Slides:

  1. Machine Learning Review
  2. Neural Networks and Optimization
  3. Neural Networks in Practice
  4. Convolutional Neural Networks
  5. Image Classification
  6. Network Architectures
  7. Segmentation and Detection
  8. Dealing with Sequential Data
  9. BERT/Transformers
  10. Optimizers
  11. Make it Work
  12. Deep Reinforcement Learning
  13. Generative Adversarial Networks (GANs)

Homeworks:

We will have 3 homework assignments, which will be listed below as they are assigned. The assignments will be given out every week starting week 2.

Undergraduates may work individually or in pairs. Graduate students must work individually. For pairs, only one needs to submit the assignment on Canvas.


Note that there is a deadline for each assignment. Anything uploaded after the deadline will be marked late. Please be careful to not overwrite an in time assignment with a late assignment when uploading near the deadline.

Each student has four penalty-free late days for the whole quarter; other than that any late submission will be penalized for each day (-5 per day up to -50%) it is late. Late group assignments will count as late for both students.

Please let the TA know if you cannot access any of the pages.


Turn in all homeworks on canvas.

Final Project:

There will be a final project worth 40% of your final grade. The project can be done individually or in teams of up to 3. For larger teams we expect larger projects. We will have a poster session in the CSE Atrium.

For your final project you should explore any topic you are interested in related to deep learning. This could involve training a model for a new task, building a new dataset, improving deep models in some way and testing on standard benchmarks, etc. You project should probably involve some implementation, some data, and some training. The amount of effort and time should be approximately 2 homework assignments.

Apart from the poster session, each group will turn in a 2-4 page summary of their project. This summary should mention the problem setup, data used, techniques, etc. It should include a description of which components were from preexisting work (i.e. code from github) and which components were implemented for the project (i.e. new code, gathered dataset, etc).

  • Project Proposals (10%) - Due 10/30/2019 11:59pm
  • Project Milestone (10%) - Due 11/20/2019 11:59pm
  • Poster Session (50%) - December 4th 3:30-6:30
  • Project Writeup (30%) - Due December 4th 11:59pm


Example Projects from last year:

  • Next frame prediction using an LSTM
  • Clothes type detector with live demo
  • Time series analysis of rat brain excitations
  • Detecting the sound of water with live demo
  • Automatic Colorization of Line Sketch images with live demo
  • Classifying Tweets as from a celebrity or not with live demo
  • Converting low fidelity wireframes into high fidelity digital design sketches with live demo
  • Signature verification with live demo
  • Predicting retweet counts with live demo
  • Knowledge Grounded Dialog System with live demo

Grading:

The final grade will consist of homeworks (60%) and a final project (40%)

Course Administration and Policies

  • All homework assignments can be completed in pairs for undergraduate students. Both students should contribute and understand all the material for each homework. In the graduate section (599G1) homeworks should be done individually.
  • Collaboration is encouraged! Feel free to discuss howemork and class material with other students. However, make sure you understand the concepts. Do not directly or indirectly copy other students' work.
  • If you are working together or helping another student, work on teaching them concepts and answering general questions, not directly telling them what code to write. You're all smart; you should understand the line between productive collaboration and giving someone answers.
  • Each student has four penalty-free late day for the whole quarter. Beyond that, late submissions are penalized.
  • Comments can be sent to the instructor or TA using this anonymous feedback form here. We take all feedback very seriously and will do whatever we can to address any concerns.