The OxCSML group
@OxfordStats
is participating at
@NeurIPSConf
this
week, with 25 papers in the main program, 3 workshop papers and 1 paper
in dataset track! Do stop by our posters and presentations! Check out or the following :thread:for further details
We are inviting applications for fully-funded PhD studentships in statistics and machine learning, joint with
@OxfordStats
@imperialcollege
and many industrial partners.
Happy to announce that members of the
@oxcsml
group
@OxfordStats
will be presenting 19 papers at NeurIPS this year, many in collaboration with a wide variety of other research groups. Good work everyone! Links to arxiv preprints here:
We welcome friends, old and new, to Oxford attending the Bayesian Nonparametrics conference! It's an exciting line-up, including keynotes by Aad van der Vaart, Tamara Broderick, and Long Nguyen!
Hello world! As our first tweet we're happy to announce OxCSML group's participation at
#NeurIPS2018
: 13 main conf papers co-authored, and workshop invited talks, orgs, and papers.
The OxCSML group
@OxfordStats
is participating at
@icmlconf
this week, with 13 papers in the main conference and a Test of Time Award
@yeewhye
!
Do stop by our posters and presentations!
Check out the following 🧵 for further details ⬇️
Congratulations to
@jovana_mitr
for passing her viva titled ‘representation learning with kernel methods, applications in likelihood-free inference, causal discovery and neural networks’.
@sejDino
@yeewhye
Join us for our OxCSML seminar on May 7th, at 3:30 pm (UK time). Aki Vehtari
@avehtari
(Aalto University) will be speaking about Practical pre-asymptotic diagnostic of Monte Carlo estimates in Bayesian inference and machine learning.
We have just collated recent ML conference publications from our group here : 2
@iclr_conf
, 2 UAI, 5
@Aistats2020
and 8
@icmlconf
papers! Congrats everyone!
Join us for the next OxCSML seminar today by
@victorveitch
on Causal Estimation with Machine Learning without (Simple) Unconfoundedness! UK time 330pm.
Join us this Friday Feb 26th at 3.30pm GMT for our OxCSML seminar by
Karolina Dziugaite
(Element AI)
who will talk about Distribution-dependent generalization bounds for noisy, iterative learning algorithms
Zoom details available following registration
Join us this Friday at 3.30pm BST for our next OxCSML seminar!
We're very pleased to welcome
@JulyanArbel
, who will be speaking on 'Approximate Bayesian computation with surrogate posteriors'.
Zoom details available upon registration:
Join us this Friday 26th March at 3:30pm GMT for our next seminar!
Benjamin Guedj
@bguedj
will be talking about ``A primer on PAC-Bayesian learning *followed by* News from the PAC-Bayes frontline''
Zoom details available following registration:
Join us this Friday at 3.30pm GMT for our next OxCSML seminar by
Veronika Rockova (UChicago)
who will be talking about 'METROPOLIS-HASTINGS VIA CLASSIFICATION'!
Zoom details available following registration at
📢JOB VACANCY: Postdoctoral Research Assistant for the ERC project General Theory for Big Bayes, reporting to Professor Judith Rousseau.
📆Deadline for applications noon on 12 March.
An analysis of
#COVID19
incidence and mortality in Santiago, Chile, highlights major consequences of healthcare disparities in a highly segregated city, showing that these inequalities disproportionately affected younger people
@OxfordStats
@HarvardChanSPH
Join us this Friday Feb 19th at 3.30pm GMT for our next OxCSML seminar by
Francois-Xavier Briol
@fx_briol
(UCL)
who will be talking about 'Kernel-based robust inference for intractable likelihood models'
Zoom details available following registration at
Join us tomorrow at 3.30pm GMT for our next OxCSML seminar by
@sineadwilliamso
who will be talking about 'Bayesian nonparametric models for interaction networks'!
Zoom details available following registration at .
Delighted that our paper Towards A Unified Analysis of Random Fourier Features received an honorable mention at
#ICML2019
. Great work by Zhu Li,
@jeanfrancois287
and
@DinoOglic
. Check out
@DinoOglic
's talk at 2pm today
I'm excited to finally release source code for SQAIR (which was accepted as a spotlight at NIPS this year):
Thanks
@hyunjik11
,
@IngmarPosner
and
@yeewhye
for help with this!
Join us this Friday for our next OxCSML seminar!
@RogerGrosse
will be talking about ``Self-tuning networks: Amortizing the hypergradient computation for hyperparameter optimization''!
Zoom details available following registration:
Delight to announce that our
#icml2019
paper Amortized Monte Carlo Integration (see below) has been awarded an honourable mention for best paper! Don't miss
@adam_golinski
's talk at 11.40am in room 101.
Deep Adaptive Design (DAD) enables fast, adaptive experimentation. Learning a design policy net, DAD removes need for costly computations at each step of the experiment and makes decisions in <1sec using a single forward pass.
@AdamEFoster
@desirivanova
Ilyas Malik
@tom_rainforth
Join us this Friday 12th March at 3:30pm GMT for our next seminar!
@MuratAErdogdu
will be talking about `` Convergence of Online SGD under Infinite Noise Variance, and Non-convexity''
Zoom details available following registration:
This friday 3:30 pm (London time) we have the great Caroline Uhler (MIT) in our OxCSML seminar, who will be speaking about Causality and Autoencoders in the Light of Drug Repurposing for COVID-19. You are all invited!
Equivariant Learning of Stochastic Fields: Gaussian Processes and Steerable Conditional Neural Processes
We study vector valued stochastic processes with Euclidean symmetris, and apply the results to Guassian and Neural processes.
@PHolderrieth
,
@MHutchinson141
@yeewhye
First Exit Time Analysis of Stochastic Gradient Descent Under Heavy-Tailed Gradient Noise (poster Wed Dec 11th) by Thanh Huy Nguyen,
@umutsimsekli
, Mert Gurbuzbalaban and Gaël Richard at
@ParisTech_News
and Rutgers.
Join us this Friday May 14th for our OxCSML seminar. This time we have Cynthia Rudin (Duke) with "Do Simpler Models Exist and How Can We Find Them?"
@CynthiaRudin
VIREL: A Variational Inference Framework for Reinforcement Learning (Spotlight Wed Dec 11th 10:25 -- 10:30 AM @ West Exhibition Hall B) by Matthew Fellows, Anuj Mahajan,
@timrudner
,
@shimon8282
with
@whi_rl
@CompSciOxford
.
We are joining forces in
#SupportingEllis
: The European Laboratory for Learning and Intelligent Systems (ELLIS) announced the formation of its professional association at
#NeurIPS
:
@bschoelkopf
We have prominent support from the world´s top
#AI
companies!
Also, don’t miss out on the invited talk A Bayesian Perspective on
Neural Processes by Yee Whye Teh
@yeewhye
at Bayesian Deep Learning
Workshop on Dec 14!
Asymmetric Heavy Tails and Implicit Bias in Gaussian Noise Injections
Injecting Gaussian noise = bias in SGD because of heavy-tailed asymmetric noise on gradients.
@CamutoDante
· Xiaoyu Wang · Lingjiong Zhu ·
@cholmesuk
· Mert Gurbuzbalaban ·
@umutsimsekli
StatML CDT Admissions 21/22
The deadline for the 1st round - FRIDAY 13 NOVEMBER is approaching fast (only unlucky for some!!). Go to our website for information on how to apply. We look forward to receiving your application.
Join us this Friday for our next OxCSML seminar!
We're very pleased to welcome
@JamesJohndrow
who will be talking about ``Spectral gaps and error estimates for infinite-dimensional Metropolis-Hastings with non-Gaussian priors''
Register for zoom link:
Join us this Friday 5th March at 11am GMT for our next seminar! Linda Tan (National University of Singapore) will be speaking about the 'Use of model reparametrization to improve variational Bayes'.
Zoom details available following registration:
Wonderful surprise: my work done at
@OxfordStats
on Posterior Bootstrap received the
#isba2021
Best Student/Postdoc Contributed Paper Award. Thank you
@ISBA_events
and everyone who was supporting me with this project!
#WomenInSTEM
Generalized Sliced Wasserstein Distances (poster Tue Dec 10th) by Soheil Kolouri, Kimia Nadjahi,
@umutsimsekli
, Roland Badeau, Gustavo Rohde at
@ParisTech_News
HRL Laboratories LLC and U Virginia
🏃♀️🧪Active Testing: Sample-Efficient Model Evaluation
Active learning is label-efficient only in training. Active testing introduces strategies for sample-efficient evaluation.
🗣 ICML Poster Session 5
✍️
@janundnik
@sebfar
@yaringal
@tom_rainforth
📄
Join us this Friday (June 25th) for our next OxCSML seminar!
We're very pleased to welcome
Quan Zhou (Texas A&M University)
who will be talking about ``Complexity of local MCMC methods for high-dimensional model selection''
Register for zoom link:
We also have three workshop papers:
🤖 PCA Subspaces Are Not Always Optimal for Bayesian Learning 🤖
📄Paper:
🎙️Talk:
Alexandre Bense
@amirjoudaki
@timrudner
@vincefort
Oops. 14 papers!
On Signal-to-Noise Ratio Issues in Variational Inference for Deep Gaussian Process
We show that importance-weighted VI for deep GPs can lead to arbitrarily poor gradient estimates and how to prevent this from happening
@timrudner
@yaringal
@tom_rainforth
BayesIMP: Uncertainty Quantification for Causal Data Fusion
We propose a Bayesian Conditional Mean Embedding to embed the interventional distribution in the RKHS to estimate the average treatment effect with uncertainty quantification under the causal data fusion setting.
Variational Bayesian Optimal Experimental Design (spotlight Tue Dec 10th 04:40 -- 04:45 PM @ West Ballroom C) by Adam Foster, Martin Jankowiak, Elias Bingham, Paul Horsfall,
@yeewhye
@tom_rainforth
, Noah Goodman. Work done by Adam during his internship at
@UberAILabs
.
Deconditional Downscaling with Gaussian processes
We frame downscaling of indirectly matched spatial fields as an inverse conditional expectation problem and propose novel probabilistic solution based on Gaussian processes and kernel mean embeddings
@Chau9991
@shbouabid
@sejDino
I’ve been very fortunate throughout my career to work with many fantastic scientists and wonderful people, but even amongst that the team
@Genomicsplc
are amazing. I love every minute of it! Thanks!!
Asymptotic Guarantees for Learning Generative Models with the Sliced-Wasserstein Distance (poster Thu Dec 12th) by Kimia Nadjahi, Alain Durmus,
@umutsimsekli
and Roland Badeau at
@ParisTech_News
and
@ENS_ParisSaclay
.
📢Three exciting jobs! Florence Nightingale Bicentennial Fellow in Statistics, Probability or a related subject (x2) and Glasstone Research Fellowship in Science, both for researchers at an early stage of their career. More details:
Provably Strict Generalisation Benefit for Equivariant Models
The first strictly non-zero improvement in generalisation for equivariant models.
@BrynElesedy
@ShehZaidi
Poster session 1, Spot C6!
link:
Diffusion Schrödinger Bridge with Applications to Score-Based Generative Modeling (spotlight)
A principled generalization of score/diffusion based generative modeling using iterative reverse diffusions, with connections to optimal transport.
Uncertainty Quantification in End-to-End Implicit Neural Representations for Medical Imaging
Francisca Vasconcelos,
@bobby_he
,
@yeewhye
Oral at MedNeurIPS:
Poster at Bayesian Deep Learning Workshop:
Paper:
Oxford Statistician & Mathematician Alison Etheridge has been appointed Chair of the Council for the Mathematical Sciences, the organisation that represents the wide breadth of the mathematical sciences in the UK.
@OxUniMaths
With folks
@oxcsml
and Royal Society DELVE Initiative, we've been working on a model and website for tracking Covid-19 in British local authorities. Check out
With lockdown lifted and Christmas in sight, we hope this will be useful for everyone in GB!
On Locality of Local Explanation Models:
We investigate why popular model-agnostic local explanation models like SHAP are not locally faithful and how their locality can be increased with neighbourhood sampling. With
@SGhalebikesabi
,
@TerLucile
,
@karlado
,
@cholmesuk
Probabilistic Programs with Stochastic Conditioning
@dtolpin
,
@yuaanzhou
,
@Tom
,
@hyang144
We formalize and show how to condition programs on variables taking a particular distribution, rather than a fixed value
#ICML2021
Following the Italian successes at the Olympics (I might take this a bit too further, I know), I'll go for a self-congratulatory tweet. I won one of the Best Student/Postdoc Contributed Paper Award at
@ISBA_events
! Thanks, this was unexpected (I actually thought it was a scam)!
Augmented World Models Facilitate Zero-Shot Dynamics Generalization From a Single Offline Environment
A simple self-supervised context adaptation algorithm which allows generalisation to novel environments from offline data on a single environment.
Don Rubin giving a distinguished seminar today
@OxfordStats
on essential concepts for causal inference in randomised experiments and observational studies: a remarkable past and an intriguing future.
Paper info:
This preliminary work presents the first large-scale study of implicit neural representations with uncertainty quantification. Well-calibrated, accurate image reconstruction algorithms can improve doctor diagnoses and reduce patient radiation exposure.
NEO: Non Equilibrium Sampling on the Orbit of a Deterministic Transform
Paper:
Poster session: Thu 9 Dec 9:30 a.m. CET — 11 a.m. CET
Achille Thin,
@ArnaudDoucet1
Excited to announce that I will be joining the department of Mathematics at
@NUSingapore
as Peng Tsu Ann Assistant Professor in August. I would like to thank my PhD supervisors
@ArnaudDoucet1
and
@RousseauJudith
and all my amazing collaborators!