Totally agree!
-
-
Replying to @tyrell_turing @AdamMarblestone and
What's a molecularly annotated connectome? I'm on the side that a connectome is not going to be that useful but happy to be proved wrong. Certainly the brain has myriad *unseen* mechanisms that contribute to function but not evident in a connectome.
2 replies 0 retweets 5 likes -
Replying to @JasonSynaptic @tyrell_turing and
In C. elegans, it is useful for hypothesis generation and for realizing how ubiquitous signaling relationships that ignore the connectome are. Does very little work constraining circuit models.
1 reply 0 retweets 8 likes -
Replying to @MHendr1cks @JasonSynaptic and
The best sketch I have online is this: https://arxiv.org/abs/1404.5103 Think of it as layering a lot of in-situ spatial transcriptomics & proteomics, on top of connectivity, probably all obtained optically & with the benefit of expansion microscopy... would include modulatory receptors.
2 replies 0 retweets 6 likes -
Replying to @AdamMarblestone @MHendr1cks and
Without a really solid foundation in molecular biology having a bunch of in situ sequencing layered on top of connectivity will be pretty
. With sequencing you have to have a pretty good idea what you are looking for; e.g. antisense lncRNA and promoter choice of protocadherins.1 reply 0 retweets 0 likes -
Replying to @wholebrainsuite @AdamMarblestone and
I disagree. Work by Josh Huang and the Allen Institute has given characterized great markers that in conjunction help identify many inhibitory and excitatory cell types. That will go a long way towards interpreting reconstructed EM volumes.
1 reply 0 retweets 0 likes -
Replying to @neurosutras @AdamMarblestone and
But is the goal of molecular biology in neuroscience to identify cell types? Very depressing if so.
1 reply 0 retweets 0 likes -
Replying to @wholebrainsuite @neurosutras and
Fortunately there is a lot of mol bio of learning/memory to build on. And can compare before/after various learning. Point is you still want all this. Even if you interpret computationally by seeking ~ architectures & cost functions that will get optimized, and ask which & how.
1 reply 0 retweets 0 likes -
Replying to @AdamMarblestone @neurosutras and
Right. But wouldn’t one want to be way way more explicit about how one actually thinks about the mechanisms by which these architectures and their structure is brought about. Simply labeling large parts of neuroscience as stamp collectors seems unhelpful at best.
1 reply 0 retweets 0 likes -
Replying to @wholebrainsuite @neurosutras and
I do prefer this 2 come w/ large dose of circuit hypothesis detail. Refactor much neuro as looking 4 cost fxns, optimizers, conserved dev programs + what doesn’t fit that. As Blake’s dendritic DL work does. & yes “tuning” gives insight to it. Pytorch code=result, not sole input.
2 replies 0 retweets 1 like
(Though an appealing aspect is one can *also* try to make pytorch code based mostly on other considerations and see how it does, i.e., much current cognitively oriented AI, if you want to try to ~ignore biology.)
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.