Mastering Neural Networks and Model Regularization

Projects You’ll Build (With Expert Guidance)

With ready-to-use Jupyter notebooks and working code examples, Dr. Guven will walk you through creating…

Neural Network from Scratch
Construct a multilayer perceptron using only NumPy and PyTorch tensors—manually coding forward propagation, backpropagation, and gradient descent.
Then train your model on the MNIST dataset to classify handwritten digits, reinforcing a deep, code-level understanding of how neural networks learn.

Image Classification with Convolutional Neural Network
Use PyTorch to build a CNN that classifies MNIST digits with high accuracy by detecting strokes, edges, and loops. You’ll learn how convolutional layers extract spatial features and why CNNs outperform fully connected networks on image tasks.

Audio Classification with Spectrograms
Transform raw audio into spectrogram images and train a CNN to classify environmental sounds. You’ll experiment with dropout, batch norm, and architecture tuning to build models that handle messy, real-world data.

Scene Recognition with Neural Networks
Train both a fully connected network and a CNN to classify natural scene images—and see why spatial features matter. This side-by-side comparison highlights how architecture impacts accuracy, especially on complex image data.

Stack Overflow Code Classifier
Use TF-IDF features and a PyTorch neural network to predict the programming language behind Stack Overflow posts. You’ll build a multi-class text classifier and compare it to an SVM to learn when deep learning pays off—and when it might not.