- Paper discussion seminars
- PyTorch Workshop for General ACM Members (TBD, not sure if it's happening or not)
- Aim for Spring EOH Projects
- Welcome icebreaker.
- Split up into Vision/NLP/RL groups to discuss plans for the semester.
- Some people decided to work on Stanford's Computer Vision (CS231n), Berkeley's Reinforcement Learning (CS 294).
- Split into groups for discussion about various projects, targeting Spring 2017 Open House
- RL Group went over Berkeley's CS 294 on Sunday (8/17)
- Dominic talked to the Computer Vision Group about his research
- Computer Vision began to learn Tensorflow, having a workshop on Thursday (8/21)
- Reinforcement Learning Group is having a PyTorch Workshop/Lecture, replacing CS 294's Tensorflow Lecture
- NLP has began to talk about what projects to potentiall work on (speech synthesis/chat bot)
- Note the presentation had a bug with LeNet. Both the slides and examples have been updated.
- Presentation GitHub
- Presentation Slides
- Covered the following topics:
- Tensorflow Session, Variables, Placeholders, Graphs, and Ops.
- Supervised Learning Review
- Typical Loss Functions
- Linear Regression/Classification
- Implementing LeNet
- Meeting Slides
- Display feedback results -- hopefully we have 1111 this time.
- Tensorflow workshop review.
- Will gave a presentation about his research (trains and using inverse reinforcement learning to derive a human-like reward function).
- Dominic gave a presentation about SSD (Single Shot Detectors).
- NLP Group will be giving assignments to members, as a start to their project
- Meeting Slides
- Brief Discussion about the Intergroup Discussions planned for around November
- Some points brought up (in addition to the slides)
- People need prior knowledge to understand some of the papers, prereqs should be listed somewhere
- Most people don't know how to read papers, we'll need a good split on the people that can actually read vs. people that can.
- RL Group Covered:
- Value Function and on-policy/off-policy review.
- Covered Q-Learning, SARSA, Double Q-Learning.
- Advantages and Disadvantages of on/off policy
- Some intuition on how to actually program Q-Learning. (Small introduction on epsilon greedy, mentioned Experience Replay, Hyperparameter Selection, all the gross stuff)
- NLP Group Covered: