Logo

Research at the Image and Video Understanding Lab (IVUL) - Shared screen with speaker view
Asaad Alghamdi
31:54
What is the correlation between visual computing and self-driving?
Konstantin Burlachenko
47:33
"Images" is a matrix of pixels, which is pretty unstructured. For such class of input variables function composition(deep learning) seems extremely good function family (empricially) from which we select approximation....What is a function family which seems good well for video?
Konstantin Burlachenko
58:52
I have learned in the past that any prediction problem from statistical learning theory is described by 3 things:1. Signal and noise ratio (observable, non-observable data)2. Target function3. Amount of data and suitability of this specific taskCan you please comment about each of this points in context of video understanding. Because I think maybe we can not just take "all videos" and push them into "one bucket" ?
Konstantin Burlachenko
01:01:50
If system provides infinite amount of data how sample train samples(samples) from it?
Konstantin Burlachenko
01:07:44
Drones, self driving cars and we have no idea how it works (of course in case of using DL, ML). Is it ok? (SpaceX Falcon 9 First stage landing is performed by Convex Optimization + First principle based models by PhD student of Stephen Boyd)
Konstantin Burlachenko
01:12:30
About measurement "system works better compare to human". I guess "averaging" is incorrect behavior, because we are interesting in behavior of the model in a specific "input". So how such "averaging" system behavior measurement in AI/ML/DL correlated with reality?