Inf368 - preliminary plan

The use of deep neural networks has revolutionized many machine learning tasks, including image analysis (object recognition and location), natural language processing, and game playing.  This course will give practical and theoretical overview of some of the current methods.

The course will focus on practical and theoretical aspects of modern deep learning neural networks.  We will use real large scale data sets, and cloud-based GPU computing resources, and use standard Python-based frameworks like Tensorflow and Keras.

The core of the course will be a series of exercises where the students develop solutions for data classification and segmentation using deep convolutional networks, and explore new techniques like generative adversarial networks and siamese networks.  Traditional lectures will be supplemented with online resources, and students will be expected to independently explore and make use of online information. Grading will be based on a combination of exercises and final exam.

Machine learning/neural networks (Inf283/261 or similar), linear algebra and calculus, Python programming.

- Deep learning classification of images

- Object location and segmentation

- Vector space embeddings

- Generative adversarial networks