UCL CSML

UCL CSML @ELLISforEurope Unit regrouping @GatsbyUCL, @uclcs and @stats_UCL. We organise the @JumpTrading CSML Seminar Series.

[Discrete Diffusion Reading Group @diffusion_llms]

This Monday, Oct 19, Justin Deschenaux (@jdeschena) will present his paper:

Simplex Diffusion Models

[UCL CSML @uclcsml]

A reminder of the online talk today! Note a change of speaker, @jdeschena will be presenting instead on the same topic. Hope to see lots of you there!

[UCL CSML @uclcsml]

The UCL CSML Seminar is back! For our first talk, Valentin de Bortoli (GDM) @ValentinDeBort1 will present simplex diffusion models! The talk will be 12:30 (BST) Wed 14 of Oct at @SWC_Neuro @GatsbyUCL , and streamed at https://t.co/OjnGIlUchQ.

[UCL CSML @uclcsml]

@agalashov @ArthurGretton @ArnaudDoucet1 @liwenliang @JamesTThorn @AndrewC_ML @jdeschena The talk will be based on the following work.

[Valentin De Bortoli @ValentinDeBort1]

🔥Welcome 🔺Simplex Diffusion Models🔻!

Standard discrete diffusion models suffer from "information collapse" because they discard uncertainty at intermediate steps by sampling categorical tokens. (1/5)

[UCL CSML @uclcsml]

The UCL CSML Seminar is back! For our first talk, Valentin de Bortoli (GDM) @ValentinDeBort1 will present simplex diffusion models! The talk will be 12:30 (BST) Wed 14 of Oct at @SWC_Neuro @GatsbyUCL , and streamed at https://t.co/OjnGIlUchQ.

[UCL CSML @uclcsml]

Tagging some relevant people @agalashov @ArthurGretton @ArnaudDoucet1 @liwenliang @JamesTThorn @AndrewC_ML @jdeschena

[Valentin De Bortoli @ValentinDeBort1]

🔥Welcome 🔺Simplex Diffusion Models🔻!

Standard discrete diffusion models suffer from "information collapse" because they discard uncertainty at intermediate steps by sampling categorical tokens. (1/5)

[Gatsby Computational Neuroscience Unit @GatsbyUCL]

Interested in a #PhD in machine learning & theoretical neuroscience?

Applications for 2027 entry to the Gatsby Unit PhD programme now open!

💰 Fully-funded 4-year programme
🤝 Close links to Sainsbury Wellcome Centre & the ELLIS Unit @ UCL (@uclcsml)
ℹ️ https://t.co/Im1iRIbQx2

[Arthur Gretton @ArthurGretton]

Applying to #UCL for a PhD in AI? Then be sure to apply for ELLIS as well! A great opportunity to collaborate with leading European AI labs.

[ELLIS @ELLISforEurope]

🎓 Ready for a PhD in AI/ML? The #ELLISPhD Program connects top students with leading researchers across Europe.

The application portal opens on 1 Oct.

Register now to join the virtual info session on 30 Sep.

All info at https://t.co/dea4vtrv9E

[Antonin Schrab @AntoninSchrab]

New Aggregation Methods!
-SingleBatch SB
-TwoBatch TB
-DataDriven SB
-Sequential SB

Applications
-Multiple Hypothesis Testing
-Conformal Prediction

MMDAgg-SB HSICAgg-SB
-Exact level
-Higher power
-Admit p-value
-Minimax optimal

Schrab* Shah Gretton Kim*
https://t.co/MpsEUKBJEP

[Arthur Gretton @ArthurGretton]

Optimized MMD for Detecting Distribution Shift!

Video:
https://t.co/bTJ0ZkU0zU

and slides
https://t.co/thNvTmjUz9

from the #ICLR26 workshop
https://t.co/WpUCAtJQPO

Covers:
https://t.co/x6syFYarxL
and
https://t.co/OBhBLVwNHc

[Google Gemma @googlegemma]

Meet DiffusionGemma!

An experimental open model that explores a fast approach to text generation, released under an Apache 2.0 license.

Moving beyond sequential, token-by-token processes to generate entire blocks of text simultaneously. Here’s what’s new with DiffusionGemma: 👇

[Advances on Adaptive Experimentation Workshop @aaeworkshop]

Final deadline reminder: applications to AAE close today.

Join us at Gatsby for two days on adaptive experimentation, causal inference, bandits, RL, and sequential decision-making.

Apply: https://t.co/Zar0qWhgA8

[Krishna Balasubramanian @krizna_b]

🚨 🧵 New paper: Finite-particle rates for drifting models.

Drifting trains 1-step generators by moving model μ toward data ν with a conservative or non-conservative velocity field. Consider data on x-axis, model on the y-axis. These 2 fields yield 2 different motions. 1/7

[Arthur Gretton @ArthurGretton]

Your drifting model is secretly a fixed point for the Wasserstein gradient flow on...
...the KL?
...an approximation to the Sinkhorn?
...Is it even a Wasserstein gradient flow at all?

https://t.co/QJLh86Hi0d

@liwenliang @agalashov @JamesTThorn @ValentinDeBort1 @ArnaudDoucet1

[UCL CSML @uclcsml]

Next Wednesday (May 13) at 12:30 pm we’re excited to have Long Tran-Thanh (Warwick) present his work on “Pruning at initialization through the lens of graphon limits”. The seminar will be streamed live at https://t.co/WIZ32HhI2G.

[ELLIS @ELLISforEurope]

Come spend 2 days in adaptive experimentation, causal inference, & online learning at @ucl this June.

Co-organised by @uclcsml, it brings together researchers across theory, methodology, & applications.

📍 London 🇬🇧
📅 18-19 June
⏰ Apply by 28 May
🔗 https://t.co/bzXvXc2m1P

[UCL CSML @uclcsml]

This Wednesday (May 6th) 12:30 pm, we are excited to have Ben Dai (CUHK) present his work on “RankSEG: A consistent ranking-based framework for segmentation”. The seminar will be streamed at https://t.co/OjnGIlTEsi. Please note the event is online only.

[Advances on Adaptive Experimentation Workshop @aaeworkshop]

🚀 Excited to announce the Advances in Adaptive Experimentation (AAE) Workshop!

📅 June 18–19
🔗 https://t.co/Zar0qWhgA8

Join us at @GatsbyUCL in London for two days at the frontier of causal inference 🧩, adaptive experiments 🔁, and bandits/RL 🤖!

[Arthur Gretton @ArthurGretton]

Looking forward to this!