What Is Wrong with My Model? Identifying Systematic Problems with Semantic Data Slicing
Open Access
- 27 October 2024
- conference paper
- Published by Association for Computing Machinery (ACM)
- p. 306-318
Abstract
Machine learning models make mistakes, yet sometimes it is difficult to identify the systematic problems behind the mistakes. Practitioners engage in various activities, including error analysis, testing, auditing, and red-teaming, to form hypotheses of what can go (or has gone) wrong with their models. To validate these hypotheses, practitioners employ data slicing to identify relevant examples. However, traditional data slicing is limited by available features and programmatic slicing functions. In this work, we propose SemSlicer, a framework that supports semantic data slicing, which identifies a semantically coherent slice, without the need for existing features. SemSlicer uses Large Language Models to annotate datasets and generate slices from any user-defined slicing criteria. We show that SemSlicer generates accurate slices with low cost, allows flexible trade-offs between different design dimensions, reliably identifies under-performing data slices, and helps practitioners identify useful data slices that reflect systematic problems.Keywords
Funding Information
- NSF (National Science Foundation) (2106853, 2131477, 2206859)
- DOD U.S. Department of Defense (FA8702-15-D-0002)
This publication has 31 references indexed in Scilit:
- Improving Probability-based Prompt Selection Through Unified Evaluation and AnalysisTransactions of the Association for Computational Linguistics, 2024
- (Why) Is My Prompt Getting Worse? Rethinking Regression Testing for Evolving LLM APIs2024
- Zeno: An Interactive Framework for Behavioral Evaluation of Machine Learning2023
- Why Johnny Can’t Prompt: How Non-AI Experts Try (and Fail) to Design LLM Prompts2023
- What Did My AI Learn? How Data Scientists Make Sense of Model BehaviorACM Transactions on Computer-Human Interaction, 2023
- Improving generalizability of ML-enabled software through domain specification2022
- Auditing Algorithms: Understanding Algorithmic Systems from the Outside InFoundations and Trends® in Human-Computer Interaction, 2021
- ColBERT2020
- Bridging the gap between ML solutions and their business requirements using feature interactions2019
- Nuanced Metrics for Measuring Unintended Bias with Real Data for Text Classification2019