CS677 - Deep Learning (for the Curious Mind)
Fall 2020
Heart arteries in 3D - curiosity lead to saving thousands of lives by combining deep learning and computational fluid dynamics.
What is Deep Learning (for the Curious Mind)
Deep learning for the curious mind, is a course that explores the methods and fascinating applications of deep neural networks. We build the subject from the ground up starting from the foundations of inductive & transductive learning and explain why stochastic gradient descent and back propagation (BP) were critical to successfully train deep neural networks. We will use tools such as regularization and hyperparameter optimization to help us train networks that have the best inferential performance when they face big numbers of high dimensional inputs. We will then learn two architectures that embed innate priors: Convolutional Neural Networks (CNN) spatially and Recurrent Neural Networks (RNN) temporally. They are key in enabling applications in computer vision and natural language processing where learning is all about building the right representations. This is often achieved using autoencoders that allow us to learn the useful properties of our data and generative modeling that allow us to generate data from the underlying target function. Equipped with these principles, we then start to combine deep learning with other disciplines to appreciate how they have disrupted three key application domains. (1) Combining deep learning and control lead to Deep Reinforcement Learning (DRL) and today’s disruption in industrial automation via robotic manufacturing. (2) Combining deep learning and linguistics lead to breakthroughs in Natural Language Understanding (NLU) and our ability to retrieve information with the right context. (3) Combining deep learning with computer vision lead to disruption in transportation enabling self-driving cars and trucks to safely drive themselves in a wide range of challenging environments. Drawing from these three application areas we strike a good balance between concepts and hands-on experiments and assign projects that are challenging and at the same time a lot of fun!.
Logistics
Time/location: Converged Learning - Look at your https://njit.back2classroom.app to determine whether you are scheduled for in-person or remote learning.
Communication: We use Slack for all communications: announcements and questions related to lectures and projects. Slack info will be sent to your NJIT email accounts. Please install Slack in your smartphones as well. Office hours with the TAs and professor will be done via Slack voice channels.
Instructor
Pantelis Monogioudis, Ph.D Professor of Practice, NJIT & Adjunct NYU
Teaching Assistant
Nitesh Mistry
Grading
- Midterm (25%)
- Final (35%)
- Projects (40%)