GANs was kind of late in the game for me! I had already handed in my thesis, and by that time deep learning was basically working. I really enjoyed the intellectual atmosphere of ~2009-2012 when deep learning didn't really work and all bets were open, solving it was a mystery.
RESEARCH
-
GANs Win Test of Time Award: Reflections on 2012-2014
By
–
My GAN co-author Sherjil Ozair has written about some memories of 2012-2014 in the context of GANs winning one of this year's test of time awards, worth a read for the nostalgia if you were around back then, or for learning what it was like if you weren't
-
Contrarian approach to proving novel ML technique viability
By
–
They thought it wouldn't work, and I was contrarian enough to go home and code it up to prove that it would. But it's a good thing we didn't know about MMD (kernelized moment matching) or we would have just done that instead of inventing something new.
-
Scaling Machine Learning: Third Order Moments vs Discriminator Approach
By
–
At the party, the people on the generator / moment matching party asked me for software engineering advice: how to scale up to third order moments? I said this was doomed to fail and counterproposed the discriminator idea.
-
Deep Learning Textbook Work During Thesis Transition Period
By
–
I would add here, I was not part of the generator / moment matching project, I was in lame duck mode having handed in my thesis and was mostly writing the Deep Learning textbook from home, not doing research.
-
Indic-Parler TTS: 20 Indian Languages Speech Synthesis Model
By
–
Introducing Indic-Parler TTS – Trained on 10K hours of data, 938M params, supports 20 Indic languages, emotional synthesis, apache 2.0 licensed! 🔥
— Vaibhav (VB) Srivastav (@reach_vb) 3 décembre 2024
A collaboration w/ @ai4bharat & @huggingface – w/ fully customisable speech and voice personas!
Try it out directly below or use… pic.twitter.com/8aagHcCT61Introducing Indic-Parler TTS – Trained on 10K hours of data, 938M params, supports 20 Indic languages, emotional synthesis, apache 2.0 licensed! A collaboration w/ @ai4bharat & @huggingface – w/ fully customisable speech and voice personas! Try it out directly below or use
-
DNA Foundation Models for Medical and Veterinary Healthcare
By
–
During my sabbatical, one project I really enjoyed was training DNA foundation models with the team at @arcinstitute
. I've long believed that deep learning should unlock unprecedented improvements in medicine and healthcare — not just for humans but for animals too. I think this -
LLM numeric reasoning and inference behavior
By
–
Can’t speak to Claude but in general models are just trained to do this — doesn’t need multi-stage inference etc. You see it with a lot of LLMs if you ask if 9.11 or 9.9 is greater. Common to get variations of “9.11, because […] so actually it’s 9.9.”