Deep Learning Notes

Backpropagation

• Backpropagation on scalars from scratch
• Manual Backpropagation on tensor

Loss function

• Maximum likelihood estimate as loss function
• Why we add regularization to loss functioñ

Optimization

• Optimization Algorithms (SGD with momentum, RMSProp, Adam)
• Optimizing loss with weight initialization
• BatchNormalization
• RMSNorm
• Diagnostic tool to look out for while training NN
• Skip Connections

Training

Misc

• Matrix Visualization
• SwiGLU activation- not mine, but offers best explanation

Architecture Implementation

• GPT implementation
• MoE
• RoPE
• KV Cache and Grouped Query Attention

GPU

• Basic intro to GPU architecture