Papers by Omar Khattab
Assisting in Writing Wikipedia-like Articles From Scratch with Large Language Models (2024.naacl-long)
Copied to clipboard
| Challenge: | Existing methods to write grounded, long-form articles have limited planning capacity and require extensive research and planning in the pre-writing stage. |
| Approach: | They propose a system for the Synthesis of Topic Outlines throughRetrieval and Multi-perspective Question Asking that models the pre-writing stage by (1) discovering diverse perspectives in researching the given topic, (2) simulating conversations where writers carrying different perspectives pose questions to a topic expert grounded on trusted Internet sources, (3) curating the collected information to create an outline. |
| Outcome: | The proposed system is based on a dataset of high-quality Wikipedia articles and evaluates the pre-writing stage. |
Problem-Oriented Segmentation and Retrieval: Case Study on Tutoring Conversations (2024.findings-emnlp)
Copied to clipboard
| Challenge: | POSR is a task of breaking down conversations into segments and linking each segment to the relevant reference item. |
| Approach: | They propose a task that breaks down conversations into segments and links each segment to the relevant reference item. |
| Outcome: | The proposed method outperforms independent segmentation pipelines and large language models on joint metrics. |
IndicIRSuite: Multilingual Dataset and Neural Information Models for Indian Languages (2024.acl-short)
Copied to clipboard
| Challenge: | IndicIRSuite is the first attempt at building large-scale Neural Information Retrieval resources for a large number of Indian languages. |
| Approach: | They introduce Neural Information Retrieval resources for 11 widely spoken Indian Languages from two major Indian language families. |
| Outcome: | Experiments show that Indic-ColBERT improves on INDIC-MARCO datasets for 11 languages, and that it can be used to improve IR for Indian languages. |
Relevance-guided Supervision for OpenQA with ColBERT (2021.tacl-1)
Copied to clipboard
| Challenge: | Recent work has focused on learning to retrieve passages for open-domain question answering . if notions of relevance are not tailored to questions, the MRC model will not reliably see the best passages . |
| Approach: | They propose a retrieval model that uses coarse-grained vector representations of questions and passages to adapt it to OpenQA. |
| Outcome: | The proposed system improves OpenQA retrieval on Natural Questions, SQuAD, and TriviaQA. |
Backtracing: Retrieving the Cause of the Query (2024.findings-eacl)
Copied to clipboard
| Challenge: | a number of online content portals allow users to ask questions to supplement their understanding. |
| Approach: | They propose a task of backtracing to retrieve the text segment that most likely caused a user query. |
| Outcome: | The proposed method improves on the backtracing task in three domains . the results show that there is room for improvement and new retrieval approaches . |
PAPILLON: Privacy Preservation from Internet-based and Local Language Model Ensembles (2025.naacl-long)
Copied to clipboard
| Challenge: | Existing research has studied privacy in LLM training data memorization, but it does not prevent users from disclosing PII at inference time. |
| Approach: | They propose a task for chaining API-based and local LLMs that uses public data to construct a benchmark that contains personally identifiable information (PII) |
| Outcome: | The proposed model maintains high response quality for 85.5% of user queries while restricting privacy leakage to only 7.5%. |
Optimizing Instructions and Demonstrations for Multi-Stage Language Model Programs (2024.emnlp-main)
Copied to clipboard
Krista Opsahl-Ong, Michael Ryan, Josh Purtell, David Broman, Christopher Potts, Matei Zaharia, Omar Khattab
| Challenge: | Language Model Programs (LMs) require crafting prompts that are jointly effective for all modules. |
| Approach: | They propose a novel algorithm for optimizing language model (LM) prompts for all modules by using program- and data-aware techniques and stochastic mini-batch evaluation functions. |
| Outcome: | The proposed algorithm outperforms baseline optimizers on five of seven diverse LM programs by as high as 13% accuracy. |
Moving Beyond Downstream Task Accuracy for Information Retrieval Benchmarking (2023.findings-acl)
Copied to clipboard
Keshav Santhanam, Jon Saad-Falcon, Martin Franz, Omar Khattab, Avi Sil, Radu Florian, Md Arafat Sultan, Salim Roukos, Matei Zaharia, Christopher Potts
| Challenge: | Neural information retrieval (IR) systems have progressed rapidly in recent years . many IR benchmarks focus on downstream task accuracy, concealing costs incurred . |
| Approach: | They propose to include efficiency considerations on IR benchmarks to help drive progress . eral et al. propose to incorporate query latency and cost budgets into evaluation . |
| Outcome: | a new study shows that the best IR system varies according to how efficiency considerations are chosen and weighed . the proposed benchmarks would allow for more thorough exploration of possible system designs . |
UDAPDR: Unsupervised Domain Adaptation via LLM Prompting and Distillation of Rerankers (2023.emnlp-main)
Copied to clipboard
Jon Saad-Falcon, Omar Khattab, Keshav Santhanam, Radu Florian, Martin Franz, Salim Roukos, Avirup Sil, Md Sultan, Christopher Potts
| Challenge: | Existing methods for information retrieval tasks require large labeled datasets for fine-tuning, but they can experience significant drops in accuracy due to distribution shifts from the training to the target domain. |
| Approach: | They propose a method for using large language models to generate large numbers of synthetic queries cheaply using an expensive LLM. |
| Outcome: | The proposed method boosts zero-shot accuracy in long-tail domains and achieves substantially lower latency than standard reranking methods. |
ColBERTv2: Effective and Efficient Retrieval via Lightweight Late Interaction (2022.naacl-main)
Copied to clipboard
| Challenge: | Neural information retrieval (IR) methods encode queries and documents into single vectors, but late interaction models produce multi-vector representations at the granularity of each token. |
| Approach: | They propose a retrieval method that couples an aggressive residual compression mechanism with a denoised supervision strategy to improve the quality and space footprint of late interaction. |
| Outcome: | The proposed retriever improves quality and space footprint of late interaction models while reducing space footprint by 6–10x. |
ARES: An Automated Evaluation Framework for Retrieval-Augmented Generation Systems (2024.naacl-long)
Copied to clipboard
| Challenge: | Evaluating retrieval-augmented generation systems relies on hand annotations for input queries, passages to retrieve, and responses to generate. |
| Approach: | They propose an automated evaluation framework for retrieval-augmented generation (RAG) ARES fine tunes lightweight LLM judges on synthetically generated queries and answers . |
| Outcome: | The proposed framework evaluates RAG systems using only human annotations . it can be used to improve system understanding and create targeted solutions . |
LangProBe: a Language Program Benchmark (2025.findings-emnlp)
Copied to clipboard
Shangyin Tan, Lakshya A Agrawal, Arnav Singhvi, Liheng Lai, Michael J Ryan, Dan Klein, Omar Khattab, Koushik Sen, Matei Zaharia
| Challenge: | Composing language models into multi-step language programs is a mainstream paradigm for building AI systems, but tradeoffs in this space have only scarcely been studied before. |
| Approach: | They propose a benchmarking tool to evaluate the architectures and optimization strategies for language programs . they find that optimized language programs offer strong cost-quality Pareto improvement . |
| Outcome: | The proposed framework evaluates the impact of program architectures and optimizers on quality and cost. |
Fine-Tuning and Prompt Optimization: Two Great Steps that Work Better Together (2024.emnlp-main)
Copied to clipboard
| Challenge: | Recent work shows the potential of building more powerful Natural Language Processing systems by composing multiple skills of LMs into pipelines. |
| Approach: | They propose to combine weight and prompt optimization strategies to optimize a modular LM pipeline. |
| Outcome: | The proposed strategies outperform optimizing weights and prompts alone by 60% and 6% on average across LMs and tasks. |