Exam Information Inroduction to Deep Learning Oliver Schulte

Exam Information Inroduction to Deep Learning Oliver Schulte

Topics Included • Everything up and including basic RNN model (not LSTM, GRU) • Both material from the slides and the book • Check lecture schedule for details on the book

Topics Excluded • Gates in RNNs • Word Embeddings • Keras

Sample Topics Overview (I) • Linear Classifiers (about 20%) – General Form – Loss Functions – Gradient Descent training – Non-separability • Feedforward Neural Networks (about 20%) – Activation functions – Backpropagation and gradient descent • Tensorflow Basics (Ch. 2) (about 10%)

Sample Topics Overview (II) • Training topics (about 25%) – regularization, drop-out, batch normalization – Babysitting the learning process, evaluating the loss function • Convolutional Neural Networks (about 20%) – Grid data 1 D, 2 D – Basic principles: filters, hierarchical features • Recurrent Neural Networks (about 5%) – Hidden Markov Model – RNN with hidden state = hidden layer

Format • 50 -60 min in-class • You can use a cheat sheet: 1 (one) 8. 5 x 11 sheet of notes for reference, both sides. • Closed book, no electronics. • Mix of – – – Multiple choice Max the concept Compute an answer Short answer Recognition (e. g. what does this piece of code do? )