Model Influence Analysis
Sameer discusses the nuances between influence functions and nearest neighbors in model analysis. Nathan questions the effectiveness of using nearest neighbors on final layer representations, prompting a comparison with influence functions. The conversation delves into the importance of understanding what makes an instance influential in model predictions, especially in the context of pre-trained language models like Bert.In this clip
From this podcast
NLP Highlights
117 - Interpreting NLP Model Predictions, with Sameer Singh
Related Questions