Jiefeng Chen
Jiefeng Chen is a Research Scientist at Google Cloud AI Research, working on building more efficient and reliable Large Language Model (LLM)-based agents for real-world applications. For complete list of publications and latest updates, please check out his primary homepage or visit his Google Scholar page.
Research Areas
Authored Publications
Sort By
Cost-effective Agent Test-Time Scaling
Samira Daruki
I-Hung Hsu
Jin Miao
Tengxiao Liu
2026
Preview abstract
Scaling test-time computation improves performance across different tasks on large language models (LLMs), yet mainstream scaling methods remain challenging for tool-augmented LLM agents. Sequential scaling tends to yield shallow tool use and under-exploration, whereas parallel scaling inflates cost through repeated tool calls. The dual costs of tokens and tool calls further complicate cost accounting and hinder fair comparison. In this work, we study the test-time scaling of widely used, tool-reliant search agents under resource constraints, analyzing performance with a unified cost metric that incorporates both tokens and tool calls. To this end, we propose Cost-effective Agent Test-Time Scaling (CATS), a budget-aware framework designed to support more cost-effective scaling by guiding resource allocation between sequential and parallel exploration. Experiments across search-intensive benchmarks show that CATS produces more favorable scaling curves, attaining higher accuracy with fewer tool calls and lower overall cost. Our work introduces a cost-conscious design for agent test-time scaling and contributes empirical insights that enable a more transparent and principled understanding of scaling in tool-augmented agents.
View details
FabScore: Fine-Grained Evaluation of Fabrications in Automated AI Research
James Xu Zhao
See-Kiong Ng
Dongfu Jiang
Bryan Hooi
Qianyun Guo
Hui Chen
Pang Wei Koh
Muhao Chen
Yiwei Wang
2026
Preview abstract
In automated AI research, scientific rigor is not merely a matter of producing coherent papers and executable code; rather, it fundamentally requires that the experimental results claimed in a paper are faithfully supported by the accompanying implementation and verifiable through actual execution logs. When this alignment breaks down, AI-generated research may contain fabrications: discrepancies between the methods described in the paper and what the code actually implements, or between the reported results and those obtained by running the code. This motivates a systematic evaluation of fabrications that systematically examines the consistency between a paper's claimed experimental outcomes and its accompanying code files. In automated AI research, scientific rigor is not merely a matter of producing coherent papers and executable code; rather, it fundamentally requires that the experimental results claimed in a paper are faithfully supported by the accompanying implementation and verifiable through actual execution logs. When this alignment breaks down, AI-generated research may contain fabrications: discrepancies between the methods described in the paper and what the code actually implements, or between the reported results and those obtained by running the code. This motivates a systematic evaluation of fabrications that systematically examines the consistency between a paper's claimed experimental outcomes and its accompanying code files. This is a placeholder. This is a placeholder. This is a placeholder.
View details
Preview abstract
Understanding long visual documents, where information is distributed across extensive pages of text and graphics, remains a critical challenge for modern Vision-Language Models (VLMs). This difficulty is rooted in two fundamental obstacles: poor evidence localization and a high tendency for model hallucination. To address these issues, we propose DocLens, a multi-agent framework that decomposes the task into two specialized stages.
First, a Lens Module leverages document parsing tools for fine-grained, hierarchical evidence localization at both the page and element level. Second, a Reasoning Module employs a sampling-adjudication mechanism to systematically analyze the localized evidence, mitigating hallucination and synthesizing reliable answers.
Paired with Gemini-2.5-Pro, DocLens achieves state-of-the-art performance on MMLongBench-Doc and FinRAGBench-V, even surpassing human experts. Furthermore, our framework offers a highly cost-effective variant that delivers comparable performance to strong baselines at a five-fold reduction in cost.
View details
Preview abstract
Autonomous research agents can now produce competitive solutions and complete manuscripts, yet their papers routinely contain fabricated citations, method descriptions disconnected from the code, and scores on incorrect scales---failures invisible to evaluations that assess fluency rather than evidentiary grounding. The core problem is verifiability: no existing system maintains a traceable chain from each claim in the paper to its grounding evidence, and current evaluation protocols assess output fluency rather than evidentiary grounding. We address this with Chaine-of-Evidence (CoE), a verifiability standard requiring every claim to trace to its grounding evidence, and instantiate it in Scientist One, an end-to-end research system that maintains evidence chains natively, and CoE Audit, an evaluation protocol with four integrity checks targeting the most damaging chain failures. Auditing 60 papers from four systems, we find every baseline exhibits at least one failure: phantom citations at 4--25%, method-code alignment in at most 2/15 papers. Scientist One achieves zero phantom citations (0/830), the highest alignment rate (7/15), and competitive solver scores.
View details
Preview abstract
Automating AI research differs from general software engineering due to computationally expensive evaluation (e.g., model training) and opaque performance attribution. Current LLM-based agents struggle here, often generating monolithic scripts that ignore execution costs and causal factors. We introduce MARS (Modular Agent with Reflective Search), a framework optimized for autonomous AI research. MARS relies on three pillars: (1) Budget-Aware Planning via cost-constrained Monte Carlo Tree Search (MCTS) to explicitly balance performance with execution expense; (2) Modular Construction, employing a "Design-Decompose-Implement" pipeline to manage complex research repositories; and (3) Comparative Reflective Memory, which addresses credit assignment by analyzing solution differences to distill high-signal insights. MARS achieves state-of-the-art performance among open-source frameworks on MLE-Bench under comparable settings, maintaining competitiveness with the global leaderboard's top methods. Furthermore, the system exhibits qualitative "Aha!" moments, where 63% of all utilized lessons originate from cross-branch transfer, demonstrating that the agent effectively generalizes insights across search paths.
View details
TUMIX: Augmenting LLM Reasoning with a Dynamic Tool-Use Mixture
Chuchu Fan
Na Li
Chi Wang
Ji Yin
Yongchao Chen
2025
Preview abstract
Integrating tools like Code Interpreter and Search has significantly improved Large Language Models (LLMs) reasoning, as shown by leading models such as OpenAI's ChatGPT Agent, Google's Gemini-Pro, and XAI's Grok4. However, the research community still lacks practical guidance on fully leveraging these tools. The main challenge lies in finding an effective method to fully exploit the benefits of textual reasoning, coding, and searching when facing distinctive questions. To address this, we propose an ensemble-based framework that runs multiple agents in parallel, each exploring different answer paths with distinct tool-use strategies. Agents iteratively share and refine their answers by considering the original question and previous responses. Our proposed method Tool-Use Mixture (TUMIX) achieves significant gains over other representative tool-augmented test-time scaling methods such as Self-MoA, Symbolic-MoE, DEI, SciMaster, and GSA. With near equal inference costs, TUMIX delivers an average +3.55% accuracy improvement over the best baseline on Gemini-2.5-Pro and Gemini-2.5-Flash across key reasoning benchmarks (HLE, GPQA, AIME 24&25), where coding and search can effectively support reasoning when applied properly. We find that agent diversity and quality are crucial, and can be further improved by querying LLMs to automatically optimize agent designs. To reduce costs, TUMIX halts refinement once sufficient confidence is reached, preserving nearly the same performance at just 49% of the inference cost. With further scaling, TUMIX can achieve even higher performance, though at substantially greater cost.
View details
Preview abstract
Inference-time scaling has been successful in enhancing large language model (LLM) performance by increasing computation at test time, but it often relies on external verifiers or is not optimized for manageable computational budgets. To address these, we propose DynScaling, which addresses these limitations through two primary innovations: an integrated parallel-sequential sampling strategy and a bandit-based dynamic budget allocation framework. The integrated sampling strategy unifies parallel and sequential sampling by constructing synthetic sequential reasoning chains from initially independent parallel responses, promoting diverse and coherent reasoning trajectories. The dynamic budget allocation framework formulates the allocation of computational resources as a multi-armed bandit problem, adaptively distributing the inference budget across queries based on the uncertainty of previously sampled responses, thereby maximizing computational efficiency. By synergizing these components, DynScaling effectively improves LLM performance under practical resource constraints without the need for external verifiers. Experimental results demonstrate that DynScaling consistently surpasses existing verifier-free inference scaling baselines in both task performance and computational cost.
View details
Preview abstract
While Large Language Models (LLMs) have shown remarkable advancements in reasoning and tool use, they often fail to generate optimal, grounded solutions under complex constraints. Real-world travel planning exemplifies these challenges, evaluating agents' abilities to handle constraints that are explicit, implicit, and even evolving based on interactions with dynamic environments and user needs. In this paper, we present ATLAS, a general multi-agent framework designed to effectively handle such complex nature of constraints awareness in real-world travel planning tasks. Our framework introduces a principled approach to address the fundamental challenges of constraint-aware planning through dedicated mechanisms for dynamic constraint management, iterative plan critique, and adaptive interleaved search. ATLAS demonstrates state-of-the-art performance on the TravelPlanner benchmark, improving the final pass rate from 17.8% to 44.4% over its best alternative. More importantly, this is the first work to be evaluated in and demonstrate quantitative effectiveness on real-world travel planning with live information search and multi-turn feedback. In this realistic setting, ATLAS demonstrates its ability to adapt to multi-turn user feedback, achieving an 84% final pass rate which significantly outperforms baselines including ReAct (59%) and a monolithic agent (27%).
View details
SETS: Leveraging Self-Verification and Self-Correction for Improved Test-Time Scaling
Xinyun Chen
Transactions on Machine Learning Research (TMLR) (2025)
Preview abstract
Recent advancements in Large Language Models (LLMs) have created new opportunities to enhance performance on complex reasoning tasks by leveraging test-time computation. However, existing scaling methods have key limitations: parallel methods like repeated sampling are often inefficient and quickly saturate, while sequential methods like SELF-REFINE struggle to improve after a few rounds. Although combining these approaches shows promise, current methods require fine-tuned reward and revision models. This paper proposes Self-Enhanced Test-Time Scaling (SETS), a simple yet effective approach that overcomes these limitations by strategically combining parallel and sequential techniques and fully leveraging LLMs' self-improvement abilities. SETS exploits the inherent self-verification and self-correction capabilities of LLMs, unifying sampling, verification, and correction within a single framework. This facilitates efficient and scalable test-time computation for enhanced performance on complex tasks without any model training. Our comprehensive experimental results on challenging benchmarks spanning planning, reasoning, math, and coding demonstrate that SETS achieves significant performance improvements and more advantageous test-time scaling behavior than the alternatives.
View details
Astute RAG: Overcoming Imperfect Retrieval Augmentation and Knowledge Conflicts for Large Language Models
Fei Wang
The Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025) (2025) (to appear)
Preview abstract
Retrieval-Augmented Generation (RAG), while effective in integrating external knowledge to address the limitations of large language models (LLMs), can be undermined by imperfect retrieval, which may introduce irrelevant, misleading, or even malicious information. Despite its importance, previous studies have rarely explored the behavior of RAG through joint analysis on how errors from imperfect retrieval attribute and propagate, and how potential conflicts arise between the LLMs' internal knowledge and external sources. We find that imperfect retrieval augmentation might be inevitable and quite harmful, through controlled analysis under realistic conditions. We identify the knowledge conflicts between LLM-internal and external knowledge from retrieval as a bottleneck to overcome in the post-retrieval stage of RAG. To render LLMs resilient to imperfect retrieval, we propose Astute RAG, a novel RAG approach that adaptively elicits essential information from LLMs' internal knowledge, iteratively consolidates internal and external knowledge with source-awareness, and finalizes the answer according to information reliability. Our experiments using Gemini and Claude demonstrate that Astute RAG significantly outperforms previous robustness-enhanced RAG methods. Notably, Astute RAG is the only approach that matches or exceeds the performance of LLMs without RAG under worst-case scenarios. Further analysis reveals that Astute RAG effectively resolves knowledge conflicts, improving the reliability and trustworthiness of RAG systems.
View details
