Day 1:

What I learned from Clare Lyle at Finding the Frame workshop:

1. Clare is in pretty good shape, comparison to her made realize I definitely should give my best to stay in a good really shape.
2. She does blogging here explaining math/reesrch materissl and also life lessons. It made me realize I should do that too to have a clearer mind
3. As of August 5th, 2025, I’m learning stuff that she learned in her 2nd year of undergrad. This made me realize that I should temper my expectations, be much more humble, try harder to acquire those knowledge too. We’ve roughly started our undergrad at the same time and compared to how I was behaving in my 2nd year, she is miles ahead and throughout the years the drift has exponentially grown; I shouldn’t expect to close the gap overnight. Can I close the gap? Let’s be straight, N0. But, the best time to plant a tree was 20 years ago. The second-best time is now.
4. She was asked: “How do you choose a topic/research question out of this huge world to do research on?” Her answer:

   I usually take something that works really well and test it on a setting that it doesn’t work well anymore. Then, I try to figure out why it isn’t working. This Approach results in a cycle of questioning and answering that leads to interesting research.

Mark Ho:

Psychology is hard because we can’t observe the mental state of people but only the outcome.

Day 2:

Leslie:

You can denote a histroty dependent policy by: $\pi: (O, A)^* \to A$

Dale:

Develop Computer Science perspective as well as the machine learning perspective.

Day 3:

• An average person like me that doesn’t have solid scientific understanding/background, if that person isn’t in a good shape physically, then that person wouldn’t stand a chance to remain stable mentally when faces someone like Clare Lyle. Because she’s scientifically and physically in an extremely good shape and the comparison wouldn’t give a run on their money. I must at least get in a good shape physically to not lose it at leasr in that regard.

Joelle Pineau:

The definition of value is different in Economics, Ethics, sociology, etc.

I really consider her as a theoretician but her notation on value function was really sloppy. Takeaway is that even people who do formalisms can have sloppy notations on keynote talk.

Bandit orals:

• Use empirical experiments to demonstrate your theoretical results.

Michael Littman:

You can represent a dataset by: $D = \langle s, a, r, s’\rangle^+$

Day 4:

Peter Dayan:

• A book to read: “Advice for young investigator”

He revisited the idea of liking vs wanting and blew me away. Wanting is a beneficial process, but slow. Liking is fragile and not necessarily beneficial. The analogy is tongue vs. stomach. Tongue swiftly process sweet food and most people like (instant gratification), but it’s stomach who actually process what your body wants and results in long-term satisfaction.

Rich:

The biggest limitation in the quest of AI is inefficient learning algorithms.

Rich’s Dyna with function approximation.