Model Training Insights
Alvin discusses the challenges of training translation models using standard web corpora, highlighting the limitations of non-simultaneously translated data. He also delves into the complexities of deep learning models, emphasizing their "black box" nature and the unexpected behaviors they exhibit, like providing high confidence in nonsensical inputs. This conversation sheds light on the importance of understanding model interpretability in the context of AI advancements.In this clip
From this podcast
The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence)
Pathologies of Neural Models and Interpretability with Alvin Grissom II - TWiML Talk #229
Related Questions