Search - Amazon Science

Jingzhou Hu

Applied Scientist

Hongfu Liu

Applied Scientist

Zach Bamberger

Applied Scientist

Pruning network nodes on the fly to improve LLM efficiency

Jing Liu, Grant Strimel

July 21, 2025

Language models inspired by specialized processing regions in the brain offer significant time and cost savings.

Conversational AI

Think clearly: Improving reasoning via redundant token pruning

Daewon Choi, Jimin Lee, Jihoon Tack, Woomin Song, Saket Dingliwal, Sai Muralidhar Jayanthi, Bhavana Ganesh, Jinwoo Shin, Aram Galstyan, Sravan Babu Bodapati

ICML 2025 Workshop on Efficient Systems for Foundation Models

2025

Recent large language models have shown promising capabilities in long-form reasoning, following structured chains of thought before arriving at a final answer. However, we observe that these reasoning paths tend to include substantial redundancy; analyzing attention patterns reveals that attention scores are widely scattered, particularly incorrect answers exhibit greater attention sparsity. In this paper

Conversational AI

TopoSem: In-context planning with semantically-informed tooling graph similarity

Shengjie Liu, Alex Lu, Li Dong, Jason Zhu, Manish Gawali, Alice Zhou

KDD 2025 Workshop on LLM4ECommerce

2025

Designing intelligent assistants for e-commerce sellers presents significant challenges, primarily due to the abstract nature of seller queries and the complexity of orchestrating multiple internal tools. In-context planning (ICP) has emerged as a promising adaptive problem-solving approach for this setting. However, selecting effective exemplars for ICP remains a difficult problem, largely because of the

Conversational AI

Anirudh Nair

Machine Learning Engineer

Tournament of prompts: Evolving LLM instructions through structured debates and Elo ratings

Anirudh Nair, Adi Banerjee, Laurent Mombaerts, Matthew Hagen, Tarik Borogovac

KDD 2025 Workshop on Prompt Optimization

2025

Prompt engineering represents a critical bottleneck to harness the full potential of Large Language Models (LLMs) for solving complex tasks, as it requires specialized expertise, significant trial-and-error, and manual intervention. This challenge is particularly pronounced for tasks involving subjective quality assessment, where defining explicit optimization objectives becomes fundamentally problematic

Conversational AI

Building analyst-like agents: A self-improving multi-agent framework for financial reasoning in the enterprise

Sabrina Zhang, Daksha Yadav, Tom Jin, Miriam Teng

ICML 2025 Workshop on Multi-Agent Systems

2025

Enterprise accounting data is complex, ambiguous, and shaped by evolving systems and regulations. The institutional knowledge needed to reason over the data is sparse, scattered and rarely structurally documented—posing major challenges for LLM agents. We introduce a multi-agent financial research framework that mimics a junior analyst’s onboarding and growth. The Analyst Agent learns proactively from repeated

Conversational AI

Mamba drafters for speculative decoding

Daewon Choi, Seunghyuk Oh, Saket Dingliwal, Jihoon Tack, Kyuyoung Kim, Woomin Song, Seojin Kim, Insu Han, Jinwoo Shin, Aram Galstyan, Shubham Katiyar, Sravan Babu Bodapati

ICML 2025 Workshop on Efficient Systems for Foundation Models

2025

Speculative decoding has emerged as a promising approach to accelerating large language model (LLM) generation using a fast drafter while maintaining alignment with the target model’s distribution. However, existing approaches face a tradeoff: external drafters offer flexibility but can suffer from slower drafting, while self-speculation methods use drafters tailored to the target model but require re-training

Conversational AI

Measuring the fairness gap between retrieval and generation in RAG systems using a cognitive complexity framework

Sandeep Avula, Chia-Jung Lee, Rongting Zhang, Vanessa Murdock

SIGIR 2025

2025

In this paper, we investigate the problem of quantifying fairness in Retrieval-Augmented Generation (RAG) systems, particularly for complex cognitive tasks that go beyond factual question-answering. While RAG systems have demonstrated effectiveness in information extraction tasks, their fairness implications for cognitively complex tasks - including ideation, content creation, and analytical reasoning —

Search and information retrieval

Compress, gather, and recompute: REFORMing long-context processing in transformers

Woomin Song, Sai Muralidhar Jayanthi, Srikanth Ronanki, Kanthashree Mysore Sathyendra, Jinwoo Shin, Aram Galstyan, Shubham Katiyar, Sravan Babu Bodapati

ICML 2025 Workshop on Efficient Systems for Foundation Models

2025

As large language models increasingly gain popularity in real-world applications, processing extremely long contexts, often exceeding the model’s pre-trained context limits, has emerged as a critical challenge. While existing approaches to efficient long-context processing show promise, recurrent compression-based methods struggle with information preservation, whereas random access approaches require substantial

Conversational AI

CatalogRAG: Retrieval-guided LLM prediction for multilingual e-commerce product attributes

Bryan Zhang, Suleiman Khan, Stephan Walter

KDD 2025 Workshop on LLM4ECommerce

2025

E-commerce stores increasingly use Large Language Models (LLMs) to enhance catalog data quality through automated regeneration. A critical challenge is accurately predicting missing structured attribute values across multilingual product catalogs, where LLM performance varies significantly by language. While existing approaches leverage general knowledge through prompt engineering and external retrieval

Machine learning

Jiekai Wang

Applied Scientist

ICML 2025

The International Conference on Machine Learning (ICML) is the premier gathering of professionals dedicated to the advancement of the branch of artificial intelligence known as machine learning.

Machine learning

July 13 - 19, 2025

Vancouver, Canada

Website

FalseReject: Reducing overcautiousness in LLMs through reasoning-aware safety evaluation

Zhehao Zhang, Weijie Xu

July 18, 2025

Novel graph-based, adversarial, agentic method for generating training examples helps identify — and mitigate — "overrefusal".

Conversational AI

Xie Xiaohui

Applied Scientist

SIGIR 2025

SIGIR is the premier international forum for the presentation of new research results and for the demonstration of new systems and techniques in information retrieval. The conference consists of five days of full papers, short papers, resource & reproducibility papers, perspectives papers, system demonstrations, doctoral consortium, tutorials, and workshops focused on research and development in the area of information retrieval.

Search and information retrieval

July 13 - 17, 2025

Padova, Italy

Website

Mia Kokic

Applied Scientist

Raymond Yeung

Principal Research Scientist

Search results

Work with us