* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download Circuits, Circuits
Cortical cooling wikipedia , lookup
Dual consciousness wikipedia , lookup
Emotional lateralization wikipedia , lookup
Cognitive neuroscience wikipedia , lookup
Electrophysiology wikipedia , lookup
Recurrent neural network wikipedia , lookup
Environmental enrichment wikipedia , lookup
Aging brain wikipedia , lookup
Haemodynamic response wikipedia , lookup
Eyeblink conditioning wikipedia , lookup
Neural oscillation wikipedia , lookup
Convolutional neural network wikipedia , lookup
Neural modeling fields wikipedia , lookup
Neuroplasticity wikipedia , lookup
Neural engineering wikipedia , lookup
Neuroesthetics wikipedia , lookup
Multielectrode array wikipedia , lookup
Cognitive neuroscience of music wikipedia , lookup
Activity-dependent plasticity wikipedia , lookup
Types of artificial neural networks wikipedia , lookup
Neuromuscular junction wikipedia , lookup
Endocannabinoid system wikipedia , lookup
Synaptogenesis wikipedia , lookup
Neuroeconomics wikipedia , lookup
Sensory cue wikipedia , lookup
Mirror neuron wikipedia , lookup
Caridoid escape reaction wikipedia , lookup
Neuroanatomy wikipedia , lookup
Central pattern generator wikipedia , lookup
Nonsynaptic plasticity wikipedia , lookup
Premovement neuronal activity wikipedia , lookup
Optogenetics wikipedia , lookup
Clinical neurochemistry wikipedia , lookup
Neural correlates of consciousness wikipedia , lookup
Neurotransmitter wikipedia , lookup
Molecular neuroscience wikipedia , lookup
Chemical synapse wikipedia , lookup
Holonomic brain theory wikipedia , lookup
Time perception wikipedia , lookup
Single-unit recording wikipedia , lookup
Neural coding wikipedia , lookup
Metastability in the brain wikipedia , lookup
Development of the nervous system wikipedia , lookup
Channelrhodopsin wikipedia , lookup
Neuropsychopharmacology wikipedia , lookup
Stimulus (physiology) wikipedia , lookup
Feature detection (nervous system) wikipedia , lookup
Synaptic gating wikipedia , lookup
Circus, Circuits Interesting Neural Networks: Some actually occur in brains; some are hypotheses Owl Audition • The Barn Owl uses delay lines and coincidence detectors (neurons that only fire when both pre-synaptic axons are simultaneously depolarized) to locate objects in horizontal and vertical plane. Far Right From Left Ear A B C D E From Right Ear Far Left Center-Surround Cells “ON center OFF surround” cell S S S C Visual Field S S C C C S S S S C Center S CS Brain Ganglia CS firing pattern Strong Retina Medium S Surround Inhibit Weak On-Center -vs- Off-Center Retinal Ganglion Cells • The primary visual receptors (rods & cones) actually turn OFF when hit by photons (light) and are ON when they detect dark spots (Hubel, Eye, Brain and Vision, 1988, pg. 54) On-Center (Off-Surround) Off-Center (On-Surround) Bipolar Cells Light Light S S S Receptors S C C S C C S S Receptors S S S Excite Inhibit Retinal Ganglion These are non-intersecting pathways but are drawn together to illustrate their similarities. Line Detectors Retinal Ganglia Visual Cortex On-Centers 45o Line Off-Centers To higher levels of the visual cortex Motion Detectors Riechard Detector (1961) - based on the fly’s visual system • Works best when delay = t2 - t1 = t3 - t2 • = normal (non-delayed) transmission time 450 bar moving left to right t3 t1 t2 t3+ Delay Delay t3+2 Delay t2+ Coincidence detectors => only fire when all inputs are ON simultaneously. Lateral Inhibition Lines Output Excite Inhibit 1 2 3 4 5 Input Neuron • Neurons that stimulate themselves and inhibit their near neighbors function as filters Lateral Inhibition in Visual Pathways • Grossberg, S. (2003) in The Handbook of Brain Theory and Neural Networks, pp. 594-600. V1(2/3) V2(2/3) V1(4) V2(4) V1(6) V2(6) LGN Retinal ON Cell • 6 - 4 - 2/3 pathway/loop is self-excitatory • Similar lateral inhib topology in V1 & V2 Excite Inhibit Interneuron Central Pattern Generators (CPGs) Neural circuits for generating simple, repeated patterns of activity. E.g. gait patterns in N-legged animals. Ian Steward (1998). Life’s other secret. Ch.9 Overhead view of horse, goat, dog?? Standard Notation: Fractions = Phase diffs t2 t4 1/4 3/4 t1 t3 0 1/2 Walking gait: First move left rear leg, then left front, then right rear, then right front. Generic Gait Generator • • • • • Each animal species can perform many different gaits. Do we need a different wiring pattern for each gait? No! (Golubitsky, Stewart, Collins, Buono (1997)) Goal: A single circuit with adjustable delay times. Solution: For an N-legged animal, 2 cross-linked N-neuron loops. AL2 AR2 Inter-loop delay AL1 AR1 Intra-loop delay Left Front Right Front Left Rear Right Rear By adjusting these TWO delay times, we can generate all standard gait patterns for N-legged animals!! Walking Jumping 3/4 1/4 3/4 3/4 1/2 0 1/2 1/2 1/4 3/4 1/4 1/4 0 1/2 0 0 1/4 1/2 1/4 0 Pacing 0 Trotting 1/2 1/2 0 1/2 0 1/2 0 0 1/2 1/2 0 0 1/2 0 1/2 0 1/2 1/2 1/2 Brain Clocks • Wright, Karen,”Times of our Lives”, Scientific American, Sept. 2002 • In the cerebral cortex, a collection of neurons with different firing patterns enables us to record and reuse specific time intervals. A Time Signatures B C D t1 t2 t3 t4 A B C D t1 0 1 0 1 t2 1 1 1 0 t3 1 1 0 0 t4 0 0 1 1 Timing Circuit Neural Oscillators from 10-40 Hz Cerebral Cortex A B C D 1. A start signal (e.g. Dance instructor says ”Begin”): STN excites SNr, which then inhibits all cortical oscillators, so they essentially RESET to off. 2. Oscillators then resume their normal diverse firing patterns, from same init state. 3. A stop signal (e.g. Dance instructor…): SNc releases dopamine into striatum, causing striatal cells to record the current time signature via Hebbian Learning STN S SNr Striatum Excite Inhibit SNc Dopamine Signal => Learn! Learning a Time Signature High B Low Low C High STOP!! LEARN!! • • • C D D A A • B S S Non-associate Learning: Strengthen pre-synaptic axon since: a) it fired/depolarized, and b) significant event (STOP) signalled. After learning, S will only fire when B & D are active (i.e. after a time interval of duration = t1). Details are unclear as to whether A & C develop inhibitory links to S. In future (e.g. when repeating the dance), the instructor still says ”Go”, which again resets the cortical oscillators, but now the brain generates its own ”STOP” signal in the striatum, when S fires => student has learned t1! Given enough diverse oscillators, student can learn ANY interval. Cricket Phonotaxis • Webb, B. (2001). Biorobotics: Methods & Applications, Ch. 1. • Female Crickets only respond to songs with particular carrier frequencies and syllable durations. Left Ear Drum • Syllable Duration • Carrying Frequency = 1/Inter-syllable period Right Ear Drum Bug Off! Preferred Carrier Frequency Distance between the two ear-drums is the critical determinant. If it’s ONE QUARTER the song’s inter-syllable wavelength, then the eardrums vibrate most strongly. Here P = period of the sound wave. Eardrums R L Time T Peak • Time T+P/2 Trough • • • • From T to T+P/4, the peak travels across the body and meets the right eardrum, causing it to vibrate, thus generating a new peak. From T+P/4 to T+P/2, the new peak travels exactly 1/4 wavelength = ear-to-ear distance. At time T+P/2, the left ear has a) a trough on the outside, and b) a peak on the inside. That’s a max pressure difference => the eardrum is maximally stimulated. The cricket is happy!! Preferred Syllable Duration • Appears to be determined in the brain, but details only partially known. • Biorobotics researchers (Webb et. al.) provide minimal ANNs that are sufficient explanations. Turn Right Turn Left Motor Neurons MNR MNL Auditory Neurons ANR ANL Right Ear Left Ear • Each auditory neuron stimulates the corresponding motor neuron and inhibits the opposite motor neuron. • Each of the 4 neurons has a very detailed (but standard) model: leaky integrate-and-fire • AN => MN synapses are temporarily depressed after the AN fires Leaky Integrate-and-Fire Neural Models Leak Integrate tmdVi/dt = b(EL - Vi) + awijzj zj = (1 + eVi)-1 {Standard sigmoidal transfer function} Vi = voltage inside the neuron EL = voltage outside the neuron (standard value: -55mV) zj = firing rate of neuron j wij = synaptic weight from neuron j to neuron i. a: excitation factor, b: leakage factor, tm = time scaling factor z1*wi1 z2*wi2 zi Vi Leak z3*wi3 EL AP = Voltage Spike • Although the voltage of a neuron changes constantly, only large abrupt changes (action potentials) can be transmitted to other neurons. Overshoot +40 mV K+ gates open. K+ leaves cell. Na+ gates still open Na+ gates close. K+ gates still open. 0 mV Na+ gates open. Na+ enters cell. Rising Phase Falling Phase K+ gates close. -65 mV Resting Potential Undershoot Habituation When a neuron fires weakly, but frequently, its axonal synapses weaken. After a little rest, the synapse returns to normal strength. tmdwij/dt = c(wij(*) - wij) - S(zj) wij(*): base value for wij S(zj) = stimulus function; lower zj => higher S wij S t Vj zj zj wij t Vi Preferred Syllable Duration • • Assume a stimulus on the left side of the cricket. High frequency (short wavelength) sound has a quickly-decaying amplitude with distance, so the left ear gets a stronger signal than the right. Syllable Incoming sound • Neuron ANL integrates the inputs from the left ear drum and fires groups of pulses with durations = syllable durations. ANL Response • This inhibits motor neuron MNR but stimulates MNL, which integrates the inputs from ANL and eventually begins to fire. However, it integrates more slowly than ANL and therefore fires less frequently. MNL Response • The cricket turns left. It is attracted to the song. Null Poeng • Stimulus again from left side, but now the syllables are very short and frequent.. Syllable Incoming sound • Neuron ANL integrates the inputs from the left ear drum and fires constantly, with very few significant gaps. ANL Response • This inhibits motor neuron MNR and stimulates MNL. • But, now the ANL-MNL synapse habituates due to the constant firing of ANL (and hence no break in which to regain strength). • So the signals that ANL sends to MNL are WEAK, and MNL never integrates enough charge to fire. MNL Response • The cricket is not interested. Another Loser • Stimulus again from left side, but now the syllables are very long, with a large gap between syllables.. Syllable Incoming sound • Neuron ANL integrates the inputs from the left ear drum and fires long sets of pulses with long gaps. ANL Response • This inhibits motor neuron MNR and stimulates MNL. • But, now the gap is too long: MNL almost fires during a syllable, but then a lot of voltage LEAKS out during the inter-syllable gap. • So although ANL’s signals are strong, MNL leaks too much and can never integrate enough charge to fire. MNL Response • This is cricket is very picky!