GUI environments, which integrate text, images, and spatial relationships, as well as the variability in action spaces across different pages and tasks. To address these limitations, we propose MobA, a novel MLLM-based mobile assistant system. MobA introduces an adaptive planning module that incorporates a reflection mechanism for error recovery and dynamically adjusts plans to align with the real environment contexts and action module's execution capacity. Additionally, a multifaceted memory module provides comprehensive memory support to enhance adaptability and efficiency. We also present MobBench, a dataset designed for complex mobile interactions. Experimental results on MobBench and AndroidArena demonstrate MobA's ability to handle dynamic GUI environments and perform complex mobile tasks.

cs.MA↗

arXiv2024

Depression Diagnosis Dialogue Simulation: Self-improving Psychiatrist with Tertiary Memory

Mental health issues, particularly depressive disorders, present significant challenges in contemporary society, necessitating the development of effective automated diagnostic methods. This paper introduces the Agent Mental Clinic (AMC), a self-improving conversational agent system designed to enhance depression diagnosis through simulated dialogues between patient and psychiatrist agents. To enhance the dialogue quality and diagnosis accuracy, we design a psychiatrist agent consisting of a tertiary memory structure, a dialogue control and reflect plugin that acts as ``supervisor'' and a memory sampling module, fully leveraging the skills reflected by the psychiatrist agent, achieving great accuracy on depression risk and suicide risk diagnosis via conversation. Experiment results on datasets collected in real-life scenarios demonstrate that the system, simulating the procedure of training psychiatrists, can be a promising optimization method for aligning LLMs with real-life distribution in specific domains without modifying the weights of LLMs, even when only a few representative labeled cases are available.

cs.CL↗

arXiv2024

ChemDFM-X: Towards Large Multimodal Model for Chemistry

Rapid developments of AI tools are expected to offer unprecedented assistance to the research of natural science including chemistry. However, neither existing unimodal task-specific specialist models nor emerging general large multimodal models (LMM) can cover the wide range of chemical data modality and task categories. To address the real demands of chemists, a cross-modal Chemical General Intelligence (CGI) system, which serves as a truly practical and useful research assistant utilizing the great potential of LMMs, is in great need. In this work, we introduce the first Cross-modal Dialogue Foundation Model for Chemistry (ChemDFM-X). Diverse multimodal data are generated from an initial modality by approximate calculations and task-specific model predictions. This strategy creates sufficient chemical training corpora, while significantly reducing excessive expense, resulting in an instruction-tuning dataset containing 7.6M data. After instruction finetuning, ChemDFM-X is evaluated on extensive experiments of different chemical tasks with various data modalities. The results demonstrate the capacity of ChemDFM-X for multimodal and inter-modal knowledge comprehension. ChemDFM-X marks a significant milestone toward aligning all modalities in chemistry, a step closer to CGI.

cs.LG↗