Congratulations!

No worries at all! I appreciate the additional context, and I agree it's very valuable for readers who might be unfamiliar with the literature. I am glad we had a chance to further clarify things 😀

(3/3) When I talk about RNNs, the key innovation is the use of methods to find the network's fixed points and study the linearized dynamics, like in Sussillo & Barak, 2013. Also, I agree that others had already trained monkeys to do similar tasks, but the analyses are what made the difference here!

(2/3) In my opinion, this is different from using unsupervised methods like PCA (or even more refined methods like demixed PCA) and is more similar e.g. to the ideas of "potent" and "null" axes like in Kaufman et al., 2014.

(1/3) Hi @jbarbosa.org! The transmitter piece was meant for a broad audience so it was hard to describe in too much detail the novel aspects of Mante & Sussillo. When I talk about population dynamics, I refer to their use of targeted dimensionality reduction to find key axes in neural space.

That sounds great!

Thank you Nicole!

Haha I wish we could work something out! Thank you for the nice words! 😀

Thank you very much @nicolecrust.bsky.social!!!

7/7 Finally, we will be presenting a poster at @cosynemeeting.bsky.social in Montreal next week (Poster 1-061). I look forward to chatting about all this, and to present some new results about how rats and RNNs learn to solve flexible decision-making. Please come say hi! 😀

6/7 In my lab @sidb-edinburgh.bsky.social in Edinburgh, we study the neural mechanisms underlying flexible decision-making, and we leverage rat models of neurodevelopmental disorders to study the link between genetic mutations and cognitive dysfunction. We are hiring! www.marinopagan.com/pagan_lab

5/7 What’s next? Using rats opens the door to the use of modern genetic and molecular tools available in rodents to study context-dependent evidence accumulation. It also offers an unique opportunity to study how genetic mutations can affect the neural circuits underlying flexible decision-making.

4/7 Infinite thanks to my wonderful postdoctoral Mentor, Carlos Brody, who was an inexhaustible source of guidance, wisdom and support! 🙏🙏🙏 Also very grateful to our fantastic collaborators, @sussillodavid.bsky.social , Valerio Mante, Mikio Aoi, @jpillowtime.bsky.social , and Vincent Tang.

3/7 Also, we found that recurrent neural networks (RNNs) can fail to capture the heterogeneity observed in biological brains. We introduce new methods to design RNNs that cover the full space of solutions, bridging the gap between artificial systems and the diversity seen in biological computation.

2/7 Most neuroscience research focuses on what is common across subjects, but individual differences can be large for tasks that allow many different internal solutions! Our work highlights the importance of characterizing individual variability when studying complex cognitive behaviours.