petitegeek.bsky.social
Computing Science prof in multimodal embodied AI, emotion, interaction at SFU in Vancouver 🇨🇦🇵🇠Director of the Rosie Lab www.rosielab.ca Robotics nerd. Previously at SoftBank Robotics 🤖 FR/JP
172 posts
3,116 followers
1,389 following
Regular Contributor
Active Commenter
comment in response to
post
Thank you!
comment in response to
post
This work was started by Hatice Gunes' group at Cambridge
arxiv.org/abs/2302.06514
The goal is to model the multiplicity of reactions given that "a person can react differently according to their internal state and other contextual factors".
comment in response to
post
Yes! We also have the bodily influence (heart rate and variability, gut) - is there a way to model these in addition to neural activations? Or is it always mediated through the brain in the end?
comment in response to
post
This one! Just finished open access unfortunately
www.nature.com/articles/s44...
comment in response to
post
Wow, this paper is so cool. Collaborative filtering to model responses like preferences - love it! Reminds me of drift diffusion to model emotion response in this recent nature review paper. Still learning about it, but drift diffusion is also used to model preferences if I understand correctly.
comment in response to
post
I like the wider range of emotions in that study (16) - granularity in the range of emotions that people could be experiencing also seems important. We have been using 26 (although not evoked emotion, perceived ToM). arxiv.org/abs/2310.19995
comment in response to
post
Maybe if we knew the entire personal history of Bob and how he might appraise a situation, eg. once he had a cat that died, so now he gets a little emotional when there are cats in a video.
comment in response to
post
Perhaps the emotional ToM problem is much more complex because groundtruth is harder to nail down. This is one of the issues I have in affective computing - people don't annotate in-the-wild, ambiguous images the same. And I think it's a feature, not a bug (noise). Kind of like the blue-yellow dress
comment in response to
post
Also, this paper has been interesting for me
www.nature.com/articles/s44....
Drift diffusion modeling is also used for preferences I think?
comment in response to
post
What do you think of its similarities to color perception?
comment in response to
post
Generally my sentiment on house ownership 😅
comment in response to
post
There were these in Japan, speaking of drainage making a house look better! en.m.wikipedia.org/wiki/Rain_ch...
comment in response to
post
Great video! Have you done any work extending these analogies to people with reality construction disorders?
comment in response to
post
Have you seen any other ways AI, embodied AI or affective computing research could help? 🤔
comment in response to
post
Some research in interactive agents embody the patient's voices so that a therapist can help the patient be able to reason and talk with their voices. Better TTS could help here.
comment in response to
post
Theory of mind can be highly affected in psychosis. Delusions of persecution, "they're out to get me", show off an impaired ToM and heavy fear response. Integrative models of emotion processing could be interesting to look at
www.nature.com/articles/s44...
comment in response to
post
Psychosis is a breakdown of reality and sensory integration. Hallucinations have a stronger priors compared to bottom-up sensory processing. Robot simulators also need to create reality from predictive processing coupled with sensory input.
comment in response to
post
Thank you so much!!
comment in response to
post
@beausievers.bsky.social would you or @mehr.nz have any pointers to canonical papers on dance as a (universal) form of communication or analogies to conversation/language? Or is universal capacity the main finding?
comment in response to
post
Awesome, thank you!
comment in response to
post
I'm really excited to read this! Any podcasts that you've recorded leading up to the release? Would love to learn more.
comment in response to
post
Ah, yes just get a rechargeable Suica or Pasmo and tap -- much easier and it works on all lines.
comment in response to
post
Well, I would not group them in the toy category. Peppers did work on cruise ships, airports, sushi shops, car dealerships, train stations, etc in Japan and Europe (a bit in North America but not as much). They did the same job as informational touch screens. But not manipulation, that's for sure.
comment in response to
post
I would have thought Aldebaran had that title 🤔 What is the record for humanoid robots shipped?
comment in response to
post
Yikes, I'm doing Air New Zealand direct from Vancouver, no way to combine airlines to go the other direction?
comment in response to
post
It seems it should, I think it's based more on separation of voices based on acoustic properties, @mzmttks.bsky.social could tell you more. It also seems possible to try it using their web interface
comment in response to
post
If you are interested, the founder Dr. Takeshi Mizumoto is truly an exceptional person, speaks English fluently and I'd recommend having a short meeting to see if it would meet your needs.
comment in response to
post
If the kids are working around a table then you can go hands free. In Japan, some of my auditory scene analysis colleagues from Kyoto University created an "egg" you can place on a table that does all the speech analysis and tracking, and has deployed it in elementary schools
www.hylable.com/en/
comment in response to
post
Arghh
comment in response to
post
I have some lectures in my ethics course on value-driven design, where they have to figure out the values of the client personas and make sure that their final project aligns with those. I think they find it empowering rather than the rest of my lectures which is like, Morton Thiokol and depressing.
comment in response to
post
In the autistic-autistic, nonautistic-nonautistic and autistic-nonautistic dyad studies, I wonder if anyone has tried pairing with "people who have grown up with a non-autistic person in their household". Perhaps, these people could code switch.