NC unit secretary
2020-10-06 20:04:31

This is a discussion channel for " Reverse Engineering Visual Intelligence" by Prof. James J. DiCarlo (Massachusetts Institute of Technology) A link to the talk is the following. Please do not share the link to anyone outside of this slack workspace. Access Passcode can be found at the announcement channel.

URL: https://vimeo.com/471283865 (33 minutes)

👍 桑原優
David Rotermund
2020-10-08 06:16:21

Is it only me (my two computers) or is the talk stuck on slide 1? The face-cam and audio are working fine but I see only slide 1 for the whole duration...

Javier-tejeda
2020-10-08 08:33:56

*Thread Reply:* Same here 😣

NC unit secretary (inactive)
2020-10-08 12:10:38

*Thread Reply:* Thank you for pointing this out. We asked Prof. DiCalro for his slides. We will re-upload the talk as soon as possible.

NC unit secretary (inactive)
2020-10-08 12:12:08

As noted by David, the talk currently has an issue that only the first slide is shown. We asked Prof. DiCalro for his slides. We will re-upload the talk as soon as possible.

} David Rotermund (https://aibssymposium.slack.com/team/U01CTQNBNAC)
NC unit secretary
2020-10-10 07:31:11

@David Rotermund @Javier-tejeda Prof. DiCarlo's movie is now available. Thank you.

Shin'ya Nishida
2020-10-10 21:05:30

Hi Jim, great talk. Is there any theoretical explanation as to why V1-like initial layer contributes to adversarial robustness? How much can it be explained by spatial-frequency characteristics of contrast response function (specifically low sensitivity to high-frequency components)?

👍 Leon WAN, Taiki Miyagawa
Achler
2020-10-13 00:51:16

I appreciate the repeated analogy how the deep ANN's are partial and a glass half full while matching some neural phenomena. The brain schematic showed much more feedback and I know backprop training does not play well with direct feedback. Nor does the implied rehearsal mechanism which has to store all data and rehearse it in a non biologically plausible way. My question is whether there are plans for more efforts to be dedicated to some of those feedback connections. See poster 26 "Homeostatic recognition circuits emulating network-wide bursting and surprise".