Irina Rish
@irinarish
prof UdeM/Mila; Canada Excellence Research Chair; AAI Lab head https://t.co/UzlrC7ZrGF; INCITE project PI https://t.co/0rV7szd7rH; CSO https://t.co/XDhj6MEtUj
ID:393867458
https://www.irina-rish.com 19-10-2011 06:22:21
6,4K Tweets
8,6K Followers
994 Following
Glad to see this becoming a reality! Since 2013, we worked on Psych 4 AI & AI for Psych with my colleagues Guillermo Cecchi, Steve Heisig, Djallel Bouneffouf @doerlb Sahil Garg at Comp Psych(ology/iatry) dept IBM Research (and more recently, UdeM collaborators Guillaume Dumas
Using just MMLU for model comparison can be quite misleading/limiting as it does not capture many other model's properties and behaviors in real user interactions. The Y axis must change - or become (highly) multivariate manifold. Ethan Caballero is busy
How do model components (conv filters, attn heads) collectively transform examples into predictions? Is it possible to somehow dissect how *every* model component contributes to a prediction?
w/ Harshay Shah andrewilyas, we introduce a framework for tackling this question!
Some thoughts on AI and related issues, from the perspective of Diverse Intelligence:
noemamag.com/ai-could-be-a-…
thank you peter mellgard Claire Isabel Webb !
This is a really simple and effective way to reduce the bias in Transformers. We normally prune for efficiency but this paper shows that we can also prune for fairness! Great work by my student Abdelrahman Zayed and collaborators!