Zoom Logo

DDMB - Shared screen with speaker view
Linus Schumacher
02:23:01
Why is the diffusion operator the right way to weight short vs long distances?
James H.K. Ooi
02:35:40
With neural ODE technique, does steady state of the ODE matters to predict patients outcome?
Alan Aspuru-Guzik
02:37:40
I was going to add that in an unrelated application of Neural ODEs (for quantum computing) our system reached the fixed point or steady-state remarkably well even though it was never trained on it
Alan Aspuru-Guzik
02:37:49
But this may be system-specific (still unpublished work)
Smita Krishnaswamy
02:38:19
Interesting! I guess the steady state of immune cells is probably death or something so I didn’t extrapolate so far :)
Alan Aspuru-Guzik
02:38:48
Yes but also the fixed point of quantum systems is quite boring, they are also “dead” from the quantum computing perspective :)
James H.K. Ooi
02:48:00
Thank you Smita and Alan for weighing in! Very interesting talk.
Stephen Zhang
03:00:52
Can this kind of methodology be extended to stochastic dynamical systems?
Stephen Zhang
03:01:19
Ok thanks!
Alan Aspuru-Guzik
03:02:18
Rodrigo Vargas in my group worked on this problem with Paul Brumer on his past postdoc (Gaussian Process for molecular dynamics, he has not published but it seems connected)…
Sui Tang
03:03:09
Thanks for pointing out!
James H.K. Ooi
03:04:07
I look forward to Rodrigo V talk this Thursday :)
Smita Krishnaswamy
03:04:09
Sui have you tried a graph neural network or graph attention network for the interaction learning?
Sui Tang
03:04:43
I have not tried and I have seen some works in this direction. I will check them out! Thank you!
James H.K. Ooi
03:44:20
Hi Alan, interesting talk! Regarding the Genetic Algorithm, do you manually tune the mutation rate / crossover rate? Or does the feedback readjust these rates?
Mengyang Gu (UCSB)
03:48:54
Very interesting talk Alan, I have two questions. 1) how can ab ab initio MD simulation help identify molecule structures outside the space of a database (as extrapolation directly may be hard); 2) how to efficiently represent and learn through a long vector of inputs from Selfies respresentation
Alan Aspuru-Guzik
04:00:41
Contact alan@aspuru.com
Alan Aspuru-Guzik
04:00:47
Twitter @A_aspuru_guzik
James H.K. Ooi
04:00:51
We are using SELFIES and it save us alot of compute times.
Alan Aspuru-Guzik
04:00:59
Fantastic James
James H.K. Ooi
04:01:00
Thanks Alan :)
Alan Aspuru-Guzik
04:01:01
Love to hear
Alan Aspuru-Guzik
04:01:09
Happy to help if you need extensions of the software package
Alan Aspuru-Guzik
04:01:20
Nice to meet you all!!