Recent & Upcoming Talks

2024

Open Problems in Gaussian Process Approximation and Benchmarking

All good GP approximations should perform very similarly, since they approximate the same thing. This means that significant differences in performance are down to tuning of approximation parameters, and we should adjust benchmarking procedures accordingly.

Dec 14, 2024 12:00 PM — Jul 3, 2024 1:00 PM Vancouver, Canada

Mark van der Wilk

Bivariate Causal Discovery using Bayesian Model Selection

Can Bayesian model selection help to determine causal direction? (Yes.)

Jul 3, 2024 12:00 PM — 1:00 PM Venice, IT

Mark van der Wilk

Variational Transductive Learning?

Mar 28, 2024 12:00 PM — 1:00 PM Oxford, UK

Mark van der Wilk

2023

Meaningful Metrics for Probabilistic Predictions

How should you evaluate probabilistic predictions?

Jul 18, 2023 9:00 AM — 10:30 AM Computer Lab, Cambridge, UK

Mark van der Wilk

Causal Discovery through Bayesian Model Selection

Does Bayesian inference have anything to say about inferring causal direction? (yes, it does)

Jun 2, 2023 12:00 PM — Jun 2, 2022 1:00 PM UCL, London, UK

Mark van der Wilk

2022

OccaMLab at NeurIPS: GPs, Equivariance, Causality

Overview of our papers at NeurIPS.

Dec 8, 2022 12:20 PM — 12:30 PM Computer Lab, Cambridge, UK

Mark van der Wilk

Learning Inductive Biases in Nerual Networks

Learning invariances in neural networks to automate neural architecture design.

Jul 11, 2022 11:30 AM — 12:30 PM Computer Lab, Cambridge, UK

Mark van der Wilk

A Smarter Neuron: Automating Neural Architecture Design

Learning invariances in neural networks to automate neural architecture design.

Mar 25, 2022 4:00 PM — 5:30 PM Fully Connected, Cambridge, UK

Mark van der Wilk

Inductive Biases, Input Densities, and Predictive Uncertainty

Is predictive uncertainty in supervised learning as simple as realising when you are in a low-data region?

Mar 9, 2022 4:00 PM — 5:30 PM University of Amsterdam, Netherlands

Mark van der Wilk

Automating Gaussian Process Approximations

How mathematical guarantees lead to easier-to-use and more accurate methods.

Mar 9, 2022 4:00 PM — 5:30 PM Amazon, Berlin, Germany

Mark van der Wilk

Approximations, Inductive Biases, and their Connections in Gaussian Processes

How is the efficiency of a GP approximation related to inductive biases like invariances?

Jan 14, 2022 9:00 AM — 10:15 AM Online

Mark van der Wilk

2021

Input Densities and Predictive Uncertainties

Is predictive uncertainty in supervised learning as simple as realising when you are in a low-data region?

Oct 12, 2021 12:00 PM — 1:00 PM University of Copenhagen, Denmark

Mark van der Wilk

Data Augmentation in Infinitely Wide Neural Networks

How Gaussian process approximations can help us implement data augmentation in infinitely wide neural networks.

Sep 21, 2021 7:00 PM Google Brain (Online)

Mark van der Wilk

Reliable Training of Gaussian Process Approximations

How to select inducing inputs, and how this makes training sparse GPs much better.

Jun 2, 2021 4:00 PM G-Research, London

Mark van der Wilk

How Accurate Gaussian Processes can help Deep Learning

How Accurate Gaussian Processes can help Deep Learning.

Apr 30, 2021 4:00 PM — 6:00 PM Cambridge Computational & Biological Learning group (Online)

Mark van der Wilk

Bayesian Model Selection in Deep Learning

Various work in getting Bayesian model selection to work in deep learning.

Feb 24, 2021 4:00 PM Qualcomm Amsterdam (Online)

Mark van der Wilk

2020

Bayesian Model Selection in Deep Learning

Various work in getting Bayesian model selection to work in deep learning.

Dec 10, 2020 12:00 AM NeurIPS Bayesian Deep Learning meetup (Online)

Mark van der Wilk

Variational Gaussian Processes without Matrix Inverses

Jun 16, 2020 University of Oxford, UK

Mark van der Wilk

Learning Invariances using the Marginal Likelihood

A tutorial on Bayesian model selection, and how it can be used to learn invariances using backprop.

Jun 8, 2020 12:00 AM Deep Learning Classics & Trends Reading Group (Online)

Mark van der Wilk

Learning Invariances using the Marginal Likelihood

Kernels provide a powerful way of encoding assumptions about the class of functions that should be used for a particular learning …

Feb 17, 2020 EURECOM, Sophia Antipolis, France

Mark van der Wilk

2019

Variational Gaussian Process Models without Matrix Inverses

Can we train Gaussian processes without taking an inverse at every iteration?

Dec 8, 2019 2nd Symposium on Advances in Approximate Bayesian Inference, Vancouver, Canada

Mark van der Wilk

Sampling for Data Augmentation: Generative Models vs Invariances

Is good unsupervised learning enough for finding data augmentations that help generalisation?

Oct 10, 2019 1:00 PM Workshop on Generative Models and Uncertainty, Copenhagen

Mark van der Wilk

Invariances in Gaussian processes and How to Learn Them

When learning mappings from data, knowledge about what modifications to the input leave the output unchanged can strongly improve …

Sep 12, 2019 1:00 PM Gaussian Process Summer School, University of Sheffield, UK

Mark van der Wilk, ST John

2018

Convolutional Gaussian Processes: a Building Block for Bayesian Deep Learning

Jul 8, 2018 Numerical Analysis Group Seminar, University of Manchester, UK

Mark van der Wilk

2017

Convolutional Gaussian Processes

When learning mappings from data, knowledge about what modifications to the input leave the output unchanged can strongly improve …

Sep 28, 2017 ASML, Eindhoven, the Netherlands

Mark van der Wilk