Download 11/10/16 Memory Part 2 Reinforcement learning (12.2) • Involves a

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Cognitive neuroscience of music wikipedia , lookup

Neuroeconomics wikipedia , lookup

Donald O. Hebb wikipedia , lookup

Neurotransmitter wikipedia , lookup

Endocannabinoid system wikipedia , lookup

Aging brain wikipedia , lookup

Development of the nervous system wikipedia , lookup

Stimulus (physiology) wikipedia , lookup

Nonsynaptic plasticity wikipedia , lookup

Environmental enrichment wikipedia , lookup

Misattribution of memory wikipedia , lookup

Socioeconomic status and memory wikipedia , lookup

Eyewitness memory (child testimony) wikipedia , lookup

Emotion and memory wikipedia , lookup

Neuropsychopharmacology wikipedia , lookup

Eyeblink conditioning wikipedia , lookup

Prenatal memory wikipedia , lookup

Molecular neuroscience wikipedia , lookup

State-dependent memory wikipedia , lookup

Childhood memory wikipedia , lookup

Holonomic brain theory wikipedia , lookup

Limbic system wikipedia , lookup

Synaptogenesis wikipedia , lookup

Long-term potentiation wikipedia , lookup

NMDA receptor wikipedia , lookup

Long-term depression wikipedia , lookup

Memory consolidation wikipedia , lookup

Activity-dependent plasticity wikipedia , lookup

Chemical synapse wikipedia , lookup

Synaptic gating wikipedia , lookup

Clinical neurochemistry wikipedia , lookup

Transcript
11/10/16
Memory Part 2
Reinforcement learning (12.2)
•
•
•
•
•
Involves a reinforcing or pushing “signals” – context dependent
Reinforcement signal is information poor but motivation rich
Involves “basic” brain structures that do not “process” information
o E.g. drug addiction: deficit in the control of the reinforcing value
Reinforcement: the ventral tegmental area
o Midbrain → basal forebrain (dopamine: VTA)
o Self-stimulation studies: rats will work to exhaustion
o Self-stimulation is a form of instrumental conditioning (no perceived
reward)
Meso-limbic system
o VTA: → amygdala, nucleus accumbens, and hippocampus
o Nucleus accumbens: neural substrate of drug addiction?
Dopamine in nucleus accumbens: stimulations (12.19)
•
•
Both stimulation of VTA and natural rewards (sugar, food, sex…) increase
dopamine in nucleus accumbens
Sex chamber
o Hedonic value of dopamine: natural stimulation (makes you feel good)
o Dopamine also released in anticipation rewards
Reinforcement learning
•
•
Meso-cortical system
o VTA → primary/secondary association cortex, hippocampus, and frontal
cortex
o Frontal cortex: working memory, reinforcement of current plans and
strategies
So what controls the VTA?
o Lateral hypothalamus, amygdala, and prefrontal cortex → VTA
o LH: hunger
o Amygdala: emotional stimuli
o Prefrontal cortex: planning and strategies
Motor learning (12.3)
•
Motor learning: learning within the motor system (little perceptual input required)
– e.g. ride a bike, play tennis, tie a shoe
Relational learning
•
•
Learning relationships between stimuli or previous phenomes
o Memory without meaning ~30 minutes (Ebbinghaus)
Use of contextual information (schema) to store or retrieve new memories
•
o Schemas: set of interconnected concepts/memories
o Schemas are continuously updated
Memories can be “manipulated” by manipulating (consciously or not) their
relationship to other memories – false memories. Eye witness testimony.
Implanted memories
Memory: neural mechanisms
•
•
Neural substrate of memory: what is memory?
o Neural activity done cannot account for memories that last more than a
few seconds
o Synapses are the primary encoding mechanism
o Memory items: a pattern of synapses (between excitatory cells)
o Different kinds of memory, different kinds of learning → different brain
structures
The hippocampus: short term memory
o Basic anatomy: the tri-synaptic circuit
o Long-term potentiation
o Long-term depression
o Rat hippocampus (12.4)
§ Entorhinal cortex (inputs) → dentate gyrus → CA3 → CA1
Long-term potentiation (12.5)
•
•
•
•
Electrical test stimulus = “recall”
LTP induction = “learning”
Memory = “size of the EPSP”
LTP: synapse specific (12.7)
o The Hebb learning rule: if presynaptic and postsynaptic sites are
repeatedly active at the same time the synapse is strengthened
Hippocampus (12.6)
•
•
•
Memory is associative – so is long term potentiation
If two synapses, weak and strong, are stimulated at the same time (i.e.
association), the weak synapse becomes stronger
Associative LTP (12.9)
o Many (almost simultaneous) inputs can be associated: “dendritic spike”
o Memories are represented by patterns of synapses
Memory: LTP and NMDA receptors
•
•
•
•
•
If classical conditioning is implemented by neurons there should be a neuronal
mechanism that mimics it
Fact 1: synaptic modification occurs only if there is Ca2+ entry in a synapse
(second messenger including CaM-KII)
Fact 2: LTP occurs only when Ca2+ enters through NMDA receptors
o NMDA receptors are permeable to Na+ and Ca2+ but are normally
blocked by Mg2+
o NMDA receptors are unblocked only if there is postsynaptic
depolarization:
§ Action potential
§ Depolarization through an electrode → i.e. NMDA unblocked only if
there is postsynaptic activity
o Unblocked NMDA receptors let Ca2+ through only if there is glutamate
bound → i.e. Ca2+ enters only if there is presynaptic activity
So… Ca2+ enters through pre and post synaptic activity (Hebb rule again)
NMDA, depolarization, and calcium (12.8)
LTD
•
•
•
•
•
Long term depression: decrease in synaptic strength
Induction (i.e. learning)
o High frequency stimulation → LTP
o Low frequency stimulation → LTD
Result in a decrease in AMPA receptors
Also associative and selective
Goal: prevent overly strong synapses