Overview

Deep Learning Fundamentals is your step-by-step path into feedforward neural networks and the core ideas that make deep models actually train, generalize, and stay stable.

Classification Using Neural Networks

From linear models to multilayer perceptrons

Move from classic ML to neural networks by using fully connected layers for classification. Connect activation functions, loss functions, and backpropagation to the project's implementation.

You'll learn: Perceptrons and MLPs, activations (ReLU, sigmoid, softmax), cross-entropy loss, forward/backward pass intuition, basic multiclass classification with neural nets

Optimization

Make training actually converge

Go beyond "it trains" by understanding how optimizers and learning rate choices shape your loss curve, mirroring the Optimization project tasks. Learn to debug training runs instead of guessing.

You'll learn: Gradient descent variants (SGD, momentum, RMSProp, Adam), learning rate schedules, loss landscapes, exploding/vanishing gradients, practical tips for stable training

Error Analysis

Debug models with data, not vibes

Use systematic error analysis to understand why your model fails, directly supporting the Error Analysis project. Learn to look at examples, slices, and metrics before changing architectures.

You'll learn: Confusion matrices revisited, per-class and per-slice analysis, common failure patterns, building checklists, prioritizing fixes that actually move metrics

Regularization

Control overfitting in deep networks

Connect regularization theory to the Regularization project, seeing how each technique changes training and validation curves in practice. Learn to keep models flexible but not memorizing.

You'll learn: L1/L2 weight penalties, dropout, early stopping, data-related regularization, capacity vs. overfitting, reading learning curves to choose regularization strength

Treat each project as a lab: run small experiments, track metrics, and use error analysis and regularization to turn "it works" models into reliable baselines before you move on to CNNs.