MyTorch - Deep Learning Framework

A custom PyTorch-inspired framework built from scratch using Python and NumPy.

Functionalities

Core Components

• Neural Network Layers

  • Linear Layers
  • Convolutional Layers
  • Recurrent Layers (LSTM, GRU)
  • Attention Mechanisms
    • Self-Attention
    • Multi-Head Attention
    • Temporal Attention

• Training Components

  • Optimizers (SGD, Adam, RMSprop, AdaGrad)
  • Loss Functions
  • Activation Functions
  • Regularization (Dropout, BatchNorm, LayerNorm)

Architecture Implementations

• MLPs (Multi-Layer Perceptrons)
• CNNs (Convolutional Neural Networks)
• RNNs & LSTMs
• Transformer Models
• CNN-LSTM Hybrid Models

Course Structure

Homework Assignments

• HW1: Custom Neural Network Implementation

  • Basic NN components (Linear layers, Activations)
  • Optimizers

• HW2: Computer Vision

  • MLP and CNN implementations
  • Vision model training

• HW3: Speech Recognition

  • CNN-LSTM architecture
  • Speech processing and recognition

• HW4: Language Models & Speech

  • P1: Language Modeling with LSTM
  • P2: Transformer-based Speech Recognition