Explainable machine learning
Course Overview
Embark on a focused exploration of explainable machine learning. This course covers fundamental and advanced techniques to make models like neural networks more interpretable and explainable. Begin with the basics, understanding explainable machine learning and exploring interpretable models like linear regression and decision trees. Progress to visualizing model components of neural networks, employing DeConvNets, and mastering activation maximization. Delve into global model-agnostic methods, understanding attribution methods, feature importance, and tools like partial dependence plots, permutation feature importance, and prototypes. Transition to local model-agnostic methods, covering Shapley values, occlusion sensitivity maps, spectral clustering, RISE, and LIME. Conclude with advanced techniques like gradient-based saliency maps, CAM, Grad-CAM, layer-wise relevance propagation, TCAV, and activation space occlusion sensitivity. Enroll now to master the art of explainable machine learning and interpret complex models effectively.
Details
Lecturer
Overview
Embark on a focused exploration of explainable machine learning. This course covers fundamental and advanced techniques to make models like neural networks more interpretable and explainable. Begin with the basics, understanding explainable machine learning and exploring interpretable models like linear regression and decision trees. Progress to visualizing model components of neural networks, employing DeConvNets, and mastering activation maximization. Delve into global model-agnostic methods, understanding attribution methods, feature importance, and tools like partial dependence plots, permutation feature importance, and prototypes. Transition to local model-agnostic methods, covering Shapley values, occlusion sensitivity maps, spectral clustering, RISE, and LIME. Conclude with advanced techniques like gradient-based saliency maps, CAM, Grad-CAM, layer-wise relevance propagation, TCAV, and activation space occlusion sensitivity. Enroll now to master the art of explainable machine learning and interpret complex models effectively.
1st lecture: Introduction - Explainable machine learning
Topics covered in this video: • What is explainable machine learning? • Inherently interpretable models: linear regression and decision trees
2nd lecture: Looking into a neural network - Explainable machine learning
Topics covered in this video: • Visualizing model components (weights, activations) • DeConvNets • Activation maximization
3rd lecture: Global model-agnostic methods - Explainable machine learning
Topics covered in this video: • What are global model-agnostic methods and why are they useful? • Attribution methods and feature importance • Partial dependence plots • Permutation feature importance • Variance feature importance • Prototypes and criticism • Maximum mean discrepancy and witness function
4th lecture: Local model-agnostic methods - Explainable machine learning
Topics covered in this video: • What are local model-agnostic methods? • Shapley values • Occlusion sensitivity maps • Spectral clustering • Randomized input sampling (RISE) • Local interpretable model-agnostic explanations (LIME)
5th lecture: Interpret. by backward prop. + concept discovery - Explainable machine learning
Topics covered in this video: • Gradient-based saliency maps • CAM and Grad-CAM • Layer-wise relevance propagation • Testing with concept activation vectors (TCAV) • Activations space occlusion sensitivity
