teaching

DS 4440 Practical Neural Networks

  • Time: MW 2:50 pm - 4:30 pm
  • Location: Forsyth Building 236
  • Lecturer: Wengong Jin
  • TA: Chaitanya Agarwal (agarwal.cha@northeastern.edu)

Course Description

Offers a hands-on introduction to modern neural network (“deep learning”) methods and tools. Covers fundamentals of neural networks and introduces standard and new architectures from simple feedforward networks to recurrent and transformer architectures. Also covers stochastic gradient descent and backpropagation, along with related parameter estimation techniques. Emphasizes using these technologies in practice, via modern toolkits. Reviews applications of these models to various types of data, including images and text.

Resources

Grading

  • Attendance (10%)
  • Four project-style homeworks
    • HW1: complete PyTorch Lab
    • HW1 will not be graded. If you haven’t used PyTorch before, you must finish HW1 by the end of September. Otherwise, you won’t be able to work on HW2-HW4
    • HW2-HW4 (30% each)
    • HW2 will be released by the end of September (stay tuned)

Schedule (tentative)

Date Lecture
9/3 Introduction
9/8 Feedforward Neural Networks
9/10 Neural Network Training
9/15 Neural Network Regularization
9/17 Convolutional Neural Networks
9/22 Recurrent Neural Networks
9/24 Transformers
9/29 State space models (SSM)
10/1 Graph Neural Networks
10/6 Equivariant Neural Networks
10/8 Generative Models (Overview + autoregressive)
10/13 No class (Colombus Day)
10/15 Variational Autoencoders
10/20 Generative Adversarial Networks
10/22 Diffusion Models
10/27 Large Language Models 1
10/29 Large Language Models 2
11/3 Large Language Models 3
11/5 Reinforcement Learning 1
11/10 Reinforcement Learning 2
11/12 Explainable AI 1
11/17 Explainable AI 2
11/19 Adversarial Attack
11/24 Neural Network Pruning
11/26 No class (Thanksgiving)
12/1 AI for science
12/3 AI for healthcare
12/8 TBD
12/10 TBD