Interpreting Deep Neural Networks Towards Trustworthiness
Explore a 46-minute lecture on interpreting deep neural networks for trustworthiness, presented by Bin Yu for the International Mathematical Union. Delve into the concept of interpretable machine learning and learn about the agglomerative contextual decomposition (ACD) method for neural network interpretation. Discover the adaptive wavelet distillation (AWD) technique, an extension of ACD into the frequency domain, and its applications in cosmology and cell biology predictions. Examine the importance of a quality-controlled data science life cycle for building trustworthy interpretable models, and understand the Predictability Computability Stability (PCS) framework. Access accompanying presentation slides to enhance your understanding of these complex topics in deep learning and interpretability.