Ed Wagstaff

[Ed Wagstaff @EdWagstaff]

🤖New blog post🤖

AlphaFold 2 refines protein structures in an iterative, equivariant way. @FabianFuchsML and I wrote about the challenges of building an iterative, equivariant SE(3)-Transformer. Complete with code and a tech report!

https://t.co/L1waoO6ho8

[Fabian Fuchs @FabianFuchsML]

I wrote a blog post trying to understand TrackStar, a gradient-based method tracing LLM predictions to influential training examples.

Mostly: take the main equation, read it from right to left, and poke at the pieces with a small MNIST toy example. 🙂

☕️ https://t.co/JCWblqh956

[Ed Wagstaff @EdWagstaff]

@BrendanCGillott I don't think "recomposed from stock" is a good description of what LLMs do

[Franziska Seeger (she/her) @FranziskaSeeger]

I will be giving an overview of what we are up to @PrescientDesign during my talk @CCG_MOE #BiologicsByDesign2023 #proteinengineering #machinelearning #proteindesign

[Petar Veličković @PetarV_93]

🕸️ In "Learnable Commutative Monoids for Graph Neural Networks", Euan Ong and I show how to build potent, learnable and aligned aggregators for GNNs, building on the success of PNA (@GabriCorso @dom_beaini) and set aggregation (@EdWagstaff @FabianFuchsML @martinengelcke).

[Petar Veličković @PetarV_93]

📢 New blog post!
Realising an intricate connection between PNA (@GabriCorso @lukecavabarrett @dom_beaini @pl219_Cambridge) & the seminal work on set representations (Wagstaff @FabianFuchsML @martinengelcke @IngmarPosner @maosbot), Fabian and I join forces to attempt to explain!

[Samuel Stanton @samuel_stanton_]

BayesOpt is great for exploring the unknown, but it's hard to know when you're accurately predicting new inputs. Two big reasons it's hard --- faulty assumptions and covariate shift. Our new preprint shows conformal prediction can improve coverage!

https://t.co/kilh2hW9Jj
1/7

[Ed Wagstaff @EdWagstaff]

I've seen plenty of people flatly dismissing these sorts of concerns as "sci-fi", but no substantive arguments that they shouldn't be taken seriously. That's partly because I don't really know where to look for such arguments. Are there any?

[Anders Sandberg @anderssandberg]

Post in question: https://t.co/F8pcKgbmip

[Ed Wagstaff @EdWagstaff]

@pfau Could you say any more about which already-known facts about Transformers make this unsurprising?

[Freddie Kalaitzis @alkalait]

It is w pride (and a tear in my eye) that after 6 months of competitive assessment, negotiations & legal due diligence, I can announce that @esa has awarded a €1M contract to a program I am leading in consortium with @CompSciOxford @isp_uv_es @BrockmannCon,

project #OpenSR

[Ed Wagstaff @EdWagstaff]

@razibkhan Tuition fees were capped at £1,000/year in 2000, tripled in 2006, tripled again in 2010.

[Rohin Shah @rohinmshah]

I'm hiring for Research Scientist and Research Engineer roles on the Alignment team at DeepMind! Come talk to me at the EA Global London career fair (4-8 pm Friday 15 April), or DM / email me and I can let you know when the job ad officially goes up.

[Fabian Fuchs @FabianFuchsML]

Graph neural networks often have to globally aggregate over all nodes. How we do this can have a significant impact on performance 🎯. After we recently finished a project on this, I wrote a blog post on this topic. Let me know what you think! :)

➡️https://t.co/OGJJAF9w9C ☕️

[Michael Bronstein @mmbronstein]

Multiple open faculty positions at @CompSciOxford - join a great CS department!

https://t.co/OuWjlNNc9j

[antonio vergari ⚔️ @tetraduzione]

I am looking for troublemakers who want to start a PhD in #probabilistic #ML #AI at @ancAtEd @InfAtEd @EdinburghUni to design the foundations for the next generations of #autonomous #systems.

Drop me an email or DM here!
Please share!

[Shuyu Lin @ShuyuLin_n]

Sign up to join discussions on some very interesting topics!

To name a couple:
- “I wish I knew … at the start of my research career”
- “Is deep learning the master algorithm?”

This event is hosted as part of the social program at #NeurIPS2021 😎😎

[Ed Wagstaff @EdWagstaff]

This two-tweet thread may be the best thing I have ever seen written on the subject of i.i.d. coin flips.

[Bartosz Milewski @BartoszMilewski]

Suppose you want to generate a random binary byte by tossing coins. You can either toss 8 coins once, or toss one coin 8 times. Why do we expect the same distribution both times?

[Ed Wagstaff @EdWagstaff]

@JDHamkins Great post! My immediate thought for the triangle case was angle trisection - it's easy to triple ∠ABC but you can't go back.

[John Schwenkler @johnschwenkler]

A truly extraordinary footnote.

[Plugin Play @PluginPlay__]

Playing around with @sportsracer48's new notebook. Taking advantage of the optical flow features and water drops. Wild!

#VQGAN #vqganclip

[Ed Wagstaff @EdWagstaff]

@arthurturrell @code Any particular plugins or faffy setup needed to do this?