I no longer use Twitter. Please follow me on Bluesky (https://xitter.cf/t.co/w038WYESoz) and mastodon (@jess@neuromatch.social)
Oxford, United Kingdom
Joined August 2011
- Tweets1.9K
- Following1.1K
- Followers1.8K
- Likes4.9K
Pinned Tweet
Pleased to share my latest research "Zero-shot counting with a dual-stream neural network model" about a glimpsing neural network model that learns visual structure (here, number) in a way that generalises to new visual contents.
sufficiently excited about this preprint to arise from twitter slumber: arxiv.org/abs/2405.09953
looks like it's about numerical cognition, but it's actually about abstraction - how to build a neural network that actually learns structure. @tsonj @hannahsheahan. enjoy!
Jess Thompson retweeted
When does new learning interfere with existing knowledge in people and ANNs? Great to see this out in @NatureHumBehav! with @summerfieldlab @tsonj @ScienceLukas @jangrohn
In this article, Holton et al. show that both humans and artificial neural networks have similar patterns of transfer and interference during sequential learning.
nature.com/articles/s41562-0…
Humans and neural networks show similar patterns of transfer and interference during continual learning
New preprint! With
@summerfieldlab @tsonj @ScienceLukas @jangrohn
When does new learning interfere with existing knowledge? We compare continual learning in humans and ANNs, revealing similar patterns of transfer & catastrophic interference tinyurl.com/ykuhwtp2
I finally made a Bluesky account: bsky.app/profile/tsonj.bsky.…
Please follow me there where you can read about @EleanorHolton1 's newest preprint!
Humans and neural networks show similar patterns of transfer and interference during continual learning bsky.app/profile/eleanor-hol…
P.S. I am still on mastodon at neuromatch.social/@jess !!
Se-Bum Paik wrote a great Preview about our recent Neuron paper. It contains this figure, made by Jeonghwan Cheon, which is sooo much better than any of the figures I ever made! 😍🙏
Split-second insights from the brain’s dual-stream visual system
sciencedirect.com/science/ar…
Jess Thompson retweeted
My exclusive for Observer today: For 1st time FOI data from Oxford colleges shows majority of tutorials taught by early career academics struggling to manage on low pay & ‘Deliveroo-style’ contracts. FT profs only taught a third of tutorials last year. theguardian.com/education/20…
Proud of my fellow @OxfordUCU members for bringing this to light.
"The union says that pay is based mainly on contact hours with students, but once preparation, marking and supporting students are taken into account, this often falls below the minimum wage in real terms."
Oxford University named & shamed in the Observer.
Disgraceful that colleges with endowments stretching back to the Middle Ages are forcing staff to work Deliveroo-style contracts.
Amazing work by our @OxfordUCU @OxfordAntiCas members 👏
theguardian.com/education/20…
Jess Thompson retweeted
This is a great opportunity to secure funding for postdoc.
My lab has a few research directions that align with the IVADO initiative: "Shifting Paradigms for a Robust, Reasoning, and Responsible Artificial Intelligence and its Adoption".
Reach out if you’re interested!
🔷IVADO is proud to launch a support program for outstanding national and international postdoctoral researchers.
More info: ivado.ca/en/scholarships-and…
#AI #ArtificialIntelligence #R3AI #Funding #postdoc
Jess Thompson retweeted
Neural networks need a dorsal stream to learn structure in natural scenes! really proud of this work led by @tsonj
How does the brain represent the structure of visual scenes independent of their contents? In “Zero-shot counting with a dual-stream neural network model” (@NeuroCellPress), we propose a model based on the parallel pathways of the primate visual system. shorturl.at/Zp5EA
How does the brain represent the structure of visual scenes independent of their contents? In “Zero-shot counting with a dual-stream neural network model” (@NeuroCellPress), we propose a model based on the parallel pathways of the primate visual system. shorturl.at/Zp5EA
The theory that was formalised and tested in this work was initially described in "Structure learning and the posterior parietal cortex" by @summerfieldlab, Fabrice Luyckx, and @hannahsheahan sciencedirect.com/science/ar…
Thank you to co-authors @hannahsheahan, @atomsrivet, Julian Sandbrink, Manuela Piazza, and @summerfieldlab!
Jess Thompson retweeted
Multiple job openings in brain and cognitive science at MIT! One is for experimentalists studying neural circuits for cognition, a second targets computational neuroscientists, and a third is aimed at computational cognitive scientists. Deadline is Nov 15. Links in next tweet...
Jess Thompson retweeted
We are looking for new PhD / MSc students! Machine learning innovations for tackling climate change - e.g. remote sensing, climate informatics, materials discovery, & biodiversity monitoring. Apply by submitting applications to *both* @Mila_Quebec (by Dec 1) & @mcgillu (Dec 15).
Jess Thompson retweeted
@JacquesPesnot and I are very happy to share this paper. osf.io/preprints/psyarxiv/ry…. Building on work by @scychan_brains we show that human in-context and in-weight learning vary with the distributional properties of the training data in a very similar way to transformer networks.
