Do you think this is possible in the absence of functional annotations?
-
-
Replying to @neurowitz
I think so. But it is hard.
1 reply 0 retweets 0 likes -
Replying to @KordingLab @neurowitz
We tried. Tough to infer connections, and to infer *changes* in connections is way harder. Must measure many synapses at 2 times.
2 replies 0 retweets 2 likes -
Replying to @xaqlab @KordingLab
I think
@SebastianSeung has been thinking about potential ways around this problem...1 reply 0 retweets 1 like -
Would be very curious to see his ideas.
2 replies 0 retweets 0 likes -
Replying to @KordingLab @neurowitz and
I may come across as molecular annotation extremist, but I bet with enough static molecular detail one could get something indicating dS/dt where S is synaptic strength.
3 replies 0 retweets 3 likes -
Replying to @AdamMarblestone @KordingLab and
supposing one has only a coarse anatomical "correlate" of strength (number of synapses; cumulative area of contact maybe) how good can one do at predicting circuit behavior(s)? iow, are there any cheeky shortcuts where we can do "good enough" without knowing precise strengths?
1 reply 0 retweets 1 like -
Replying to @jipkin @AdamMarblestone and
That is a great question. And I have no idea why the connectomics field does not take it seriously!
1 reply 0 retweets 1 like -
Replying to @KordingLab @jipkin and
@tyrell_turing probably dS/dt is more important than specific strengths S in your picture, since gets more to learning rules?1 reply 0 retweets 3 likes -
Replying to @AdamMarblestone @KordingLab and
Yes, exactly! dS/dt over multiple time scales is the key measurement, IMO, and should be a better reflection of what is being optimized. Also, note that the learn alg may involve non-synaptic modifications as well, so even if we could measure sn strengths, we might miss key info.
1 reply 0 retweets 2 likes
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.