PhD Student
University of Edinburgh

Hey, I'm Ben.

Research

Here's some information about the work I do, that I find particularly interesting.
Send me a message if we have overlapping interests.
🎯 Zero-Shot Generalisation
Show a monkey how to stack some bricks, and it’ll ace the job no matter the shape, color, or pose. Teach the same trick to a DL model, and watch it have an aneurism when the colour of the brick changes. Personally, I'm not a fan of drowning models in heaps of training data, so how do we craft domain-invariant policies?

Zero-shot Generalisation in DRL Survey - really nice survey on this topic by the lot in DARK labs, UCL.
Open-Ended Learning - Even if you don't want to read this, at least get lost in its figures 🤤.
Domain Generalisation Survey - a broader look at the problem of domain invariance.
Generalisation in Cooperative MARL - an interesting look into combinatorial generalisation.
🧠 Representation Learning
When you learn something new, what does it even mean to understand it? It's something we do quasi-proactively; half the time, it's like a mental magic trick we perform almost on autopilot. This notion of learning in DL is usually achieved with representation learning, where a model learns representations from raw data. This currently works pretty dang well in single-domain supervised learning tasks, but how can we create representations that are more representative of the mess of a world we live in?

I-JEPA - can you even talk about representation learning without mentioning Yann's work?
V-JEPA - V0.2 of Yann's vision.
Representation Learning Review - old but gold.
Multimodal Representation Learning - what it says on the tin.
🔄 Continual Learning
Is there a myth of the objective? Does it even exist, or are we just chasing shadows? This leads onto the real question: how do we design an algorithm that will last for the next thousand years?

Continual Learning Survey
Another Continual Learning Survey - Both of these surveys are great
A Definition of Continual Learning - a definition of continual DRL by one of my PhD supervisors.
📜 Magnus Finis 
Stop doing research for the sake of research. What was the end goal again? How do we get there? Well, here are a few interesting takes.

Can Machines Think? - the man himself.
The Alberta Plan - Richard Sutton's take.
Path to AMI - Yann LeCun's take.
The Big World Hypothesis - short but sweet.

experience