Part 3: Autonomous Agents 11/3/04 Reorganization of Cortex Orientation Columns • Median nerve sectioned to show fluidity of cortical organization • (C) before • (D) immediately after • (E) several months later 11/3/04 (fig. < McClelland & al, Par. Distr. Proc. II) 1 11/3/04 (fig. < Nicholls & al., Neur. to Brain) 2 Slow Potential Neuron Orientation Columns 11/3/04 (fig. < Nicholls & al., Neur. to Brain) 3 11/3/04 (fig. < Anderson, Intr. Neur. Nets) 4 1 Part 3: Autonomous Agents 11/3/04 Variations in Spiking Behavior Frequency Coding 11/3/04 (fig. from Anderson, Intr. Neur. Nets) 5 11/3/04 6 Chemical Synapse Synapses 1. 2. 3. 4. 11/3/04 video by Hybrid Medical Animation 7 11/3/04 (fig. from Anderson, Intr. Neur. Nets) Action potential arrives at synapse Ca ions enter cell Vesicles move to membrane, release neurotransmitter Transmitter crosses cleft, causes postsynaptic voltage change 8 2 Part 3: Autonomous Agents 11/3/04 Typical Receptor Axon Hillock 11/3/04 (fig. from Anderson, Intr. Neur. Nets) 9 11/3/04 (fig. from Peters, Palay & Webster) Dendrite & Dendritic Branches 11/3/04 (fig. from Peters, Palay & Webster) 10 Dendrite & Dendritic Spine 11 11/3/04 (fig. from Peters, Palay & Webster) 12 3 Part 3: Autonomous Agents 11/3/04 Neuropil axon terminal dend Myelinated Axon Making Synapse on Dendrite rite 11/3/04 (fig. from Peters, Palay & Webster) 13 Excitatory Synapse Between Axon Terminal and Dendritic Thorn Various Synapses 11/3/04 11/3/04 (fig. from Peters, Palay & Webster) 15 11/3/04 axon (fig. from Peters, Palay & Webster) 14 axon terminal dendritic thorn synapse (fig. from Peters, Palay & Webster) 16 4 Part 3: Autonomous Agents 11/3/04 Dendro-dendritic Synapses Electrotonic Synapse Type II (symmetric) Type I (asymmetric) 11/3/04 (fig. from Peters, Palay & Webster) 17 11/3/04 18 (fig. from Peters, Palay & Webster) 5B Typical Artificial Neuron connection weights Artificial Neural Networks inputs output (in particular, the Hopfield Network) threshold 11/3/04 19 11/3/04 20 5 Part 3: Autonomous Agents 11/3/04 Typical Artificial Neuron linear combination Equations activation function Net input: net input (local field) New neural state: n hi = w ij s j j=1 h = Ws si = ( hi ) s = (h) 11/3/04 21 11/3/04 What to do about h = 0? Hopfield Network • • • • • • There are several options: Symmetric weights: wij = wji No self-action: wii = 0 Zero threshold: = 0 Bipolar states: si {–1, +1} Discontinuous bipolar activation function: 1, ( h ) = sgn( h ) = +1, 11/3/04 22 (0) = +1 (0) = –1 (0) = –1 or +1 with equal probability hi = 0 no state change (si = si) • Not much difference, but be consistent • Last option is slightly preferable, since symmetric h<0 h>0 23 11/3/04 24 6