Deep Learning Lab

My Deep Learing course assignments.

Content

• Lab1: Implements core NumPy operations and basic image processing including grayscale conversion, Sobel edge detection, and patch shuffling using matrix manipulation.
• Lab2: Demonstrates polynomial regression with gradient descent optimization, comparing underfitting/overfitting scenarios and L2 regularization effects through synthetic sinusoidal data.
• Lab3: Builds linear models (LSM/Ridge Regression) and compares them with scikit-learn/numpy implementations, extending to polynomial feature space for nonlinear regression.
• Lab4: Implements fundamental classification algorithms - logistic regression (binary), softmax regression (multi-class), and perceptron - evaluated on breast cancer and iris datasets with accuracy metrics.
• Lab5: Implements gradient descent optimization for functions y=x² (starting from x=1) and z=x²+y² (starting from x=1,y=3), with visualization of gradient descent trajectories.
• Lab6: Manually implements a three-layer neural network with backpropagation algorithm, then replicates using PyTorch framework. Compares performance differences between manual and PyTorch implementations on Breast Cancer dataset through training/testing metrics.
• Lab7: Implements classical convolutional neural network models (LeNet-5 and AlexNet) for image classification. Trains and evaluates models on MNIST and CIFAR-10 datasets, comparing architectural differences and performance characteristics.
• Lab8: Implements RNN, LSTM, and GRU for MNIST image classification and IMDB sentiment analysis (score prediction). Compares basic RNN versions with custom "Hymmn0s" variants (e.g., incorporating embeddings for IMDB) and evaluates with accuracy (MNIST) and MAE/sentiment accuracy (IMDB).
• Lab9: Implements and visualizes optimization algorithms (AdaGrad, RMSprop, AdaDelta, Momentum, NAG, Adam) on saddle point ( $f(x,y) = x^2 - y^2$ ) and sharp minimum ( $f(x,y) = A(1 - e^{-(x^2+y^2)/k})$ ) functions to compare their navigation of non-convex landscapes.

Usage

This project uses uv as a Python package manager. Setup instructions:

Prerequisites

• Python 3.x
• uv installed (curl -LsSf https://astral.sh/uv/install.sh | sh)

Setup

# Clone repository
git clone https://github.com/timedegree/Deep-Learning-Lab.git
cd Deep-Learning-Lab

# Create and activate virtual environment
uv venv
source .venv/bin/activate  # Linux/macOS
# or . .venv\Scripts\activate  # Windows

# Install dependencies from pyproject.toml
uv sync