18/02/2026 4:30 PM - Lecture
18/02/2026 4:30 PM - Lecture
Gestalt and Perceptual Grouping
Things that have properties that cannot be derived from component parts.
We endow things with meaning.
Thinking that we don't really build things from the ground up, we compose wholes and think in those wholes.
Visual Gestalt: Often used is advertising to get you to look at things.
Equivolcation - Perceptual ambiguity. Old vs young lady. Multiple valid interpretations. Neckar cube.
Continuance - Connect similar phenomena, construct a timeline. Making up a story that connects disconnected objects. Is it the same object in these places?
Closure - We complete incomplete images. Like a cutoff image being completed. Kanisza triangle. If it fits I sits.
Grouping - We can group based on lots of stuff. Alignment, color, shape
* Common fate - Things moving in the same direction get grouped
Scale constancy - We do not assume people are smaller because they are further away.
* Context around allows us to make things appear big or small based on scale constancy. Things the same size in the image may appear wildly different sizes based on scale constancy assumptions
Color Constancy - When we know things are the same color, we make it the same color. In shadow? Still the same color.
Composition - Elements together make up a whole
Figure-ground: Is this the foreground? Or background? There are sometimes equivocations with this where swapping which is which leads to different interpretations.
Doug's Talk
Actuallly find? If not pitch competetions, where?
What changes when you are a hard problem fund?
Do you generally know if the market exists before talking to the founders? Do they convince you?
In most AI companies, what is the moat? If the true insight is that AI can be applied to a new area, how can you know that somebody else will not quickly replicate the dataset and then model?
* Acutally dataset and go to marketability is growing more important as programming gets easier with AI.