Hey everyone! 👋
Welcome to the AI & ML Society for the new academic year! Our events will be starting again in November.
1. Technic Sessions 🧑💻 We'll dive into state-of-the-art AI (selected content for fundamental algorithms in generative models, LLM and embodied intelligence includes Transformers, PPO, neural operator, flow matching, PINN and application in virtual embodied intelligence). We'll together watch a tutorial or read a paper, then code the main ideas and discuss. The total estimated time for each session would be 1 hour.
Sessions
Time & Where
Content
Plan
1
Monday, 17th Nov, LH small common room
One foundation in Embodied intelligence and Large Language Model (e.g GPT):
Reinforcement learning and Proximal Policy Optimization Algorithms
Reading:
RL concept 5 mins and discuss 5 mins:
https://spinningup.openai.com/en/latest/spinningup/rl_intro.html
Policy Optimization 10 mins and discuss 5 mins:
https://spinningup.openai.com/en/latest/spinningup/rl_intro3.html
PPO paper reading 15 mins and discuss 5 mins:
https://arxiv.org/abs/1707.06347
Aim: understand the objective function
Coding 15 mins:
Share one example of PPO in character animation
2
Monday, 24th Nov, LH small common room
Virtual embodied intelligence of single agent control
https://arxiv.org/abs/1707.02286
Reading:
5 mins see the research problem.
And discuss if the RP is old or new, if old what’s the improvement. What’s their contribution;
30 mins see the details of the method, evaluation, and think limitation.
And discuss 5 mins, what’s the input and output of the frame work, how many steps in the frame work, what’s the details of their method, what’s limitation, what’s novalty;
Coding 15mins:
Share my reimplement
3
Monday, 1th Dec, LH small common room
Virtual embodied intelligence of multiple agent control
https://www.science.org/stoken/author-tokens/ST-751/full
https://arxiv.org/abs/1707.02286
Reading:
5 mins see the research problem.
And discuss if the RP is old or new, if old what’s the improvement. What’s their contribution;
30 mins see the details of the method, evaluation, and think limitation.
And discuss 5 mins, what’s the input and output of the frame work, how many steps in the frame work, what’s the details of their method, what’s limitation, what’s novelty;
Coding 15mins:
Share my reimplementation
4
Monday, 8th Dec, LH small common room
Generative model - Flow matching
(theory session)
https://neurips.cc/virtual/2024/tutorial/99531
Watching the tutorial together and discuss
Flow matching basics
Advanced design of flow matching
5
Monday, 15th Dec, LH small common room
Generative model - Flow matching
(application session)
https://neurips.cc/virtual/2024/tutorial/99531
Watching the tutorial together and discuss
Flow matching basics
Advanced design of flow matching
Code
6
Term 2, TBD
Solving PDE From data and Fourier Neural Operator
https://www.youtube.com/watch?v=0Ve9xwNJO2o&t=985s
Watching the tutorial from 1st author of FNO
Coding 5 mins:
Share my reimplementation
7
Term 2, TBD
Solving PDE by Physics informed neural network (PINN)
https://arxiv.org/abs/2412.09009
TBD, better to find a good tutorial video instead of reading
8
Term 2, TBD
One foundation in Large Language Model (e.g GPT)
Transformer
Two videos from 3blue1brown
1. https://www.3blue1brown.com/?v=gpt
2. https://www.3blue1brown.com/?v=attention
Share my coding for 5 mins.
Monday 01 December 2025 7:30pm - 8:30pm
LH Small Common Room