What do neural network functions look like as they learn? GINN: Geometric Illustrations for Neural Networks. bayeswatch.com/2018/09/17/GIN… and arxiv.org/abs/1810.01860 #bayeswatch

CINIC-10: an extended plug in for CIFAR-10 Report: arxiv.org/abs/1810.03505 Blog: bayeswatch.com/2018/10/09/CIN… Data: datashare.is.ed.ac.uk/handle/10283/3… Github: github.com/BayesWatch/cin…

It is the early part of Neural Network training that matters: Fisher pruning appears ineffective without retraining.

Machine Learning faculty posts Informatics, Edinburgh, UK. Substantial expansion in ML&AI, New Bayes Centre for Data Science. Speak to us (Amos Storkey, Chris Williams) at NeurIPS. bit.ly/2DVxRGU #NIPS #NeurIPS #Edinburgh #DeepLearning #mljobs @InfAtEd

My latest blog post on meta-learning in general and "How to train your MAML" in particular is now out. bayeswatch.com/2018/11/30/HTY… The post thoroughly explains MAML, some of its problems, and proposes some solutions. In addition visualizes the learned per-step per layer learning rate

Edinburgh postdoc position available. Novel methods for Deep Learning in Brain Imaging. bit.ly/2WhdSIn . Joint between Informatics bayeswatch.com Dundee bit.ly/2WfvwfO and Centre for Clinical Brain Sciences bit.ly/2sS3Lg1 . #Bayeswatch infed org

Scott Rudin kills off performances of To Kill a Mockingbird across the world with threats of law suits. Acceptable? Not to me. capitaltheatres.com/whats-on/mocki… #boycottRudinplays

First paper, first tweet! Here we consider an increasingly common problem: model compression when training data is not available (e.g. private or too large). Our solution is to adversarially search for pseudo data, and use it for distillation. Paper & code arxiv.org/abs/1905.09768

Humans can learn new notions quickly from a small number of labelled examples. However, they can also adapt those notions in light of new unlabelled data, to improve their generalization. Inspired by this, we develop 'Self-Critique and Adapt'. arxiv.org/abs/1905.10295

Our work on Gaussian Processes and Deep Neural Networks for few-shot learning.

Lecturer/Reader (equiv. Assist./Assoc. Prof.) in Machine Learning. School of Informatics, University of Edinburgh. Deadline: 5pm 22 Jan 2020. vacancies.ed.ac.uk/pls/corehrrecr…

Lecturer/Reader (equiv. Assist./Assoc. Prof.) in Machine Learning. School of Informatics, University of Edinburgh. Deadline: 5pm 22 Jan 2020. #mljobs #AIjobs #facultyjobs #academicjobs #jobsacuk #DeepLearning vacancies.ed.ac.uk/pls/corehrrecr…

Meta-learning and AutoML folks. Are you tired of being limited to a few inner steps when learning hyperparameters with gradients? Our latest paper enables gradient-based HPO over CIFAR-10 sized problems, without introducing greediness. arxiv.org/abs/2007.07869

I am advertising a PhD studentship in Better Sample Efficiency in Deep Reinforcement Learning, joint with Kamil Ciosek and Microsoft Research Cambridge. ed.ac.uk/informatics/po…

One or two deep learning postdocs in Edinburgh with Amos Storkey: continual few-shot learning and beyond - vacancies.ed.ac.uk/pls/corehrrecr… #AI #MachineLearning #MLjobs #DataScience #AIjobs

Deep Learning postdoc, Informatics, Edinburgh. Closing soon (Mar 31). vacancies.ed.ac.uk/pls/corehrrecr…

State of the art one shot learning via a Data Augmentation GAN: DAGAN arxiv.org/abs/1711.04340…

Interested in PhD place in Data Science, Machine Learning, AI? x.com/EdiDataScience…

Postdoc -- Informatics, Edinburgh: Deep Learning Methods for Constrained Environments. Start Date Flexible. Deadline to be extended 1 week to 25 May. bit.ly/2IJWOa2