Caglar Gulcehre

@caglarml

Scientist MAI, Prof @ EPFL, Ex-research scientist @ DeepMind, MSR, IBM Socials: bsky.app/profile/caglar…, mastodon.social/@caglarml

Lausanne, Switzerland

caglar.ai

Joined July 2016

1,334 Following

5,564 Followers

Replies Reposts Media

Pinned
Caglar Gulcehre
@caglarml
Oct 10
Incredible! I'm really proud of this work and what @im_td and @AnjaSurina have achieved. I was actually thinking of tweeting about this paper today, calling it the most underrated paper I've worked on this year!
Kaiyue Wen
@wen_kaiyue
Oct 10
Replying to @wen_kaiyue
3/ 🏆 Best Paper The Future of Facts: Tracing the Factual Generation-Verification Gap @im_td @AnjaSurina @caglarml openreview.net/forum?id=181wn…
1
Caglar Gulcehre
@caglarml
Oct 23
I really like simple ideas that work well. This is a really good example for that. Great work and execution @jdeschena, @nadjushenka_, Adrien et al!
Nadezhda Ilieva
@nadjushenka_
Oct 22
Can a language model be based purely on self-conditioning? 🤔 Yes! Meet LAW 👋 (a.k.a. Loopholing All the Way!) 🧵1/8
2
Caglar Gulcehre
@caglarml
Oct 22
Old is new again... 👀 We introduced a NOP memory cell that lets the controller attend to "nothing" when no meaningful read/write is needed in our 2016 D-NTM paper. It basically has the same purpose and goal with the transformer attention sinks which is to absorb attention
1
Caglar Gulcehre
@caglarml
Sep 20
As AI starts making significant dents in science, I think the role of scientists will gradually change. Scientists may spend less time solving problems themselves, and more time deciding which problems are actually worth solving. 🧵 1/X
8
Caglar Gulcehre
@caglarml
Jul 6
Go and check out our paper at ICML and make sure to ask any of your questions to @XiuyingWei966!
Xiuying Wei
@XiuyingWei966
Jul 6
Excited to present our #ICML 🇰🇷 paper tomorrow morning at Hall A #2911! 🧩RAT+: an upstream dense architecture that enables more flexible and accurate downstream sparse inference. Compared to attention: 1) RAT+ enables flexible dilated pattern at inference (up to 64x KVCache
1

Caglar Gulcehre

Caglar Gulcehre’s posts