Overview
Neural networks have re-emerged as the dominant technology in machine learning. Such models now achieve “state-of-the-art” results on important tasks (e.g., translation; self-driving car navigation) and arguably less important ones (playing video games).
This course will cover the basics of neural networks and standard architectures, with an emphasis on the use of modem programming frameworks (fensorFlow + Keras, PyTorch) for defining and fitting such models. In general the emphasis will be on practical, implementation aspects.
Format
Lectures will cover the basics of neural networks (starting with supeNised learning and Qradient descent, and building up to backpropagation), and implementations the aforementioned frameworks.
Homeworks will mostly be hands-on implementation of models. For this we will rely on GooQle colab which provides free cloud access to GPUs via Jupyter notebooks. The course will culminate in a final project, affording an opportunity to explore novel uses of state of the art neural networks.
Pre-reqs
DS 4400 is a co-req, or permission of instructor.
Instructor
Byron Wallace
Website
ht1o://www .byronwallace .com/DS4440