Jean-Baptiste Tristan & Michael L. Wick, Harvard University
Time: Mon/Wed 1:30-2:45pm
Location: MD G115
| Date | Instructor | Area | Topic | Reference | Assignment | Project |
|---|---|---|---|---|---|---|
| Sep. 4 | Wick | Foundations | Introduction 1 |
|
||
| Sep. 9 | Wick | Introduction 2 |
|
A1 out | ||
| Sep. 11 | Tristan | Learning theory 1 |
|
|||
| Sep. 16 | Tristan | Learning theory 2 |
|
|||
| Sep. 18 | Wick | Optimization-based ML 1 |
|
|||
| Sep. 23 | Wick | Optimization-based ML 2 |
|
A1 due, A2 out | ||
| Sep. 25 | Tristan | Deep learning | Stochastic Gradient Descent |
|
||
| Sep. 30 | Tristan | Optimization for neural networks 1 |
|
|||
| Oct. 2. | Tristan | Optimization for neural networks 2 |
|
|||
| Oct. 7. | Wick | Convolutional neural networks |
|
A2 due, A3 out | ||
| Oct. 9 | Midterm |
|
||||
| Oct. 14 | Columbus day |
|
||||
| Oct. 16 | Wick | Convolutional neural networks |
|
|||
| Oct. 21 | Wick | Recurrent neural networks |
|
A3 due | ||
| Oct. 23 | Wick | Recurrent neural networks |
|
|||
| Oct. 28 | Tristan | Probabilistic machine learning | Bayesian machine learning |
|
||
| Oct. 30 | Tristan | Markov Chain Monte Carlo 1 |
|
|||
| Nov. 4 | Tristan | Markov Chain Monte Carlo 2 |
|
Proposal due | ||
| Nov. 6 | Tristan | Variational inference |
|
|||
| Nov. 11 | Wick | Deep learning | Deep learning and linguistics |
|
||
| Nov. 13 | Wick | Deep Q-Learning |
|
|||
| Nov. 18 | Swetasudha Panda | Invited talk | RL and game theory |
|
||
| Nov. 20 | Rediet Abebe | Invited talk | ML and sociology |
|
||
| Nov. 25 | Reading exam |
|
Status report due | |||
| Nov. 27 | Thanksgiving |
|
||||
| Dec. 2 | Project feedback |
|
||||
| Dec. 16 |
|
Paper due |
| Week | Type | Topic | TF | Reading |
|---|---|---|---|---|
| Sep. 9 | Review | Math | Celine and William | |
| Sep. 16 | Review | Python/Scikit-learn | Zach and Jason | |
| Sep. 23 | Reading | William | ImageNet:A Large-Scale Hierarchical Image Database | |
| Sep. 30 | Reading | Lev | Adaptive Methods for Nonconvex Optimization | |
| Oct. 7 | Review | Pytorch | zach | |
| Oct. 21 | Reading | Zach | Neural Machine Translation by Jointly Learning to Align and Translate | |
| Oct. 28 | Reading | Yuting | Conditional Random Fields:Probabilistic Models for Segmenting and Labeling Sequence Data | |
| Nov. 4 | Reading | Jason | Human-level concept learning through probabilistic program induction | |
| Nov. 11 | Reading | Celine | Hierarchical Dirichlet Processes | |
| Nov. 18 | Reading | Lev | Playing Atari with Deep Reinforcement Learning |
| Date | Location | Topic | Materials |
|---|---|---|---|
| Sep. 1, 10-11am (Mark) | Pierce 301 | Math Review (Linear Algebra, Calculus, Probabilistic Theory) | |
| Sep. 4, 5-6pm (Zhirui) | Pierce 301 | Math Review (Linear Algebra, Calculus, Probabilistic Theory) | |
| Sep. 7, 5-6pm (Rachit) | Pierce 320 | Code Review (Python, Numpy, Matplotlib, PyTorch) | |
| Sep. 8, 11-11:59am (Rachit) | MD 223 | Code Review (Python, Numpy, Matplotlib, PyTorch) |
The objective of this final project is to explore new research in machine learning. The ideal outcome would be a paper that could be submitted a top machine learning conferences.
You should work in a group of 3 or 4 people. Groups of 2 will be considered with permission. Larger groups will not be permitted.
There are three deliverables.
Write a two-page document describing the plan for your project. This should clearly state what problem you are trying to solve. If you have developed a new model, explain what models this work will build on and how it resolves deficiencies. If it is a new algorithm for inference, explain the regimes for which you think it well be well-suited. If you are developing a new theoretical contribution, discuss the theorems you will prove. For problem-driven papers, discuss the data and the unique challenges that make this interesting. Identify relevant work and algorithms you intend to implement as baselines. This does not need to be a comprehensive document and I expect that it will be speculative. Your focus should be on identifying the questions you wish to answer about your data or your method and specifying clearly what success will mean.
Write a four-page document to describe the status of your project. What have you proved? What baselines have you established? Have there been unexpected results, good or bad?
Using the NeurIPS conference paper format (available at http://nips.cc), write a paper of up to ten pages. This paper should have a typical conference style, with abstract, introduction, etc. You should clearly state what problem you are trying to solve, introduce and explain your approach, and review the relevant literature. It should explain in detail the experiments that were run, show their results and discuss conclusions that can be drawn.