|
|
|
|
Our Papers Latest preprints : Hebbian Crosstalk Prevents Nonlinear Unsupervised Learning PDF and the arXiv link Kingsley Cox and Paul Adams This paper shows that in a popular model of nonlinear Hebbian learning, there is a critical level of synaptic update error beyond which the system is unable to extract any useful information. This is the first evidence that a critical error threshold may well be a limiting factor regarding learning in the cortex and supports our contention that a large part of cortical circuitry is involved in error correction by proofreading.
Hebbian Inspecificity in the Oja Model PDF and the arXiv link Anca Radulescu, Kingsley Cox and Paul Adams This paper explores with analysis and simulations the role of Hebbian synaptic update error in the Oja Model, which is a linear system. We find that error, as expected, degrades performance but that there is no sudden breakdown, in contrast to the nonlinear case.
Recent Meeting and Posters/Abstracts
|