Abstract: While large deep learning models have become increasingly accurate, concerns about their (lack of) interpretability have taken a center stage. In response, a growing subfield on interpretability and analysis of these models has emerged. While hundreds of techniques have been proposed to “explain” predictions of models, what aims these explanations serve and how they ought to be evaluated are often unstated. In this talk, I will first present a framework to quantify the value of explanations, which allows us to compare different explanation techniques. Further, I will highlight the need for holistic evaluation of models, sharing two tales on (i) how geographically representative are artifacts produced from text-to-image generation models, and (ii) how well can conversational LLMs challenge false assumptions?

Bio: Danish Pruthi is an Assistant Professor at the Indian Institute of Science (IISc), Bangalore. He received his Ph.D. from the School of Computer Science at Carnegie Mellon University. He is broadly interested in the areas of natural language processing and deep learning, with a focus towards inclusive development and evaluation of AI models. He completed his bachelors degree in computer science from BITS Pilani, Pilani. He is also a recipient of the Schmidt Sciences AI2050 Early Career Fellowship, Siebel Scholarship, CMU Presidential Fellowship and industry awards from Google and Adobe Inc. Until recently, his legal name was only Danish—an “edge case” for many deployed NLP systems, leading to airport quagmires and, in equal parts, funny anecdotes.