This preview shows page 1. Sign up to view the full content.
Unformatted text preview: and chemical architectures.
Big questions remain regarding sensible chemical inference, the
right choice of neural architecture, how to implement complex
models and energy costs.
Bayesian inference can be burdened by prior information in an
un-modeled non-stationary world.
NEXT: Control. Laplace transforms to study ODE systems. Nick Jones Sampling and the Brain: Inference, Control and Driving Bibliography
The paper below is the basis for this lecture. There are a few bugs
but it’s both interesting and well written. There are a couple of
implementation disconnects. In their paper: a) the term απji in the
Hamiltonian should be −απn b) they actually used a smaller
lattice than they state (a 5 by 5 image - that helps explain Fig.
2a) c) they suppose the input image has values +1 or −1 and they
used w in a discretized fashion.
There are also some diﬀerences between the form we discuss and
implement mostly for reasons of simplicity.
Multistability and Perceptual Inference - Samuel J. Gershman
Edward Vul Joshua B. Tenenbaum - Neural Computation 2012
Society of America A, 20(7):14341448, 2003. Nick Jones Sampling and the Brain: Inference, Control and Driving...
View Full Document
This document was uploaded on 03/01/2014 for the course EE 208 at Imperial College.
- Spring '14