top of page

Deep Neural Networks

 A course in deep neural networks. Topics included:

  • Probability and Information Theory reviews

  • Feedforward networks

  • Convolutional networks (non-linear, pooling, and batch normalization layers; CV applications)

  • Recurrent networks (vanilla, LSTM, GRU; NLP applications)

  • Reinforcement Learning, Deep Q-Network

  • Auto encoders, manifold learning, sparsity

  • Graphical NNs, GANs, biologically inspired networks

  • Optimization (stochastic optimization, gradient descent with momentum, RMS prop, Adam)

  • Regularization (norm regularization, dropout, data augmentation)

  • Computation (backprop, TensorFlow, Keras)

The TensorFlow/Keras framework was used throughout this course

Problem Set Topics

Problem sets included:

  • implementing fully-connected, feedforward networks for function approximation

  • Handwritten digit (MNIST) classification

  • CNN implementation

  • Filter engineering to perform common image processing operations (e.g. grayscale conversion and edge detection)

  • ResNet implementation

  • RNN implementation

  • GRU implementation

  • Sentence modeling and text prediction with RNNs

  • Sentiment analysis via RNNs


A final paper was also a requirement of this course. I wrote mine on comparing architectures within the LeNet/AlexNet family of feedforward CNNs for classifying the CIFAR-10 dataset. The paper can be found below

Final Paper
bottom of page