Sascha Frölich
@saschafrolich.bsky.social
Postdoc @ Cognitive Computational Neuroscience Lab, TU Dresden, Germany.
Interested in Cognition, Neuronal Computation, Statistics and ML.
Interested in Cognition, Neuronal Computation, Statistics and ML.
Another limiting factor is the size of the brain. What functions are so complex that the human brain is simply too small to approximate? And how does this translate to generative models?
November 7, 2023 at 9:03 PM
Another limiting factor is the size of the brain. What functions are so complex that the human brain is simply too small to approximate? And how does this translate to generative models?
Great thought. In theory, finite layered neural networks can approximate any function: en.m.wikipedia.org/wiki/Univers...
Of course, biological neurons don't have activation functions that are defined on [-inf, +inf], but my guess is that the result is a (perhaps negligible) loss in precision.
Of course, biological neurons don't have activation functions that are defined on [-inf, +inf], but my guess is that the result is a (perhaps negligible) loss in precision.
Universal approximation theorem - Wikipedia
en.m.wikipedia.org
November 7, 2023 at 8:58 PM
Great thought. In theory, finite layered neural networks can approximate any function: en.m.wikipedia.org/wiki/Univers...
Of course, biological neurons don't have activation functions that are defined on [-inf, +inf], but my guess is that the result is a (perhaps negligible) loss in precision.
Of course, biological neurons don't have activation functions that are defined on [-inf, +inf], but my guess is that the result is a (perhaps negligible) loss in precision.