Deep Learning with PyTorch

Go from zero deep learning knowledge to building your own image classifier in just three days. This workshop takes you on a hands on journey: you will understand how neural networks actually work under the hood, learn PyTorch as your tool of choice, train models on real data, and finish by building a complete CNN pipeline that classifies images. Every concept is introduced from scratch with live code examples. No prior machine learning experience required.

Day 1: Understanding Neural Networks and Your PyTorch Toolbox

Morning: Neural Networks from First Principles. What is a neuron? How do layers of simple math operations learn to recognize patterns? You will build a tiny neural network by hand (just Python and NumPy) so you truly understand forward passes, activation functions, and backpropagation before any framework magic.

Afternoon: PyTorch Fundamentals. Now that you know what happens under the hood, you will learn PyTorch as your power tool. Tensors (think: multi dimensional arrays on steroids), automatic differentiation (autograd does the calculus for you), and nn.Module (the building block for every model you will ever write).

Day 2: Training Real Models

Morning: Training Neural Networks. You will learn how models actually learn: loss functions (how wrong is my prediction?), gradient descent (how do I get less wrong?), and overfitting (why does my model memorize instead of generalize?). Lots of live experiments where you watch training curves in real time.

Afternoon: Build Your First MNIST Classifier. Time to put it all together. You will load the famous handwritten digits dataset, build a neural network from scratch in PyTorch, train it, evaluate it, and see it recognize digits you draw yourself. This is your "hello world" moment in deep learning.

Day 3: Convolutional Neural Networks for Images

Morning: CNNs for Image Understanding. Why do regular neural networks struggle with images? You will learn about convolutions (sliding filters that detect edges, textures, shapes), pooling (making your model robust to position changes), and how to stack these into architectures that "see" images the way humans do.

Afternoon: Building Image Training Pipelines. The grand finale. You will build a complete image classification pipeline: loading real world image datasets, applying data augmentation (flip, rotate, color jitter to get more training data for free), training a CNN, evaluating performance, and using transfer learning to get state of the art results with minimal data.