Emile Mathieu

Our London team at @Xaira_Thera is looking for an ML/AI Research Engineer to build and scale antibody design 🦈🦙. Reach out if you enjoy working in an interdisciplinary team and want to push the boundaries of science (maybe 👀) job-boards.greenhouse.io/xairatherapeut…

I’m really excited to be attending NeurIPS and presenting our work on efficient fine-tuning of pre-trained diffusion models for SOTA conditional generation. Come chat with us on 12th Dec (Thursday) at 11am! Thread below (🧵) -

Glad to move to 🦋 with the same handle 'mathieuemile' for some chats on protein design, cycling and whatnot

Il y a 30 ans…Happy Birthday, Eurostar!

@LucaAmb @jwvdm > I am definitely worried about the rise of conspiratorial thinking. Well pretty sure this 2nd mandate is _not_ going to help with that! 🙃

@jwvdm I am definitely worried about the rise of conspiratorial thinking. I hope he will listen to the right people. In general, I did appreciate his administration actions in promoting the development of the vaccine. By the way, I don't support him. I was and am on the fence.

At the end of the day, I am happy that Trump won. Here I said it. Don't hate me :-p

@LucaAmb @jwvdm He was 'indicted' on four charges regarding the January 6 Capitol attack. it didn't go further because of his immunity. #Donald_Trump" target="_blank" rel="nofollow noopener">en.wikipedia.org/wiki/Criminal_…

@jwvdm I have heard many rumor, and as a rule, I don't trust them without external verification. People have political motives.

@charlieharris01 I cannot find @Xaira_Thera🧐 How is a 'TechBio' defined? Thanks for this useful resource! :)

🚨 New blog/open resource alert 🚨 Struggling to keep track of the exponentially growing number of new TechBio companies? So was I! So I decided to start tracking them and I eventually decided to make it an open community resource. You can access via the below article👇 (1/n)

I love @googlemaps 'Public Transport' layer for London! Can eyeball potential routes and make fun or necessary plans 👌 Yet gmaps is exclusively showing main roads at high zoom level! 'Only' 45% of brits have access to a car whilst everyone can hop on a train or a bus. @GoogleUK

A man has been charged with murder after running over Paul Varry in Paris. Paul told the driver to get out of a bike lane. They drove over him. Paul was 27, a @ParisEnSelle campaigner and our friend. No words. See the protests #JeSuisPaulVarry 🇫🇷🇬🇧💔 lemonde.fr/en/police-and-…

@wud1888 @leoklarner @njwfish would you know whether the code is public or not? :)

@leoklarner This is a super exciting paper on constrained diffusion models! I'd love to explore it further. However, I noticed that the link to the code seems to be missing, and I couldn't locate it on your GitHub either. Would it be possible for you to share the link to the code? Thank you!

Generative models for scientific data can often benefit from using domain-informed constraints to reduce their problem space. In our new #NeurIPS2023 paper, we simplify constrained (Riemannian) diffusion models to make them better, faster & easier to use. arxiv.org/abs/2307.05439

wowwwwwwww 🔥 #NobelPrize2024

@crisbodnar Perhaps we should work 1000% more to move 1000% faster then? Following this linear model assumption of course

@MathieuEmile Also I think the math is pretty obvious, your competition is also smart and if they work 40% more than you, they will move 40% faster than you. So if you want to change anything by joining a startup, you either dedicate to it or not. Otherwise it's a waste of time.

Someone told me today about this European AI startup that got to 3M ARR in their first year. Then they switched to 4 day working week and plateaued. Serious startups should work on weekends.

Always love going to Walthamstow wetlands, but last weekend was surprise to see lots of wildlife rubbish scattered everywhere by overflowing bins + wind on Lockwood way entrance 👀 @E17Wetlands @wfcouncil

@dpkingma @AnthropicAI Congrats!👏

Personal news: I'm joining @AnthropicAI! 😄 Anthropic's approach to AI development resonates significantly with my own beliefs; looking forward to contributing to Anthropic's mission of developing powerful AI systems responsibly. Can't wait to work with their talented team, including a number of great ex-colleagues from OpenAI and Google, and tackle the challenges ahead!

New edition of 'NeurIPS in Paris' meeting to discuss and discover new advances in Machine Learning ! ⌚️Registrations are open until 5th November

Any plan to make Tufnell Park junction safe? Pretty certain that the bicycles boxes are’t going to cut it (especially as drivers don’t respect them) since I rarely see vulnerable road users cycling through it. Pretty unpleasant and unsafe atm @TPC2S @CycleIslington @IslingtonBC

Recently been made aware of the campaign for a safe Finsbury Park junction, it’s due time as it’s a really dangerous and scary one 😬 Any plan for Seven Sisters rd? Keeping my hopes high 🙌 @CycleIslington @hackney_cycling @willnorman @IslingtonBC @hackneycouncil

@json_yim @alexechu_ I def concur that is it an open problem how to do this best! I really like the approach from @json_yim and @AndrewC_ML, and from the multi-marginal paper from @msalbergo, where one amortises the NN on every/many noising path at training time and thus can choose any at inference

@alexechu_ Or you can have 3 processes in multiflow ;) arxiv.org/abs/2402.04997 I agree how to handle multiple diffusions is not well understood. This is a good fundamental ML problem to work on -- not just for proteins.

in frame diffusion you have two jointly evolving _correlated_ diffusion processes (trans & rot). i always thought this problem was more fundamental than equivariance, but found the methods usually focused much more on the equivariance rather than how to handle the interaction of

@alexechu_ I guess the advantage vs modelling the joint p(trans, rot) as p(trans) p(rot | trans) (or vice versa), is that a) you're solving only one differential equation hence amortising the call to the NN, b) your noised modalities explicitly inform each others while denoising