
[19] Dumitru Erhan - Understanding Deep Architectures and the Effect of Unsupervised Pretraining
The Thesis Review
00:00
Visualizing Pre-training Challenges
This chapter explores the evolution and challenges of visualizations in unsupervised pre-training, particularly in computer vision and natural language processing. The speakers discuss optimization intricacies in deep learning, including the concept of basin of attraction and the impact of experimental strategies on model development. Additionally, the chapter examines the limitations of saliency maps in machine learning, questioning their true effectiveness in providing insights into model behavior.
Transcript
Play full episode