3 ms·

Honestly, it just sounds like the field has managed to go from taking a big blob ('cpu'/'gpu') and finding basic relationships ('matlab vs games') to working wi

by ColanR 9y ago

Honestly, it just sounds like the field has managed to go from taking a big blob ('cpu'/'gpu') and finding basic relationships ('matlab vs games') to working with much smaller blobs that are closer to the base principles at work. Illustrating from your example, current state of the art seems to have managed to break the big blobs into smaller blobs (e.g., now we're looking at 'memory that binds multiple features between objects'), and then found more complex relationships ('looking at separation or completion operations in similar memories'). That still doesn't tell us how the actual programming works. We barely understand the role of dendritic spines, and we're still trying to get a handle on the utter complexity of single neuron interactions in the neocortex. He might have oversimplified, but I don't think he's wrong. (Just playing devil's advocate. Would love to know how badly I misunderstood.) Edit: nearly all of your links are paywalled.

SubiculumCode 9y ago
Oh sure there is some blobology to it still...but the nature of the function attributed is becoming more precise.--not a memory blob, a blob associated with signal orthoganalization. But that is just the functional neuroimaging in humans in which we are working downwards, while neuroscientists using methods that influence activity at the neural level, or even the dendritic level, work upwards. Each informs the other. For example, the computational models of pattern separation and completion are derived by observations of the structure of connections between neurons, which led to hypotheses about what kinds of operation it could support, which led to tests of that hypothesis in rats, and eventually in humans. But at the heart of it, I don't think we have to be completely reductionistic to gain understanding. We don't have to fully understand in every detail how dendrites operate to understand that a collection of neurons can support a particular data transformation.

by ColanR 9y ago

SubiculumCode 9y ago

Oh sure there is some blobology to it still...but the nature of the function attributed is becoming more precise.--not a memory blob, a blob associated with signal orthoganalization. But that is just the functional neuroimaging in humans in which we are working downwards, while neuroscientists using methods that influence activity at the neural level, or even the dendritic level, work upwards. Each informs the other. For example, the computational models of pattern separation and completion are derived by observations of the structure of connections between neurons, which led to hypotheses about what kinds of operation it could support, which led to tests of that hypothesis in rats, and eventually in humans. But at the heart of it, I don't think we have to be completely reductionistic to gain understanding. We don't have to fully understand in every detail how dendrites operate to understand that a collection of neurons can support a particular data transformation.