I am in Vancouver at ICML, and tomorrow I will present our newest paper "Partially Observable Reinforcement Learning with Memory Traces". We argue that eligibility traces are more effective than sliding windows as a memory mechanism for RL in POMDPs. 🧵

GitHub - NishanthVAnand/eligibility_state: Eligibility state experiments. Eligibility state experiments. Contribute to NishanthVAnand/eligibility_state development by creating an account on GitHub.

When I first started working with Doina as a master’s student 7 years ago, this was the first idea that I tried (github.com/NishanthVAna...). But, I gave up on it too soon!

I’m happy and pleased to see you explore this path in depth with a solid paper! E-traces are the best idea to come out of RL!!

Mila Launched Canada’s First AI Computing Cluster Dedicated to Academic Research | Mila Mila, in partnership with Université Laval and Calcul Québec, officially launched TamIA, the first AI computing cluster specifically designed to support academic research in the country.

Introducing TamIA, the first AI computing cluster in Canada dedicated to academic research! A collaboration of Mila, Digital Research Alliance of Canada, CIFAR, Amii, @vectorinstitute.ai, Calcul Québec, Université Laval, @ualberta.bsky.social and University of Toronto. mila.quebec/en/news/mila...

This week, Mila researchers will present more than 90 papers at @iclr-conf.bsky.social in Singapore. Every day, we will share a schedule featuring Mila-affiliated presentations.
Day 1 👇 #ICLR2025
mila.quebec/en/news/foll...

Neat! What about two neural networks that have different architectures? Are there any metrics that you’re aware of could help compare them?

Does this mean we can evaluate the representation quality of two different neutral networks and objectively say one is better than the other?

Doina lab group picture

We had a lab social and the mandatory pic after 💜

I am hoping to contribute to this thread equally!

add me too please: @itsnva7.bsky.social