(1/17) Have you ever wondered why animals learn simple tasks at vastly different rates? Probably not, but after a pretty bizarre observation a few years ago, @kishoreneuro, @ostojic_srdjan, and I stopped getting frustrated with training our mice and thought a lot about it.pic.twitter.com/b6a4YenhKL
-
-
(4/17) We interleaved reinforced trials with non-reinforced probe trials throughout training, and found that animals consistently reached expert performance levels faster in the absence of reinforcement!pic.twitter.com/UmOtl6PE3l
Show this thread -
(5/17) Reinforcement feedback is obviously critical for learning but, paradoxically, masks the expression of underlying knowledge.pic.twitter.com/2PT9HGZYOG
Show this thread -
(6/17) To ensure this effect was not due to some idiosyncratic element of our task, we performed additional studies in rats, mice, and ferrets. Remarkably, the dissociation between acquisition and expression generalized across all species and a large variety of task designs.pic.twitter.com/aJOEAJBpqP
Show this thread -
(7/17) To better understand this curious behavior, we asked what computational mechanisms may underlie the dissociation between learning curves in reinforced and probe trials. We focused on a network implementation of reinforcement learning for go/no-go tasks, and hypothesized...
Show this thread -
(8/17) ... that learning of action values takes place only during reinforced trials, while the changes between contexts (reinforced and probe trials) do not change the learned values of different options, but modulate only the read-out parameters.pic.twitter.com/yvg8KO3UHl
Show this thread -
(9/17) In particular, we found that selectively scaling the gain of the inhibitory decision unit in our model provided an excellent fit to all of our behavioral data – across species and task designs.pic.twitter.com/Gve9J7WtA9
Show this thread -
(10/17) Finally, we were curious whether the variability in learning and expression was similar across animals. Did “smart” animals perform better in both contexts? Strikingly, while the variability in performance in the presence of reinforcement was huge…
Show this thread -
(11/17) … there was almost no variability in learning in on the probe trials. All animals appeared to acquire the ‘knowledge’ in a roughly identical time frame, but required vastly different amounts of training to effectively express this knowledge in the testing context.pic.twitter.com/WivoCPNQoZ
Show this thread -
(12/17) All together, probing learning by omitting reinforcement appears to uncover latent knowledge that is acquired quite rapidly and in a stereotyped way. Importantly, it identifies context, and not “smartness”, as the key driver of individual variability.
Show this thread -
(13/17) Why is there such a big difference in learning rates and variability in the presence versus absence of reinforcement? Perhaps it’s like studying for a test....
Show this thread -
(14/17) ... you might feel pretty good about the material at home when you’re nice and relaxed, only to absolutely blow it in the classroom when you’re just a bit too anxious….
Show this thread -
(15/17) … or maybe it’s more adaptive. Exploring your environment, even if it leads to errors, can be unexpectedly rewarding. Do you want to try that new dish at your favorite restaurant or go for what you know is good?pic.twitter.com/cNbuLT2ECB
Show this thread -
(16/17) Our data suggest that it’s a mix of maladaptive state-dependent effects and adaptive factors that dissociate acquisition and expression. If that sounds interesting to you, check out the full story!https://www.biorxiv.org/content/early/2018/12/09/489450 …
Show this thread -
(17/17)… and if the neural mechanisms that enable that lie of at the heart of this dissociation are of interest to you too, I’d suggest checking out the Kuchibhotla lab at
@kishoreneuro@JohnsHopkins !https://www.kishorelab.org/Show this thread
End of conversation
New conversation -
-
-
Nvm, I just saw the video. They just lick in the air where the tube was previously!
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
The whole story is really cool but I am confused about something. If you remove the lick tube, how do you get behavioural readout? How do you know whether the animal identified the target stimulus correctly?
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.