Oliver Vikbladh

Preprint for "Consolidation of Sequential Planning" with @evanrussek and @NeilBurgess10. We demonstrate the neural representations supporting sequential planning and the effect of memory consolidation upon them: biorxiv.org/cgi/content/sh…. As talk, see also: youtube.com/watch?v=0QcNtz…

@EdvardMoser Really impressive identification of this functional circuit! Even better it just needs continuous attractor dynamics & firing rate adaptation (or similar) to work, which also makes some new predictions (model with @zilong_ji, Tianhao Chu, Si Wu) sciencedirect.com/science/arti...

Cognitive models of behavior are a key part of neuroscience. But discovering them is hard! In new work from @GoogleDeepMind Neuroscience and collaborators @HHMIJanelia, we demonstrate an approach that uses LLMs and large datasets to discover models automatically.

I’m excited to share our new, open-access paper published today in @NatMentHealth, “Latent mechanisms of language disorganization relate to specific dimensions of psychopathology”. nature.com/articles/s4422…

New preprint with @DaniSBassett and @nathanieldaw! How do humans learn predictive representations? We propose a trial-by-trial learning rule that incorporates trace updating to learn the SR. biorxiv.org/content/10.110…

👀🍟Now out in Cognitive Science 😋🤖 We present a new approach that combines cognitive models and neural networks to predict latent preferences. onlinelibrary.wiley.com/doi/10.1111/co…

@paul_b_sharp @evanrussek lastly, the task doesn't cover all planning settings but it seems to me to be highly representative of how some planning proceeds, (when start state is fixed, its not obvious where the rewarding states are). e.g. "do you know some place we might get food around here?"

@paul_b_sharp @evanrussek whether the consolidation is going on through offline sampling or online planning is also something we should dig into more. Of course, its a lot easier to detect rollouts here this way than offline because we can time lock their onset to the start of trial.

Hey @vikbladh and @evanrussek, this work is great & inspired further qs. One is whether the MEG analysis fails for Ps that switch b/w rolling out forward & backward. Thus, is an assumption here that roll-outs follow the same direction within-subject over trials? I guess strong evidence against this being empirically a problem is that you see a correlation with the MB parameter estimated from behavior & the MEG signature. It seems too the task pushes Ps to do forward roll-out given that only the starting state is overtly defined, versus the reward or stopping states that are uncertain, as the single feature defining them could apply to several unique states. I have plenty of other qs but it would be more fun/fruitful I think to talk about it in person at some point (e.g., is reactivating a sequentially-encoded memory the same thing as planning (and more about what is being consolidated), qs about how people might sample offline, qs about representativeness of this kind of planning problem).

@paul_b_sharp Thats cool :). Look forward to hear what you think.

@vikbladh Yes this looks fantastics -- I actually RT'd the biorxiv tweet :)

The 2-step task could be solved by using a cog map to assign credit to non-taken action post rare transition. This differs from using a cog map to plan action values at decision time. Does existing work test which is used, & do we think the former is an instance of planning?

@evanrussek @NeilBurgess10 By showing the evolution of the representational basis for rollouts (MTL to PFC), we address the fundamental question of what systems consolidation is actually for. Not simply how representations change, but directly, how they contribute to subsequent flexible behaviour.

@evanrussek @NeilBurgess10 Using this method we can not only decode the contents of rollouts through the transition structure, but also, across trials within participants, decode the speed of sequential simulation. Faster reaction times are related to faster rollouts - i.e. thinking fast means acting fast.

Preprint for "Consolidation of Sequential Planning" with @evanrussek and @NeilBurgess10. We demonstrate the neural representations supporting sequential planning and the effect of memory consolidation upon them: biorxiv.org/cgi/content/sh…. As talk, see also: youtube.com/watch?v=0QcNtz…