Perception as Inference: The Brain and Computation | Theory Shorts
Вставка
- Опубліковано 9 лип 2024
- Theory Shorts is a documentary web series that explores topics from the Simons Institute’s research programs.
Episode 1, “Perception as Inference: The Brain and Computation,” explores the computational processes by which the brain builds visual models of the external world, based on noisy or incomplete data from patterns of light sensed on the retinae.
HOST
Bruno Olshausen
DIRECTOR
Christoph Drösser
EDITOR
Michaelle McGaraghan
PRODUCERS
Kristin Kane
Michaelle McGaraghan
SCIENTIFIC ADVISOR
Shafi Goldwasser
ANIMATORS
Caresse Haaser
Christoph Drösser
Lukas Engelhardt
GRAPHIC DESIGNER
Barry Bödeker
VIDEOGRAPHERS
Drew Mason
Omied Far
Michaelle McGaraghan
Matt Beardsley
PRODUCTION ASSISTANTS
Christine Wang
Bexia Shi
Lior Shavit
THEME MUSIC
“Plastic” by Purple Moons
Courtesy of Marmoset in Portland, Oregon
OTHER MEDIA COURTESY OF
Bruce Damonte
Arash Fazl
Anders Garm
Jean Lorenceau and Maggie Shiffrar
Beau Lotto
A. L. Yarbus
Bruno Olshausen
videocobra / Pond5
BlackBoxGuild / Pond5
nechaevkon / Pond5
DaveWeeks / Pond5
CinematicStockVideo / Pond5
BananaRepublic / Pond5
MicroStockTube / Pond5
shelllink / Pond5
AudioQuattro / Envato Market
HitsLab / Envato Market
FlossieWood / Envato Market
plaincask / Envato Market
MusicDog / Envato Market
Loopmaster / Envato Market
Ryokosan / Envato Market
Images used under license from Shutterstock.com
© Simons Institute for the Theory of Computing, 2019
I liked the way he scratched the back of his neck right after he said "we're only still just scratching the surface..."
Fantastic video, content, presentation and production 10/10.
Wonderfully clear explanation. Hope you post more such videos!
There is a bistable percept in music, when listening to polyrhythmic music, the brain can jump back and forth between various perceptions of the rhythm - 3/4 vs 6/8 for the same half measure of 12/8, in certain African music, for instance. In the flamenco rhythm of bulerias, "getting the rhythm" entails cultivating the ability to control when this shift in perception is made, so that we consciously and intentionally shift from one perception to the other every half measure (usually). Thus the "feel" of the rhythm, and consequent shared socio-musical/cultural experience, is one of collective, specifically controlled shifts in the perception of the bistable percept. Thank you for giving me language to express this thought I've had for a while.
@@saskiapannekoek ha! Small world. I did CogSci at Cal, and got thinking about this stuff ages ago, but never had language for it. Great video!
Super cool love the video thanks
Very interesting and informative video. Thank you.
love this, great intro !~
Amazing video!
Nice.
god i feel so smart right now
around 16:40, the bistable percept of the moving diamond image. I understood the explanation, but my mind wouldn't "see" the sideward moving object when the bars became white again. When I slowed the playback to 0.75 speed, then my mind was able to go back to the late vision impression of a diamond (and back again to the vertically moving lines). So my question is: does this effect have to do with higher cognition of objects via "mental" states of a human cognizing an object as an instance of the type of a diamond concept undergoing sideward motion; or can the effect be demonstrated in other mammals where we don't need to presuppose anything mental or concept-like? How far down the phylogenetic tree does this effect go?
It seem to me that seeing the sideward motion of the late-vision constructed image captures what lay people think of as "conscious" seeing, as opposed to seeing without explicit awareness. Like I can hit back a tennis ball when it approaches me, but I don't "consciously" construct a round shape. But to have the sideward moving diamond pop out into my awareness after the initial up-down lines impression, I need to activate my human concept of a diamond shape. So does this type of example call for an explanation that distinguishes the non-conscious percept in early vision from the somehow "conscious" vision of a mental state of seeing, in conjunction with late vision, something as instantiating a particular concept ? Maybe the content of early vision is not penetrable to conceptual processing, while the content of late vision processing and associated percepts are penetrable to conscious concepts. I do research on natural language processing, and having a clear, empirically-grounded distinction between percept and concept would be helpful for language modeling.
So the retinal pathway sends to the brain a highly processed image. Has there been any research about what is the structure of this 'image'? Mimicking these structural properties may be useful for more robust learning in ANNs.
The structure seems to be an activation pattern of two-way signalling -- the perceived features flow from retina to late vision, while the competing hypothesis of what a brain region expects to see sends signals in the opposite direction. Within each area of visual feature processing, there is a retinotopic map of the scene or object in the visual field, with each successive map becoming less like the original shape on the retina as higher features are picked out. So there seems to be a bidirectional graph of vision processing areas, with signals of retinotopic maps transformed with each area transfer, and constrained by backward signalling about expected data. Taking a clue from how linguistic perception might be structured as "typed feature structures" (in a mathematically precise type theory), perhaps early and late vision can also be modeled as a typed feature structure of retinotopic maps.
Bistability == attractor of period 2.
who is this lecturer?
Muahhhh to the banana republic