Large Language Models

From Few to Many: Self-Improving Many-Shot Reasoners Through Iterative Optimization and Generation
Self-improving reasoners taking advantage of long-context capabilities (ICLR 2025)
Xingchen Wan, Han Zhou, Ruoxi Sun, Hootan Nakhost, Ke Jiang, Sercan Ö. Arık
PDF Cite Abstract
From Few to Many: Self-Improving Many-Shot Reasoners Through Iterative Optimization and Generation
Multi-Agent Design: Optimizing Agents with Better Prompts and Topologies
Automatically discovering performant multi-agent systems end-to-end. (arXiv preprint)
Han Zhou, Xingchen Wan, Ruoxi Sun, Hamid Palangi, Shariq Iqbal, Ivan Vulić, Anna Korhonen, Sercan Ö. Arık
PDF Cite Abstract
Multi-Agent Design: Optimizing Agents with Better Prompts and Topologies
Teach Better or Show Smarter? On Instructions and Exemplars in Automatic Prompt Optimization
Comprehensive study powering the Google Cloud Vertex AI Prompt Optimizer (NeurIPS 2024).
Xingchen Wan, Ruoxi Sun, Hootan Nakhost, Sercan Ö. Arık
PDF Cite Abstract OpenReview
Teach Better or Show Smarter? On Instructions and Exemplars in Automatic Prompt Optimization
UQE: A Query Engine for Unstructured Databases
Neural Information Processing Systems (NeurIPS), 2024
Hanjun Dai, Bethany Yixin Wang, Xingchen Wan, Bo Dai, Sherry Yang, Azade Nova, Pengcheng Yin, Phitchaya Mangpo Phothilimthana, Charles Sutton, Dale Schuurmans
PDF Cite Abstract OpenReview
UQE: A Query Engine for Unstructured Databases
Fairer Preferences Elicit Improved Human-Aligned Large Language Model Judgments
EMNLP 2024
Han Zhou, Xingchen Wan, Yinhong Liu, Nigel Collier, Ivan Vulić, Anna Korhonen
PDF Cite Code Abstract
Fairer Preferences Elicit Improved Human-Aligned Large Language Model Judgments
Astute RAG: Overcoming Imperfect Retrieval Augmentation and Knowledge Conflicts for Large Language Models
Overcoming RAG imperfections and knowledge conflicts with debating and reconcilation (arXiv preprint).
Fei Wang, Xingchen Wan, Ruoxi Sun, Jiefeng Chen, Sercan Ö. Arık
PDF Cite Abstract
Astute RAG: Overcoming Imperfect Retrieval Augmentation and Knowledge Conflicts for Large Language Models
Batch Calibration: Rethinking Calibration for In-Context Learning and Prompt Engineering
Mitigating prompt biases and unifying existing calibration approaches without labeled data (ICLR 2024)
Han Zhou, Xingchen Wan, Lev Proleev, Diana Mincu, Jilin Chen, Katherine Heller, Subhrajit Roy
PDF Cite Google Research Blog Abstract OpenReview
Batch Calibration: Rethinking Calibration for In-Context Learning and Prompt Engineering
Working Memory Capacity of ChatGPT: An Empirical Study
AAAI Conference on Artificial Intelligence (AAAI), 2024
Dongyu Gong, Xingchen Wan, Dingmin Wang
PDF Cite Code Abstract
Working Memory Capacity of ChatGPT: An Empirical Study
AutoPEFT: Automatic Configuration Search for Parameter-Efficient Fine-Tuning
Automatic discovery of families of high-performing PEFT configurations (TACL 2024)
Han Zhou, Xingchen Wan, Ivan Vulić, Anna Korhonen
PDF Cite Code DOI Abstract
AutoPEFT: Automatic Configuration Search for Parameter-Efficient Fine-Tuning
Universal Self-Adaptive Prompting
Empirical Methods in Natural Language Processing (EMNLP), 2023
Xingchen Wan, Ruoxi Sun, Hootan Nakhost, Hanjun Dai, Julian Martin Eisenschlos, Sercan Ö. Arık, Tomas Pfister
PDF Cite Google Research Blog Abstract (Google Research) OpenReview ACL Anthology Slides & Talk
Universal Self-Adaptive Prompting