Uploaded September 2020 | Updated November 2026, 5 hours ago
Github repo: github.com/luisguiserrano/singular_value_decomposition
Grokking Machine Learning Book:
manning.com/books/grokking-machine-learning
40% discount promo code: serranoyt
In this video, we learn a very useful matrix trick called singular value decomposition (SVD), in which we express a matrix as a product of two rotation matrices and one scaling matrix.
We also show a very interesting application to image compression.
Similar videos:
Principal component analysis (PCA): youtube.com/watch?v=g-Hb26agBFg
Matrix factorization and Netflix recommendations: youtube.com/watch?v=ZspR5PZemcs
Introduction: (0:00)
Transformations: (0:50)
A puzzle: (1:27)
A harder puzzle: (2:21)
Linear transformations: (3:50)
Dimensionality reduction: (10:50)
Image compression: (23:57)
Github repo: github.com/luisguiserrano/singular_value_decomposition
Grokking Machine Learning Book:
manning.com/books/grokking-machine-learning
40% discount promo code: serranoyt
In this video, we learn a very useful matrix trick called singular value decomposition (SVD), in which we express a matrix as a product of two rotation matrices and one scaling matrix.
We also show a very interesting application to image compression.
Similar videos:
Principal component analysis (PCA): youtube.com/watch?v=g-Hb26agBFg
Matrix factorization and Netflix recommendations: youtube.com/watch?v=ZspR5PZemcs
Introduction: (0:00)
Transformations: (0:50)
A puzzle: (1:27)
A harder puzzle: (2:21)
Linear transformations: (3:50)
Dimensionality reduction: (10:50)
Image compression: (23:57)

![Low Rank Adaptation (LoRA) - How to search for a needle in a much smaller haystack
When you want to fine-tune a model with many parameters and you want to do it fast and effectively, Low Rank Adaptation (LoRA) is there to help you.
With LoRA, you trust that the base model is in a good state, and so you assume there may already be many dependencies among the parameters. Thus, you look for better parameters in a much smaller dimensional space.
Learn all about LoRA in this friendly video!
CORRECTION: (thanks @ryanhewitt9902!) At 30:14 the vector on the right should be [pa, qa, ra, sa, ta], instead of [p/a, q/a, r/a, s/a, t/a]. Low Rank Adaptation (LoRA) - How to search for a needle in a much smaller haystack](https://i.ytimg.com/vi/Gy9jrVQTY4Q/mqdefault.jpg)








