Li Hongyi's 2022 course videos are all given in the form of online videos (all have been recorded, you can choose to finish them all in a short time), during class time there will be live broadcasts to explain additional content (you can not listen) and homework (recommended to do) ), which is currently updated to assignment 9.
The main content of the ninth lesson is Explainable AI, the explainability of deep learning . It mainly introduces local explanation and global explanation. The two methods analyze the interpretability of a single sample and the interpretability of the model itself. See the course video for details.
-
Course video :
Video number of station b : machine learning craftsman
URL : https://www.bilibili.com/video/BV1Z34y1C7sj?spm_id_from=333.999.0.0
-
Github : class and homework courseware and basic code
URL : https://github.com/yaoweizhang/LHY2022-SPRING
Assignment 9 requires the use of hw3 and hw7 materials. You can refer to previous articles for how to obtain the materials. At the end of the article, there is the teaching assistant code for this assignment and how to obtain the assignment introduction document.
-
Job task:
Interpretability of CNNs. Using the Food Classification model and data of hw3, there are a total of 20 questions, including Lime, Saliency Map, Smooth Grad, Filter Visualization, and Integrated Gradients.
BERT Interpretability. Using hw7's BERT model related information, there are a total of 10 questions, including Attention Visualization, Embedding Visualization, and Embedding Analysis.
-
Submit address :
gradescope, non-course students can not use it.
Deadline :
2022/05/20 23:59 (Beijing time).
-
QQ exchange group : 156013866
Assignment 9 Assistant code and data acquisition methods:
-
Follow the WeChat public account " Machine Learning Craftsman "
-
Background reply keywords: 202209