My teaching experiences as a Graduate Student Instructor for Cpmputational Models Cognitive Science. This is a super interesting course that talking about human learning rather than machine learning.
Probably nobody would see this post but just to remember the sweetest time we had through this semester. Many thanks to our fantastic staff group:
- Steven Piantadosi: One of the best Profs I have ever worked with in my life. Thank you for the buncing ball and thank you for giving me this chance being part of the staff group :))) Anyone pursuing a degree in Cognitive Science at Berkeley who saw this post, please take this class!!! It really touches the soul of human learning and give me a lot of enlightments in my own research in machine learning and transportation! Steve even gave a lecture about how to choose grad schools. He's so nice and willing to help!
- Sam Cheyette: Samrt PhD and has deep eyes. Nice to have you to hold OHs on so many Weds :)
- Aummul Baneen Fidvi: Graduate Student Instructor from School of IEOR and a super great friend! Wish you all the best in your career (in this nobody-knows post :p)
- Shaivya Rastogi: Organized and always smiling to everyone. Truly happy to meet you!!!
- Mugdha Bhusari: Offered me tons of help while I was just start teaching and holding office hours. Good luck for the rest of the degree!!!
- Michael Levy
Clarification: Some of the work in this repo belong to the wisdom of the staff group, not me. Thank you!
topic: Implement the Rescorla-Wagner and plot it! Nice to know :)
- Classical conditioning
- Conditioned Stimulus(CS): Something not intrinsically rewarding, such as a tone, light, or touch.
- Unconditioned Stimulus(US) : Something intrinsically rewarding/unpleasant, such as food, warmth, or a shock.
- Have CS precede US repeatedly. Result: CS causes response (e.g. salivation, fear) even before presence of US.
- Rescorla-Wagner
- Model of how associative strength changes between CS and US given observations.
- How well does the CS predict the US? Learning happens when events violate expectations.
- “Prediction error” Greater prediction error ⇒ greater learning.
- Thoughts: does this sound familiar? Dear Machine Learning?
- In machine learning, we all have a goal -> to minimize the loss between machine predictions and ground truth. Algorithms to implement this is the famous 'gradient descent'.
- This is the exact process of minimizing prediction error in animials and human beings!
- Amazing! Human do the same to learn as algorithms in machine learning process! Or reversely, smart scientists teach machine to learn just in the way of human learning.