573 Followers
58 Following
82 Posts

Research Scientist Team Lead at DeepMind

Language, Embodiment, Analogy, 90s Football, 80s Music.

Well grounded. Not composed.

https://fh295.github.io/noncompositional.html

Twitter@FelixHill84
Websitehttps://fh295.github.io/
Scholarhttps://scholar.google.co.uk/citations?user=4HLUnhIAAAAJ&hl=en
Leaving Mexicohttps://www.youtube.com/watch?v=WwgUrzxQcX4

A nice write-up of my #NeurIPS paper by GT Communications staff:

"An unexpected discovery by Mark Riedl & his collaborators"

https://neurips.ml.gatech.edu/artificial-agents-use-reinforcement-learning-to-explain-actions-a-necessary-step-as-they-get-smarter-at-accomplishing-tasks/

TL;DR: We modified an RL system to make it more explainable. It made the base RL work better.
https://arxiv.org/abs/2112.08907

(Also I appreciate that they used an image that makes me look younger)

Artificial Agents Use Reinforcement Learning to Explain Actions, a Necessary Step as They Get Smarter

RT @[email protected]

i'll be at #NeurIPS2022 tuesday through the end of the week! reach out if you want to talk language, rl, multi-agent, deepmind or just hang outdoors or watch the world cup :))

🐦🔗: https://twitter.com/996roma/status/1597289869436325888

Roma Patel (@[email protected]) on Twitter

“i'll be at #NeurIPS2022 tuesday through the end of the week! reach out if you want to talk language, rl, multi-agent, deepmind or just hang outdoors or watch the world cup :))”

Twitter

Working with Dr Silke Göbel and myself on a project looking at the structure and development of concepts including number. Testing children aged 4-12.

https://jobs.york.ac.uk/vacancy/research-trainee-506584.html

This is part of a bigger collaboration "BounCeS", with André Knops (Université de Paris) and Martin Fischer (University of Potsdam). Investigating the structure of conceptual knowledge using a variety of methods to compare “spatial” and “non-spatial” explanations.

Jobs - The University of York

I just used iMovie to edit my slides post-presentation-recording. I feel like I have powered up.

Emergence is not a new thing

(any ideas how to easily copy this to Mastodon)

https://twitter.com/FelixHill84/status/1592809021941354496?s=20&t=ufj-xJftgVaf6u3lZIlKpw

Felix Hill on Twitter

“Lots of folks are talking about *emergence* in Deep Learning as if it's a new thing, that happens only in large language models at scale. It's not! It has been happening for decades and in very small networks. 🧵 🧵 🧵 🧵 🧵 🧵 🧵 🧵 🧵”

Twitter
#introduction since I changed servers, I suppose I should introduce myself. I'm a computational psycholinguist in Gothenburg, Sweden, working on several projects, but my biggest one currently is on the role of AI techniques in analyzing political dogwhistles. I also work on more traditional areas of psycholinguistics and natural language processing.
Hello Mastodon world with a job ad: We have a 3 year postdoc position on our fact-checking ERC grant AVERITEC starting in January: https://jobs.cam.ac.uk/job/38085/ or as soon as possible after that! If you know anyone who might be interested, send them my way! I will be at EMNLP for a chat! Deadline Dec. 15!

RT @[email protected]

Today is the last chance to apply for our #ELLISPhD Program! Computer vision, health, robotics, #ML theory, privacy, NLP, climate sciences & more: We offer access to a wide range of disciplines! Submit your application now and join us soon! @[email protected] https://ellis.eu/websites/european-lab-for-learning-intelligent-systems/pages/what-it-s-like-to-be-an-ellis-phd-student

🐦🔗: https://twitter.com/ELLISforEurope/status/1592468921109872642

What it's like to be an ELLIS PhD student

The ELLIS mission is to create a diverse European network that promotes research excellence and advances breakthroughs in AI, as well as a pan-European PhD program to educate the next generation of AI researchers. ELLIS also aims to boost economic growth in Europe by leveraging AI technologies.

European Lab for Learning & Intelligent Systems
Just did a cracking tweet wish I'd done it here ffs