On the Role of Performance Prediction for Adaptive RAG Workflows
Dr. Debasis Ganguly
University of Glasgow | School of Computing ScienceQuery Performance Prediction (QPP) has mainly been studied for more than two decades as a focused sun-topic in Information Retrieval. The community has worked towards developing more effective models for predicting the performance of a diverse range of ranking models, and in this talk I’m going to provide a very brief review of the existing classes of QPP models from unsupervised to supervised to query-variant based approaches. However, a more important problem to which I want to draw the attention of the research community is that of a downstream application of QPP in developing adaptive systems. I’m going to talk about two distinct use-cases - the first, a more direct application of QPP for improving the workflow of IR systems by incorporating a dynamic query-specific pipeline. And the second, a more subtle connection between QPP and RAG, where I will be first talking about how QPP techniques might be used to estimate the usefulness of a RAG context eventually providing some high-level pointers on how this might actually be applied to develop input-specific adaptive RAG pipelines.
Leveraging Temporal Images for Biomedical Radiology Analysis
Xi Zhang
University of Glasgow | School of Computing ScienceA temporal-aware multimodal method for radiology report generation that leverages paired chest X-rays to understand disease progression and address key challenges in medical AI.
