Chapter
The Common Thread in Modern Machine Learning Methods
42:53 - 53:07 (10:14)
This podcast explores how modern machine learning methods such as GANs, VAEs, and contrastive models are related through a common language and energy function.
Clips
The structure in language makes it easier to gather related meaning across different words and their context, compared to gathering related meaning in different images of the same object due to lighting differences.
42:53 - 46:30 (03:36)
Summary
The structure in language makes it easier to gather related meaning across different words and their context, compared to gathering related meaning in different images of the same object due to lighting differences.
ChapterThe Common Thread in Modern Machine Learning Methods
Episode#206 – Ishan Misra: Self-Supervised Deep Learning in Computer Vision
PodcastLex Fridman Podcast
The idea is that machine learning models, such as GANs or contrastive models, can all be explained by an energy function they're trying to minimize or maximize.
46:30 - 51:58 (05:28)
Summary
The idea is that machine learning models, such as GANs or contrastive models, can all be explained by an energy function they're trying to minimize or maximize. By using a common language for these models, their similarities can be better understood and used to achieve higher levels of performance.
ChapterThe Common Thread in Modern Machine Learning Methods
Episode#206 – Ishan Misra: Self-Supervised Deep Learning in Computer Vision
PodcastLex Fridman Podcast
Engineering tricks are used to determine the relationship between two things in contrastive learning for self-supervised and supervised learning, such as determining if two words are related if they are in the same context, or if two crops from the same image are related.
51:58 - 53:07 (01:09)
Summary
Engineering tricks are used to determine the relationship between two things in contrastive learning for self-supervised and supervised learning, such as determining if two words are related if they are in the same context, or if two crops from the same image are related.