Finbarr Timbers

Hi, I'm Finbarr Timbers. I'm a machine learning researcher currently working to solve intelligence. This is my personal site and has no relationship to any organization I may be professionally affiliated with (disclaimer).

I am currently working on reinforcement learning research, mostly in the context of game theory. Lately, I've been focusing on fundamental RL research in options with planning.

Occasionally, I read things that aren't just machine learning papers, and list the books I particularly like here.

You can contact me by email at finbarrtimbers [at] Google's email service.


Setting prices for your business 21 March 2021
The junior tech landscape 7 April 2018
Pointer Networks 20 September 2017
Do deep networks generalise or just memorise? 4 July 2017
Outrageously Large Neural Networks: The sparsely-gated Mixture-of-Experts layer 1 July 2017
Accurate, Large Minibatch SGD: Training ImageNet in 1 Hour 20 June 2017
Tests make you write down your assumptions 5 March 2017
Random Search for Hyper-Parameter Optimization 1 March 2017
How I designed my machine learning app 21 February 2017
Securing yourself against online tracking 4 February 2017
Useful Bash One-liners 20 January 2017
Including web fonts in RMarkdown 19 January 2017
A Deep Hierarchical Approach to Lifelong Learning in Minecraft 3 January 2017
Larry Ellison on consulting costs 6 December 2016
SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5mb model size 10 November 2016
Conditional image synthesis with auxiliary classifier GANs 8 November 2016
Generative Adversarial Imitation Learning 8 November 2016
Minimal example of how to do model selection in Python 26 October 2016
Representation Learning: A Review and New Perspectives 20 October 2016
Full Resolution Image Compression with Recurrent Neural Networks 19 October 2016
Generative Adversarial Networks and Actor-Critic methods 19 October 2016
Using simulated data to train robots 18 October 2016
Safe and Efficient Off-Policy Reinforcement Learning 18 October 2016
XGBoost: A scalable tree boosting system 20 September 2016
Excellent description of how hashtables work 15 August 2015
Full example for using JSONcpp on Unix 6 September 2014
ARIMA, ARMA, what's the difference? 21 April 2014
Solving Partial Autocorrelation Functions 3 March 2014


I turned my academic research into a SaaS app that automatically analysed & deduplicate bug reports. Unfortunately, it was too expensive for me to keep running, but I have the code & can deploy it with a few hours work.

If anyone's interested in using this at their company, I'd love to talk to you.

Automated textual analysis for bug report deduplication

A system that automatically detects duplicate bug reports.


A talker on Docker: How containers can make your work more reproducible, accessible, and ready for production

Edmonton Data Science Meetup. Edmonton, AB, October 5th, 2016.

Slides: PDF.

Automating your work away: One consulting firm's experience using KnitR

UseR! 2016, Stanford University, Stanford, CA, Tuesday, June 28th, 2016.

Slides: HTML, PDF, Video.

Work Experience

Research Developer, DeepMind

My role is to provide engineering support and contribute ideas to research projects. Heavy experience with Tensorflow and distributed computing, done mostly in Python. I was the first engineer hired in the Edmonton office.

Data Scientist, Darkhorse Analytics

Used my background in academic research to come up with highly scalable algorithms based on machine learning to solve our clients needs. Solutions were typically based on the state of the art academic literature.