Explore a comprehensive video explanation of the paper "Transformers are RNNs: Fast Autoregressive Transformers with Linear Attention." Delve into the reformulation of the attention mechanism using kernel functions, resulting in a linear formulation that reduces computational and memory requirements. Discover the surprising connection between autoregressive transformers and RNNs. Learn about softmax attention, quadratic complexity, generalized attention mechanisms, kernels, linear attention, and experimental results. Gain insights into the intuition behind linear attention and understand the caveats of the RNN connection. This 48-minute video by Yannic Kilcher breaks down complex concepts, making them accessible to those interested in AI, attention mechanisms, transformers, and deep learning.
Transformers Are RNNs- Fast Autoregressive Transformers With Linear Attention
Overview
Syllabus
- Intro & Overview
- Softmax Attention & Transformers
- Quadratic Complexity of Softmax Attention
- Generalized Attention Mechanism
- Kernels
- Linear Attention
- Experiments
- Intuition on Linear Attention
- Connecting Autoregressive Transformers and RNNs
- Caveats with the RNN connection
- More Results & Conclusion
Taught by
Yannic Kilcher
Related Courses
-
Feedback Transformers - Addressing Some Limitations of Transformers with Feedback Memory
-
Linear Transformers Are Secretly Fast Weight Memory Systems - Machine Learning Paper Explained
-
Recurrent Neural Networks and Transformers
-
RWKV- Reinventing RNNs for the Transformer Era
-
Rethinking Attention with Performers
