Transcript Lab 2 slides

Computational
Cognitive
Neuroscience Lab
Today: Second session
Computational Cognitive
Neuroscience Lab
» Today:
» Homework is due Friday, Feb 10
» This homework has more projects
than the last, but fewer questions per
project
Bias Weights
» Why do we have them?
» Why are some higher than others
in the transform project?
Local vs. Distributed
Representations
» Counting on your fingers--how high
can you count??
» 10, using a localist representation
» Using a distributed representation,
such as a binary code, we can
count to 1024!
What is clamping?
» An analogy to cellular physiology,
where electrodes are inserted into
cells to control the membrane
potential
» Types of clamps: current clamp,
voltage clamp
» Clamping just means to externally
force a state upon the cell
Bidirectional connectivity
» What we have seen so far: bottomup, or stimulus-driven excitation
» Now: top-down, or hypothesisdriven excitation
» What does top-down mean?
Imagine
What does top-down
mean?
» Customary terminology: bottom is
stimulus, top is a brain-state
» Bottom-up: think about a loud
noise that makes you jump
» Top-down: What if you knew to
expect a loud noise? That
expectation might make you jump
less when you finally hear it
Inhibition
»
»
»
»
Benefits
Mechanisms
K-winners-take-all
Sparse, distributed codes