Just a journeyman jostling towards a juncture. Working on AI for science.
Joined October 2019
- Tweets 32
- Following 996
- Followers 137
- Likes 6,042
6 Photos and videos
Pinned Tweet
Jun 11
Getting to fully autonomous small-molecule design requires us to respect the constraints of synthetic chemistry.
TerraSynth has helped us compress months of waiting on synthesis and iterate extremely quickly to more potent molecules.
1
1
8
354
Jun 13
Got asked how the TerraSynth title animation was made. Short answer is coarse co-fold of BTK and designed ligands, rendered by Claude using blender w/ geometry nodes.
More details in thread.
1
5
190
Jun 13
Since it's coarse co-fold we only have the backbone structure without sidechains. I wanted to go for a "fluorescent optical fibre" rendering. Trick to getting this right (after a few iterations) was
1) transparent backbone "cartoon" tube
2) principled BSDF shader to get the self-emission glow
3) kill some competing light by bumping up surface roughness
1
2
112
Jun 13
Between frames, ligand switches instantaneously while backbone coordinates are interpolated slowly. The rest is simpler: On the right, routes from TerraSynth are superimposed. Never been fond of the default rdkit settings so swapped to cleaner ACS 1996 styling.
1
71
Jun 11
Getting to fully autonomous small-molecule design requires us to respect the constraints of synthetic chemistry.
TerraSynth has helped us compress months of waiting on synthesis and iterate extremely quickly to more potent molecules.
1
1
8
354
Miles retweeted
16 Apr 2025
What is the place of exploration in today's AI landscape and in which settings can exploration algorithms address current open challenges?
Join us to discuss this at our exciting workshop at @icmlconf 2025: EXAIT!
exait-workshop.github.io
#ICML2025
4
10
32
9,598
12 Feb 2025
Highly recommended reading. An earlier draft of PAI was my bible while studying: arxiv.org/abs/2502.05244
5
561
9 Jan 2025
Was unfairly skeptical of this one, while never actually trying it on large chemical datasets. Finally tried it. Kicks ass, and seems better calibrated than other bayesian last-layer approaches (e.g. VBLL) for large pre-trained models.
14 Apr 2022
Epistemic Neural Networks. (arXiv:2107.08924v2 [cs.LG] UPDATED) ift.tt/auYzP92
1
4
534
10 Jan 2025
The two are almost equivalent, so maybe the construction of the prior network is doing some of the lifting here.
2
372
25 Dec 2024
Look what Santa left under the tree
25 Dec 2024
Bandit Convex Optimisation ift.tt/1APhQRd
2
432
Miles retweeted
19 Nov 2024
📢 Post-Bayesian online seminar series coming!📢
To stay posted, sign up at
tinyurl.com/postBayes
We'll discuss cutting-edge methods for posteriors that no longer rely on Bayes Theorem.
(e.g., PAC-Bayes, generalised Bayes, Martingale posteriors, ...)
Pls circulate widely!
2
41
184
48,983
Miles retweeted
4 Oct 2024
Continuity is the key to continuous diffusion models; but where’s the continuity in discrete diffusion models 🔎? Drawing on tools from Lévy processes, we reformulate discrete diffusion models into stochastic integrals, and unify the theory for continuous & discrete cases.
Check out our new paper 📖 with @SC96004870, @GrantRotskoff & @lexing_ying : How Discrete and Continuous Meet: Comprehensive Analysis of Discrete Diffusion Models via a Stochastic Integral Framework. web.stanford.edu/~renyinuo/p…
8
74
11,820
Miles retweeted
26 Aug 2024
NVentures backed @Terray_Tx, has unveiled COATI-LDM—a groundbreaking #genAI model for designing small molecules with precision. This innovative example of inverse design has potential to greatly accelerate #drugdiscovery. Check out the arXiv preprint nvda.ws/3MhwqDF
2
25
87
19,703
25 Jul 2023
Some beautiful work from @NiederauerC and @KGanzinger lab that I was lucky to be involved in. Check it out!
25 Jul 2023
Dual-colour DNA-PAINT single-particle tracking to monitor membrane protein dynamics @KGanzinger @NiederauerC @_amolf @MPI_Biochem #DNAPAINT #singlemolecule
#BiotechNatureComms
nature.com/articles/s41467-0…
2
7
1,393
Miles retweeted
21 Jul 2023
Cambridge Ellis Unit Summer School on Probabilistic Machine Learning 2023
1
11
95
12,292
Miles retweeted
10 Jun 2021
The three ways to solve Fokker–Planck equations: PDE evolution, stochastic (independent) or deterministic (coupled) particle systems. en.wikipedia.org/wiki/Fokker…
7
108
558