(1) Learning (2) The proximity principle and “evolutionary learning”Schedule of PresentationsOperations in relational networksOperation of the Network in terms of cortical columnsAdditional operations: LearningRequirements that must be assumed (implied by the Hebbian learning principle)Support for the abundance hypothesisLearning – The Basic ProcessSlide 9Slide 10Slide 11Slide 12Slide 13Slide 14LearningLearning: more termsLearning: Deductions from the basic processLearning in cortical networks: A Darwinian processLearning – Enhanced understandingColumns of different sizesHypercolums: Modules of maxicolumnsFunctional columns vis-à-vis minicolumns and maxicolumnsLearning in a system with columns of different sizesQuestion on cortical columnsFunctional columns in phonological recognition: A hypothesisFunctional columns in phonological recognition A hypothesisPhonological hypercolumns (a hypothesis)Adjacent maxicolumns in phonological cortex?Adjacent maxicolumns in phonological cortex?Revisit the diagram: Each node of the diagram represents a group of minicolumns – a supercolumnSlide 31Learning – The Basic Process: Refined viewSlide 33Learning – Refined viewSlide 35Learning: Refined viewLearning Refined viewA further enhancementLearning: Refined viewLearning: refined viewSlide 41Learning phonological distinctions: A hypothesisRemaining problems – lateral inhibitionHypothesis applied to conceptual categoriesLocating Functions: The Proximity PrincipleConsequences of the Proximity PrincipleLearning and the Proximity PrincipleTwo aspects of the proximity principleHow to Explain the Proximity Principle?Proximity: Economic necessityLimits on intercolumn connectivityLocations of available latent connectionsThe role of long-distance fibersTwo Factors in LocalizationGenetically determined proximitySome innate factors relating to localizationInnate factors relating to primary areasA Heterotypical (i.e., genetically built-in) structure Visual motion perceptionA Heterotypical structure: Auditory areas in a cat’s cortexInnate factors relating to localizationApplying the proximity principleImplications of the proximity principleDeriving location from proximity hypothesisSpeech Recognition in the Left HemisphereExercise: Location of Wernicke’s areaAnswer: Location of Wernicke’s areaMore exercisesExperience-based proximityInnate features that support languageSlide 70(1) Learning (2) The proximity principle and “evolutionary learning”Ling 411 – 17Schedule of PresentationsDelclosPlanum TempBanneyerCategoriesRuby TsoWritingBosleySynesthesiaRasmussen2nd languageBrownBilingualismTsaiTonesTu Apr 13 Th Apr 15 Tu Apr 20 Th Apr 22Operations in relational networksRelational networks are dynamicActivation moves along lines and through nodesLinks have varying strengths•A stronger link carries more activation, other things being equalAll nodes operate on two principles:•Integration Of incoming activation•BroadcastingTo other nodesREVIEWOperation of the Networkin terms of cortical columnsThe linguistic system operates as distributed processing of multiple individual components•“Nodes” in an abstract model•These nodes are implemented as cortical columnsColumnar Functions •Integration: A column is activated if it receives enough activation from other columns Can be activated to varying degreesCan keep activation alive for a period of time•Broadcasting: An activated column transmits activation to other columnsExitatory – contribution to higher levelInhibitory – dampens competition at same levelReviewAdditional operations: Learning Links get stronger when they are successfully used (Hebbian learning)•Learning consists of strengthening them •Hebb 1948Threshold adjustment•When a node is recruited its threshold increases•Otherwise, nodes would be too easily satisfiedRequirements that must be assumed(implied by the Hebbian learning principle)Links get stronger when they are successfully used (Hebbian learning)•Learning consists of strengthening them Prerequisites: •Initially, connection strengths are very weakTerm: Latent Links•They must be accompanied by nodesTerm: Latent Nodes•Latent nodes and latent connections must be available for learning anything learnableThe Abundance Hypothesis•Abundant latent links •Abundant latent nodesSupport for the abundance hypothesisAbundance is a property of biological systems generally•Cf.: Acorns falling from an oak tree•Cf.: A sea tortoise lays thousands of eggsOnly a few will produce viable offspring•Cf. Edelman: “silent synapses” The great preponderance of cortical synapses are “silent” (i.e., latent)•Electrical activity sent from a cell body to its axon travels to thousands of axon branches, even though only one or a few of them may lead to downstream activationLearning – The Basic ProcessLatent nodesLatent linksDedicated nodes and linksLatent nodesLet these links get activatedLearning – The Basic ProcessLearning – The Basic ProcessLatent nodesThen these nodes will get activatedLearning – The Basic ProcessThat will activate these linksLearning – The Basic ProcessThis node gets enough activation to satisfy its thresholdLearning – The Basic ProcessThese links now get strengthened and the node’s threshold gets raised ABThis node is therefore recruitedLearning – The Basic ProcessThis node is now dedicated to function ABABABLearningNext time it gets activated it will send activation on these links to next levelABABLearning: more terms Child nodes Potential Actual Parent nodesABABLearning: Deductions from the basic processLearning is generally bottom-up. The knowledge structure as learned by the cognitive network is hierarchical — has multiple layersHierarchy and proximity:•Logically adjacent levels in a hierarchy can be expected to be locally adjacent Excitatory connections are predominantly from one layer of a hierarchy to the nextHigher levels will tend to have larger numbers of nodes than lower levelsLearning in cortical networks:A Darwinian processA trial-and-error process:•Thousands of possibilities availableThe abundance hypothesis•Strengthen those few that succeed“Neural Darwinism” (Edelman)The abundance hypothesis•Needed to allow flexibility of learning•Abundant latent nodesMust be present throughout cortex•Abundant latent connections of
View Full Document