Tim Maniquet

FWO PhD researcher, vision neuroscience

[Tim Maniquet @TManiquet]

⚠️PREPRINT⚠️ My first paper!

Do you know everything about DNNs, human vision and recurrent processing? No? We neither. But we tried to learn a bit about them, and our results are not what we expected.

Check out our work with @hans_opdebeeck & @costantino_ai ⬇️⬇️ 1/9

[Tim Maniquet @TManiquet]

Great times in Boston! What a pleasure it was to present at #CCN24. Will certainly go to Amsterdam at #CCN25 for more models & brains 🧠

@hans_opdebeeck @costantino_ai @leuvenbrainins1 @BiolPsychol

[Hans Op de Beeck @hans_opdebeeck]

We had a great time at the CCN meeting @CogCompNeuro with @TManiquet @costantino_ai and Mualla Gunay, already looking forward to next year in Amsterdam organised by @hsteven9 @iris_groen

[Tim Maniquet @TManiquet]

Great times in Rovereto! Really enjoyed presenting my project at #caos2024, as you can see on this great pic with coauthor @costantino_ai who’s literally behind the work.

See you next year!
@hans_opdebeeck @leuvenbrainins1

[LBP @BiolPsychol]

🚨LBP is hiring! We are looking for a full-time lab assistant that can help us on a daily basis.
Friendly & inclusive work environment guaranteed 👌🏼

For more information, check below ⬇️
https://t.co/J3Hc39Hkns

Please RT.

[Tim Maniquet @TManiquet]

@kparikh2001 @hans_opdebeeck @costantino_ai @filippie509 Hey! thank you for the comment. We checked PVM, it looks interesting indeed. For sure a possibility for future studies, thanks!

[Tim Maniquet @TManiquet]

⚠️PREPRINT⚠️ My first paper!

Do you know everything about DNNs, human vision and recurrent processing? No? We neither. But we tried to learn a bit about them, and our results are not what we expected.

Check out our work with @hans_opdebeeck & @costantino_ai ⬇️⬇️ 1/9

[Tim Maniquet @TManiquet]

@guo9cheng @hans_opdebeeck @costantino_ai thanks! no, we haven't checked that, thank you for the idea!

[Tim Maniquet @TManiquet]

@yonicoud @hans_opdebeeck @costantino_ai Merci! Non, pas encore. Décembre 2025 normalement 😀

[Tim Maniquet @TManiquet]

@MartinMwiener @hans_opdebeeck @costantino_ai Indeed; overall RT analyses replicated accuracy effects, so we included what felt the most relevant. The barplot here is a correlation between human RT per task and model performance per task, what we call 'task consistency'

[Tim Maniquet @TManiquet]

@MartinMwiener @hans_opdebeeck @costantino_ai Thanks! And good question: we actually find similar results for RT. We looked, but overall we replicate the same pattern across analyses.

Here's the consistency across tasks, between model performance and human RT:

[Tim Maniquet @TManiquet]

To us, this gives perspective on the way we implement recurrence in our models, and on the benchmarks we use to compare DNNs to humans.

Find out all about this and more in our preprint here:
https://t.co/NtObwZzYlL

[bioRxiv Neuroscience @biorxiv_neursci]

Recurrent issues with deep neural networks of visual recognition https://t.co/c4Px1QiXxf #biorxiv_neursci

[Tim Maniquet @TManiquet]

We can then correlate these confusion matrices between models and humans. Here’s what we found: the more recurrent models are, the less they match with humans. 9/9

[Tim Maniquet @TManiquet]

Best thing about two presentations in a week: you can use the same poster twice!
Once for @leuvenbrainins1 's meeting & twice for Neurocog 2023, where I was lucky enough to receive a poster price 🎉🎉🥳

Note how I changed sweater in between conferences