💡The neural mechanisms behind "aha!" moments are revealed in @natcomms.nature.com.
Bridging computational neuroscience and AI, this approach offers insights into learning in biological and artificial systems.
Congrats to @biyuhe.bsky.social and team! Proud to be part of this work.
Learn more 👇
We hope you enjoy the paper!
This was a multi-year team effort, co-led by
@ayakahach.bsky.social , @jonathanshor.bsky.social , Chris Liu, and co-supervised by Eric Oermann and me.
Many thanks to funding from NSF and W.M. Keck Foundation, and support by NYU Grossman School of Medicine!
We evaluated the similarity between the prior information learnt by the ANN and the neural activity in the brain. IT shows the highest correspondence. Moreover, the strength of this ANN-informed learning signal in IT predicts the reliability of the learning effect for a given image in humans.
Our novel DNN captures the magnitude of humans’ one-shot perceptual learning effect, has excellent trial-to-trial error alignment with humans, and the model’s internal features predict whether humans will successfully learn a particular image.
To elucidate potential computational mechanisms, we built a novel ANN to capture the one-shot perceptual learning phenomenon.
This ANN combines top-down signaling with a transformer-based vision backbone. Top-down signaling conveys information accumulated from past experiences.
The invariance properties of the stored information match the invariance properties of neuronal coding in inferotemporal (IT) cortex.
With iEEG, we asked which brain region has the earliest-onset activity changes before vs. after learning. IT cortex shows the earliest change.
In this study, we combined psychophysics in healthy human subjects, intracranial EEG (iEEG) in neurosurgical patients, and artificial neural networks (ANN) to tackle this question.
With psychophysics, we asked what kind of information is stored in the knowledge encoded by the brain.
We previously showed that this dramatic one-shot learning ability does not require the hippocampus, the most famous brain structure for L&M. (www.pnas.org/doi/10.1073/...)
There are widespread activity changes in the neocortex after learning. But where and how learning occurs was mysterious.
We used a one-shot perceptual learning task inspired by the Dalmatian Dog picture.
Degraded two-tone images are initially difficult to recognize but effortlessly recognized once the corresponding original picture is viewed. This learning effect is both fast and long-lasting in humans.
“Aha” moments, flashes of insight, and one-shot perceptual learning are impressive feats of the human brain.
With intertwined empirical and modeling investigation, this paper takes us a step closer to understanding this singular human capability and building it in machines.
Ever wondered how the Dalmatian Dog effect works?
The human brain can quickly learn from a single experience and generalize it to related experiences — an impressive feat so far not matched by AI.
Our new paper in @natcomms.nature.com reveals how this works.
www.nature.com/articles/s41...
Just in time for Thanksgiving!
Here is my (very) short commentary titled "On the role of theories in consciousness science" published now in
@commspsychol.nature.com :
www.nature.com/articles/s44...
Just in time for Thanksgiving!
Here is my (very) short commentary titled "On the role of theories in consciousness science" published now in
@commspsychol.nature.com :
www.nature.com/articles/s44...
Going to SfN? Check out our lab's presentations! We have a talk on Sat afternoon (1:15 pm, SDCC Rm 11, NANO004.02) by @dhasemusic.bsky.social and two posters on Wed afternoon (PSTR469.20 / OO3 and PSTR469.21 / OO4) by @ayakahach.bsky.social and Jonathan Shor.
Check out our lab’s new paper by the brilliant @luakoenig.bsky.social !
Many thanks again to all the speakers who participated, and the staff who made it possible!!
Full list of speakers can be found here:
sites.google.com/view/conscio...
My first post here!
My article summarizing our recent workshop on consciousness at the NIH is now out in
Neuron.
50 d free-access link:
authors.elsevier.com/a/1hxjg3BtfH...
This article presents a vision for consciousness research ahead that is more empirically grounded & broader in scope.