But, if we want to know how the brain works, doesn't that mean understanding how it engages in sophisticated computations? If I understand the biology, but not how it produces behaviour, do I understand the brain?
Comments
Log in with your Bluesky account to leave a comment
I’m a neophyte undergrad on the computational side, but my intuition is there has to be some level of correspondence in our models between both approaches, and that’s going to look different at different levels. Maybe not “top-down” or “bottom-up” but “middle-out”?
Yeah, I hear you, and see why that would be a reasonable intuition. But, I have generally found it's better to really get the top-level working, then push down to fill out the middle. That's just a personal observation, though.
Personally, I see getting the top-level working as a proof-of-principle test of theory plausibility. The initial hypothesis can be inspired by something in the brain ("perhaps brain phenomenon X emerges from using algorithm Y to solve problem Z under bio-constraints W")...
...it's much easier to then demonstrate the emergence of X in a system you can actually get working in practice... then add in biological specifics, than it is to build a model with many specifics (which ones?) and try to get it working in anything resembling a real-world situation.
It’s impossible to not include top down stuff if a goal is to predict unknown reasons for a thing. Even in modeling cellular, it’s in studying bifurcations/fixed points/limit cycles etc arising from varying sets of parameter values that I get options for candidate paths to what’s observed in biology
Part of the beauty of modeling is that you can find various possibilities for something and then subsequent experiments can verify or nullify them, and either one (verification or nullification) moves understanding forward.
I agree with you. And don’t worry about being an undergrad — I feel nervous responding to these conversations as a early PhD student too but imo it’s for the most part been rewarding to contribute to discussions if you have something you want to say.
Give it 10 years and you'll be firing out replies willy-nilly with no consideration of the fact it makes you look like a real ass on the internet ;p. This is why junior trainees make the best posts - they still have good questions and they care enough to write them well.
I’m not sure how sophisticated they are, and that’s a bit of the question? Biology comes up with stunningly simple computations for seemingly sophisticated behaviors? And carving a path seems (maybe deceptively) easier by starting with constrained bio bits.
I’m modeling a cellular process, even then it’s a fight between abstraction & replication, there has to be abstraction at some level. It’s maybe enough to say one receptor’s endocytosis rate is larger than another’s, without exact values. And countless other receptors, pathways etc not included.
I think about the values behind the spruce budworm model as a guide. It didn’t model every part of the budworm ecology or every predator or every tree, but its predictions can prevent forests from being decimated.
And so I agree, I don’t think any particular approach is better or worse, what matters is if it can help our understanding of the brain or get us closer to helping the quality of life for humanity
Comments