Ruoxi Sun
Ruoxi Sun completed Ph.D. in machine learning and computational neuroscience from Columbia University in 2019. She is generally interested in a broad aspect of machine learning topics:
deep learning, language models, applied science and drug discoveries, computer vision, adversarial settings, representation learning, unsupervised learning, image/video understanding.
Authored Publications
Sort By
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
From Few to Many: Self-Improving Many-Shot Reasoners Through Iterative Optimization and Generation
Han Zhou
Hootan Nakhost
International Conference on Learning Representations (ICLR) (2025)
Preview abstract
Recent advances in long-context large language models (LLMs) have led to the emerging paradigm of many-shot in-context learning (ICL), where it is observed that scaling many more demonstrating examples beyond the conventional few-shot setup in the context can lead to performance benefits. However, despite its promise, it is unclear what aspects dominate the benefits and whether simply scaling to more examples is the most effective way of improving many-shot ICL. In this work, we first provide an analysis of the factors driving many-shot ICL, and we find that 1) many-shot performance can still be attributed to often a few disproportionately influential examples and 2) identifying such influential examples ("optimize") and using them as demonstrations to regenerate new examples ("generate") can lead to further improvements. Inspired by the findings, we propose BRIDGE, an algorithm that alternates between the optimize step with Bayesian optimization to discover the influential sets of examples and the generate step to reuse this set to expand the reasoning paths of the examples back to the many-shot regime automatically. On Gemini, Claude, and Mistral LLMs of different sizes, we show that BRIDGE to significant improvements across a diverse set of tasks, including symbolic reasoning, numerical reasoning, and code generation.
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
Reasoning-SQL: Reinforcement Learning with Partial Rewards for Reasoning-Enhanced Text-to-SQL
Mohammadreza Pourreza
Shayan Talaei
Hailong Li
Azalia Mirhoseini
Amin Saberi
Conference on Language Modeling (COLM) (2025) (to appear)
Preview abstract
Text-to-SQL is a challenging task involving multiple reasoning-intensive subtasks, including natural language understanding, database schema comprehension, and precise SQL query formulation. Existing approaches often rely on handcrafted reasoning paths with inductive biases that can limit their overall effectiveness. Motivated by the recent success of reasoning-enhanced models such as DeepSeek R1 and OpenAI o1, which effectively leverage reward-driven self-exploration to enhance reasoning capabilities and generalization, we propose a novel set of partial rewards tailored specifically for the Text-to-SQL task. Our reward set includes schema-linking, AI feedback, n-gram similarity, and syntax check, explicitly designed to address the reward sparsity issue prevalent in reinforcement learning (RL). Leveraging group relative policy optimization (GRPO), our approach explicitly encourages large language models (LLMs) to develop intrinsic reasoning skills necessary for accurate SQL query generation. With models of different sizes, we demonstrate that RL-only training with our proposed rewards consistently achieves higher accuracy and superior generalization compared to supervised fine-tuning (SFT). Remarkably, our RL-trained 14B-parameter model significantly outperforms larger proprietary models, e.g. o3-mini by 4% and Gemini-1.5-Pro-002 by 3% on the BIRD benchmark. These highlight the efficacy of our proposed RL-training framework with partial rewards for enhancing both accuracy and reasoning capabilities in Text-to-SQL tasks.
View details
Preview abstract
Large language models (LLMs), optimized through human feedback, have rapidly emerged as a leading paradigm for developing intelligent conversational assistants. However, despite their strong performance across many benchmarks, LLM-based agents might still lack conversational skills such as disambiguation -- when they are faced with ambiguity, they often overhedge or implicitly guess users' true intents rather than asking clarification questions. Under task-specific settings, high-quality conversation samples are often limited, constituting a bottleneck for LLMs' ability to learn optimal dialogue action policies. We propose Action-Based Contrastive Self-Training (ACT), a quasi-online preference optimization algorithm based on Direct Preference Optimization (DPO), that enables data-efficient dialogue policy learning in multi-turn conversation modeling. We demonstrate ACT's efficacy under data-efficient tuning scenarios, even when there is no action label available, using multiple real-world conversational tasks: tabular-grounded question-answering, machine reading comprehension, and AmbigSQL, a novel task for disambiguating information-seeking requests for complex SQL generation towards data analysis agents. Additionally, we propose evaluating LLMs' ability to function as conversational agents by examining whether they can implicitly recognize and reason about ambiguity in conversation. ACT demonstrates substantial conversation modeling improvements over standard tuning approaches like supervised fine-tuning and DPO.
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
Multi-Agent Design: Optimizing Agents with Better Prompts and Topologies
Ivan Vulić
Anna Korhonen
Han Zhou
Shariq Iqbal
2025
Preview abstract
Large language models (LLMs), employed as multiple agents that interact and collaborate with each other, have excelled at solving complex tasks. The agents are programmed with {prompts} that declare their functionality, along with the {workflows} that orchestrate interactions within a structured flow. Designing prompts and workflows for multi-agent systems is inherently complex, especially when addressing a new task. It often demands expert-level knowledge and involves significant trial and error. Gaining a deep understanding of the factors that contribute to effective multi-agent systems is essential for automating the entire process. Motivated by this, we first conduct an in-depth analysis of the design spaces for multi-agent systems, focusing on the impact of prompts, scaling the number of agents, and common types of agentic modules. Our findings reveal that top-performing systems often emerge from simpler design spaces, where prompts play a critical role in enhancing agent functionality and enabling more effective scaling. Based on the insights, we propose Multi-Agent System Search (MASS), a multi-stage optimization framework that performs the optimization in a pruned design space, with prompts and an influential subset of modules. We show that MASS-optimized multi-agent systems outperform existing alterntives by a substantial margin. Based on the MASS-found systems, we finally propose design principles behind building effective multi-agent systems.
View details
Spider2.0-GUI: Can Multimodal Agents Achieve Expert Proficiency in Data Science and Engineering?
Ruisheng Cao
Fangyu Lei
Haoyuan Wu
Jixuan Chen
Yeqiao Fu
Hongcheng Gao
Xinzhuang Xiong
Hanchong Zhang
Yuchen Mao
Wenjing Hu
Tianbao Xie
Hongshen Xu
Danyang Zhang
Sida Wang
Pengcheng Yin
Caiming Xiong
Ansong Ni
Qian Liu
Victor Zhong
Lu Chen
Kai Yu
Tao Yu
2024
Preview abstract
The field of data science and engineering is crucial for harnessing large-scale data to assist both individuals and enterprises in analytical processing and automated orchestration. Despite the significance, large language model~(LLM)-based data agents remain underexplored, particularly concerning professional data engineering tools such as {\tt dbt}, {\tt Airflow}, and {\tt Airbyte}, which are complex to use and include intensive GUI operations. To bridge this gap, we introduce Spider2.0-GUI, the first benchmark focusing on enterprise data engineering softwares across a full data pipeline. It encapsulates $486$ tasks involving $20$ professional softwares, guiding through tasks such as data warehousing, ingestion, transformation, analysis, visualization, and orchestration. Each task is paired with both abstract and verbose instructions, considering different levels of user expertise. We also build a comprehensive document warehouse with $11,231$ documents for Spider2.0-GUI to support retrieval-augmented agent frameworks. The benchmark is further enhanced with a real-time, executable Ubuntu desktop environment that interacts with real-world internet, providing a realistic and dynamic testing ground. Preliminary results with state-of-the-art vision language models~(VLMs) indicate that even the most advanced model only achieves $11\%$ success rate~(SR) with abstract instructions, and $21\%$ SR with verbose instructions~(a.k.a., step-by-step tutorials). This benchmark not only investigates the competencies of data agents, but also paves the way for future advancements in real-world automated data science and engineering tasks.
View details
SQL-PaLM: Improved Large Language Model Adaptation for Text-to-SQL
Satya Gundabathula
Pengcheng Yin
Hanjun Dai
Hootan Nakhost
TMLR (2024)
Preview abstract
Text-to-SQL, the process of translating natural language into Structured Query Language
(SQL), represents a transformative application of large language models (LLMs), potentially
revolutionizing how humans interact with data. This paper introduces the SQL-PaLM
framework, a comprehensive solution for understanding and enhancing Text-to-SQL using
LLMs, using in the learning regimes of few-shot prompting and instruction fine-tuning. With
few-shot prompting, we explore the effectiveness of consistency decoding with execution-based error filtering. With instruction fine-tuning, we delve deep in understanding the critical
paradigms that influence the performance of tuned LLMs. In particular, we investigate
how performance can be improved through expanded training data coverage and diversity,
synthetic data augmentation, and integrating query-specific database content. We propose
a test-time selection method to further refine accuracy by integrating SQL outputs from
multiple paradigms with execution feedback as guidance. Additionally, we tackle the
practical challenge of navigating intricate databases with a significant number of tables and
columns, proposing efficient techniques for accurately selecting relevant database elements to
enhance Text-to-SQL performance. Our holistic approach yields substantial advancements
in Text-to-SQL, as demonstrated on two key public benchmarks, Spider and BIRD. Through
comprehensive ablations and error analyses, we shed light on the strengths and weaknesses
of our framework, offering valuable insights into Text-to-SQL’s future work.
View details
Preview abstract
Large language models (LLMs) have achieved remarkable advancements in natural language understanding, generation, and manipulation of text-based data. However, one major issue towards their widespread deployment in the real world is that they can generate "hallucinated" answers that are not factual. Towards this end, this paper focuses on improving grounding from a holistic perspective with a novel framework, AGREE. We start with the design of a test time adaptation capability that takes into account the support information generated in self-grounded responses. To effectively enable this capability, we propose that the model tuning needs to be redesigned with a novel tuning objective mimicking the test time adaptation setup for grounding. This tuning on top of the pre-trained LLMs requires small amount of data that need to be constructed in a particular way to learn the grounding information, for which we introduce a data construction method. Our results show that AGREE pushes the state-of-the-art in grounding, demonstrated across many datasets.
View details
