Learning and the Brain What happens in the brain when we learn? Learning and the Brain: Review 1. Learning is often synonymous with memory in cognitive neuroscience and, in biological terms, change (plasticity) in brain structure/connectivity/activity can be called (in biological terms) learning - including decreases! 2. All animals learn and useful learning can be achieved by quite small networks when they can change the efficiency of their connections appropriately – e.g. “Hebbian” synapses – respond to simultaneous pre/post excitation 3. Biological connections between neurons are called synapses - chemical synapses of particular interest in neurocognition. 4. Early and late processes of long-term potentiation (LTP) provide a biological basis for the Hebbian Synapse and, therefore, learning Working memory may be due to persistent neuronal firing 5. Memory (and working memory) are distributed in the brain 6. Evolutionary psychology = another perspective on categorizing different types of learning 7. Neuroplasticity: When we learn, it is not just connectivity (and function) that changes in the brain, but structure can change also. Pavlov’s dogs – classical conditioning of reflexive behaviors ……. "Pavlov's Dog". Nobelprize.org. Nobel Media AB 2014. Web. 6 Oct 2014. <http://www.nobelprize.org/educational/medicine/pavlov/index.html> Demonstration Animals can often appear to make “reasonable” decisions. Some of these may be based on classical conditioning…. © Nicolas P. Rougier / Wikimedia Commons / CC-BY-SA 3.0 The case of “Clever Hans” the counting horse By Karl Krall (Karl Krall, Denkende Tiere, Leipzig 1912, Tafel 2) [Public domain], via Wikimedia Commons But how about non-reflexive responses that require anticipation of behavioural consequence? Operant Conditioning = Instrumental conditioning - response to reward/punishment - anticipating behavioural consequence - requires more of a cognitive construction Also: How do rats know how to take the second best route when the best is closed off? How would neurons have to behave to produce such behaviour i.e. make educated guesses in new situations from a “cognitive map”? Networks of neurons – with the “right” connections – can produce complex behaviour ………… •Neurons generate & mediate electrical signals in a complex and interconnected manner. Image © Paul Howard-Jones 2014 Artificial parallel networks have human-like learning curves and can even make good guesses in novel situations. But how do networks learn those “right” connections? By en:User:Cburnett [GFDL (http://www.gnu.org/copyleft/fdl.html) or CC-BYSA-3.0 (http://creativecommons.org/licenses/bysa/3.0/)], via Wikimedia Commons The Synapse -a neural connection Neurotransmitter Synaptic vessicle By Synapse_Illustration2_tweaked.svg: Nrets derivative work: Looie496 (Synapse_Illustration2_tweaked.svg) [CCBY-SA-3.0 (http://creativecommons.org/licenses/bysa/3.0/) or GFDL (http://www.gnu.org/copyleft/fdl.html)], via Wikimedia Commons Neurotransmitter transporter Voltage gated Ca++ channel receptor Presynaptic terminal Synaptic cleft Dendrite 1. Action potential reaches the synapse. 2. Depolarizes synapse membrane -> channels open and calcium ions flow through. 3. Activation of calcium-sensitive proteins attached to vesicles (containing neurotransmitter) 4. Proteins change shape, vesicles fuse with presynaptic membrane and open 5. Some neurotransmitter escapes, some binds to receptor molecules in postsynaptic membrane. 6. Receptor molecule activates postsynaptic cell in some way (e.g. exciting, inhibiting) 7. Neurotransmitter molecules loosen and drift off 8. Neurotransmitter either reabsorbed in presynaptic cell or broken down. For the “right” connections: “neurons that fire together wire together” (anon.) Donald Hebb (1949): * When an action potential in the presynaptic neuron reaches the synapse, it causes release of neurotransmitter * This influences the postsynaptic neuron – and that may fire also * If the postsynaptic neuron was firing when the presynaptic neuron was releasing neurotransmitter, then the synaptic efficiency would be increased. NB synaptic efficiency = likelihood of being stimulated into further firing = strength of connection Artificial neural networks with this mechanism build impressive learning networks: •human-like learning curves •human-like errors •pattern recognition (inc. language) •best guesses and cognitive maps •graceful degradation Parallel distributed learning (PDP) •etc By en:User:Cburnett [GFDL (http://www.gnu.org/copyleft/fdl.html) or CC-BY-SA-3.0 (http://creativecommons.org/licenses/by-sa/3.0/)], via Wikimedia Commons (Artificial) Neural Networks But we still need a biological mechanism by which neurons change their synaptic efficiency when adjacent neurons increase their activity - i.e. we need a “Hebbian” synapse. Long-Term Potentiation (LTP) – an enduring increase in the size of the post-synaptic potential. Bliss and Lomo (1973) in Hippocampus (visuospatial memory) of rabbits – but now known to be common throughout the brain. Lomo (2003): “As for me, it was certainly not prior interest in possible memory mechanisms that led me to a discovery that was in some ways accidental and in other ways the result of an intuition that has often, I feel, brought me to look for or see phenomena that turn out to be new and interesting.” EPSP (Excitatory PostPynaptic Potential measured at REC) is low until electrical stimulation (tetanus at STIM). Green square shows EPSP immediately high (PTP – post-tetanic potentiation). The blue/grey squares show elevated EPSP measurements 3 to 60 minutes later (LTP or long-term potentiation). By Synaptidude at English Wikipedia [GFDL (www.gnu.org/copyleft/fdl.html) or CC-BY-SA3.0 (http://creativecommons.org/licenses/by-sa/3.0/)], via Wikimedia Commons LTP type is classified according to pre- and postsynaptic activity required: Hebbian LTP requires simultaneous pre- and postsynaptic depolarization (“fire together, wire together”) Non-Hebbian LTP does not require such simultaneous depolarization. LTP depends on a postsynaptic glutamate receptor: NMDA Normally: EPSP (excitatory postsynaptic potential) mediated by AMPA receptors (another glutamate receptor). NMDA receptors blocked by magnesium (Mg) ions. But: when sufficient incoming glutamate generates enough EPSP, it depolarises the postsynaptic membrane and allows Mg++ ions to be repelled from NMDA receptors -> Na+ and Ca++ ions let through. Coincidence detector (Pre + post synaptic activity -> EPSP up) Happens if: * post-synaptic activity for depolarised membrane (gate open) * AND neurotransmitter at synaptic cleft (something to go thro’ gate) But….early LTP only lasts a few hrs/wks …. Late LTP: In the late phase of LTP, calcium enters the cell and triggers Calcium modulated protein (calmodulin) -> chain of events-> the nucleus of the cell -> processes leading to protein synthesis and structural changes, i.e., the formation of new synapses and receptors Long Term Potentiation = Learning? * We know STM -> LTM transition needs protein synthesis to maintain synaptic enhancement. Suppression of protein synthesis using drugs suppresses learning. * mice treated with NMDA receptor blockers are unable to learn their way around a water maze. * Using rats conditioned to freeze to a neutral sound, synaptic potentials from the amygdala became enhanced only while the learned response continued, and this enhancement mediated by glutamate receptors & dependent upon NMDA receptors. demo Other learning mechanisms... •Also LTD (Depression) – e.g. reduction in firing rates (e.g. in the perirhinal cortex - involved in face recognition) with habituation. •New neurons can be born – at least in the hippocampus – we don’t know yet how neurogenesis relates to learning So where is memory in the brain? Psychologist attempted simple memory dichotomies: declarative/procedural, explicit/implicit , memory/habit, semantic/episodic. Biologically, declarative/non-declarative seem useful: MEMORY Declarative Facts Events Non-declarative Procedural Priming and (skills and perceptual learning habits) Simple classical conditioning Emotional responses Medial Temporal Lobe Striatum Neocortex Amygdala Nonassociative learning Skelatal responses Cerebellum Reflex pathways We walked across the Clifton Suspension Bridge! Location: parietal Working memory buffer Shape: Inferior temporal Colour: Temporal occipital However we categorise types of memory (e.g. semantic, conceptual, procedural, episodic, visual, spatial) a simple stimulus (e.g. a word) can cue all/many of these at once. “A memory” is distributed across many different brain regions – some more associated with particular aspects of a memory than others Working Memory ... Persisting neuronal activity of a prefrontal cortical neuron: Visual Fixation cue point Correct Responses cue delay period response Visual Fixation cue point Incorrect Responses cue delay period response Working memory also distributed, but particularly associated with dorsolateral prefrontal cortex (DLPFC) By Natalie M. Zahr, Ph.D., and Edith V. Sullivan, Ph.D. [Public domain], via Wikimedia Commons Evolutionary & Species perspectives Our learning mechanisms have evolved – this provides insights at a functional (backwards) design level – if we think of evolution as a “designer”: Issue (prehistoric environment) Specification (what’s needed to survive) Design (how the mind & brain work Product (mind & brain) Understanding the prehistoric issues and specification helps us understand the design that underlies our mind/brain? Or…..was there a specification that would have justified the design we think underlies our mind/brain product and how we think it is designed? - If not, we probably misunderstand the design/product! Learning: What/why was the reason (or “design specification”)? Conditioned/instrumental/“slow” learning: avoids spurious connections – associations of more frequently linked phenomena are more useful. Although…… Garcia effect: - an exceptional type of learning mechanism (or – preparedness for learning) to avoid poisons By Evan-Amos (Own work) [Public domain], via Wikimedia Commons Phobias: Innate species-specific knowledge to avoid potentially dangerous animals. (but mostly redundant in UK) – more preparedness for knowledge. http://www.flickr.com/photos/8373783@N07 / Snakecollector [CC-BY-2.0 (http://creativecommons.org/licenses/by/2.0) ], via Wikimedia Commons Neural basis of Garcia, and species-specific phobias are unknown Brain is ready for significant (even structural) change in response to changing environment Taxi drivers have larger posterior hippocampi Maguire et al. (1997) By ed g2s • talk (Own work) [GFDL (http://www.gnu.org/copyleft/fdl.html), CC-BY-SA-3.0 (http://creativecommons.org/licenses/bysa/3.0/) or CC-BY-SA-2.5-2.0-1.0 (http://creativecommons.org/licenses/bysa/2.5-2.0-1.0)], via Wikimedia Commons But such plasticity is not species-specific Avian food storers such as the scrub jay lose their expanded hippocampi when they are denied the freedom to store and retrieve food. By Aphelocoma_californica_in_Seattle.jpg: Minette Layne derivative work: Samsara (Aphelocoma_californica_in_Seattle.jpg) [CC-BY-SA2.0 (http://creativecommons.org/licenses/bysa/2.0)], via Wikimedia Commons Structural change over only 3 months Draganski et al. (2004) 1 By Backlit (Own work) [CC-BY-SA-3.0 (http://creativecommons.org/licenses /by-sa/3.0)], via Wikimedia Commons 2 3 1:before training 2:After 3 m practise 3: 3 m since practised Images © Bodran Draganski 2012 Learning and the Brain: Review 1. Learning is often synonymous with memory in cognitive neuroscience and, in biological terms, change (plasticity) in brain structure/connectivity/activity can be called (in biological terms) learning - including decreases! 2. All animals learn and useful learning can be achieved by quite small networks when they can change the efficiency of their connections appropriately – e.g. “Hebbian” synapses – respond to simultaneous pre/post excitation 3. Biological connections between neurons are called synapses - chemical synapses of particular interest in neurocognition. 4. Early and late processes of long-term potentiation (LTP) provide a biological basis for the Hebbian Synapse and, therefore, learning Working memory may be due to persistent neuronal firing 5. Memory (and working memory) are distributed in the brain 6. Evolutionary psychology = another perspective on categorizing different types of learning 7. Neuroplasticity: When we learn, it is not just connectivity (and function) that changes in the brain, but structure can change also.