29
Developmental Artificial Intelligence 27 March 2014 [email protected] r http:// www.oliviergeorgeon.com t oliviergeorgeon.com 1/29

Developmental Artificial Intelligence

  • Upload
    axelle

  • View
    54

  • Download
    0

Embed Size (px)

DESCRIPTION

Developmental Artificial Intelligence. 27 March 2014 [email protected] http:// www.oliviergeorgeon.com. t. Outline. Organization in time and space. Formalism for Spatio -temporal coupling. Cognitive architecture . Demonstrations Exercise - PowerPoint PPT Presentation

Citation preview

Prsentation PowerPoint

Developmental Artificial Intelligence27 March [email protected]://www.oliviergeorgeon.com

toliviergeorgeon.com1/291OutlineOrganization in time and space.Formalism for Spatio-temporal coupling.Cognitive architecture.DemonstrationsExerciseImplement your self-programming agent (follow up).

oliviergeorgeon.com2/292Reminder of theoretical ideasThe objective is to learn (discover, organize and exploit) regularities of interaction in time and space to satisfy innate criteria (survival, curiosity, etc.).

To autonomously construct an ontology of reality. From experience of interactionDetect and respond to increasingly sophisticated affordances(self-programming).oliviergeorgeon.com

3/293ProposeTrace-Based ReasoningTimeHierarchical abstractionMatchingSelectTry to EnactMore abstractionoliviergeorgeon.com4/29Active self-motivated hierarchical abstraction of sequence.

4Example 2

oliviergeorgeon.com5/295Examples of learned behaviors

oliviergeorgeon.com6/296

Example 2oliviergeorgeon.com7/297Spatio-sequential regularity learningoliviergeorgeon.com8/298Concept of affordanceProperty of an object or an environment that allows an individual to perform an action.To afford an action. objects push us into doing (Heinz Werner)Affordances are properties of the coupling between the agent and the environment.We know the world in terms of possibilities of interaction. oliviergeorgeon.com9/29FormalismeAgentEnvironmentObservationActionTraditionnal formalization O AAgentEnvironmentExprienceRsultatLearning by experiencing R Eoliviergeorgeon.com10/29AgentEnvironmentIntendedInteractionEnactedInteraction I IRadical Interactionism:I = E RX: Interactions compositesHow to learn the existence of possibly persistent entities in the environment?How to adapt to different categories of entities?

Spatial Radical InteractionismAgentEnvironmentIntended Interaction i IEnacted Interaction e Ioliviergeorgeon.com11/29Spatial position Spatial transformation represents the vestibular system. Can be implemented through an accelerometer. represents position information (eye convergence, interaural phase difference, etc.)

Spatial exemple 1oliviergeorgeon.com

AgentEnvironmentEnacted interaction:..

Spatial position = (1,0)Spatial transformation = (0,0)

Intendedinteraction12/29AgentEnvironment

Environment known at time tdecd Xdicd Xdep1 ip1 ipj Iepj IDecisional mechanism Recursivity problemHow to maintain recursivity ?oliviergeorgeon.comSpatial position Spatial transformation 13/29Spatial Exemple 1

oliviergeorgeon.com14/29Enactive Cognitive Architecture

The agent programs itself through experience of interaction.The architecture does not program itself.(Kantian space, e.g., Buzsaki 2013, Space, time, and memory)oliviergeorgeon.com15/29Inspiration from neurosciencesoliviergeorgeon.com

16/29More inspiration from neurosciences?oliviergeorgeon.com

Cotterill R. (2001). Progress in Neurobiology.17/29ECA agent exemple 2

oliviergeorgeon.com18/29From drives to goals

AffordedSimulatedoliviergeorgeon.com19/29Exercise 3oliviergeorgeon.com20/2920ExerciseTwo possible experiences E = {e1,e2}Two possible results R = {r1,r2} Four possible interactions E x R = {i11, i12, i21, i22} Two environmentsenvironment0: e1 -> r1 , e2 -> r2 (i12 et i21 are never enacted)Environment1: e1 -> r2 , e2 -> r1 (i11 et i22 are never enacted)Environment2 : e1 -> r2 , e2 -> r2Environment3Motivational systems:motivation0: v(i12) = v(i22) = 1, v(i11) = v(i21) = -1motivation1: etc.Implement un agent that learn to enact positive interactions without knowing its motivatins and its environnement a priori.Write a rapport of behavioral analysis based on activity traces.

oliviergeorgeon.com21/29Implmentationpublic static Experience e1 = new experience(); Experience e2 = new experience();public static Result r1 = new result(); Result r2 = new result();public static Interaction i11 = new Interaction(e1,r1, 1); etc.Public static void main() Agent agent = new Agent3(); // Agent1(); Agent2(); Environnement env = new Env3(); // Env1(); // Env2();for(int i=0 ; i < 10 ; i++)e = agent.chooseExperience(r);r = env.giveResult(e); System.out.println(e, r, value);Class AgentClass Agent3public Experience chooseExperience(Result r)Class EnvironnementClass Env3public Result giveResult(experience e)Class ExperienceClass ResultClass Interaction(experience, result, value)public int getValue()

oliviergeorgeon.com22/2922Analyse de traces dactivit.oliviergeorgeon.come1,r2,-1e2,r1,1e2,r1,1e2,r1,1e2,r1,1e2,r1,1e2,r1,1e2,r1,1e2,r1,1e2,r1,1e1,r1,1e1,r1,1e1,r1,1e1,r1,1e1,r1,1e1,r1,1e1,r1,1e1,r1,1e1,r1,1e1,r1,1Motivation1, Environnement 0.Motivation1, Environnement 1.Motivation1, Environnement 2.e1,r2,1 learn e1r1-e1r2,0e1,r1,-1 learn e1r2-e1r1,0e1,r1,-1 learn e1r1-e1r1,-2e2,r2,1 learn e1r1-e2r2,0e2,r1,-1 learn e2r2-e2r1,0e2,r1,-1 learn e2r1-e2r1,-2e1,r2,1 learn e2r1-e1r2,0e2,r2,1 learn e1r2-e2r2,2e1,r2,1 learn e2r2-e1r2,2e2,r2,123/2923Environnement 3Behaves like Environnement0 during the first 10 cycles, then like environnement1. ImplmentationIf (step < 10)If (experiment = e1) then result = r1If (experiment = e2) then result = r2ElseIf (experiment = e1) then result = r2If (experiment = e2) then result = r1Step++

oliviergeorgeon.com24/2924TempsActivatedi11Proposei11it-3it-2it-4it-1it = i11i11PRESENTFUTURPASTlearnAGENTit(it-1,it)Activateit-1oliviergeorgeon.comPrinciple of Agent 3(i11,i12)i12e1ChooseExecute(i11i11)25/2925Implementation of Agent 2At the end of time step tRecord or reinforce ic = it-1, it, w = pre(ic), post(ic), weight If ic already belongs to the set of existing interactions ItWeight ++.At the beginning of time step tConstruct the list of activated composite interactions AtAt= { i It | pre(i) = it-1 }For each activated composite interaction in AtCreate a proposition for post(ic).experience with proclivity ic.weight * post(ic).valenceFor each experience, sum up the proclivity of all its propositions.Choose the experience that has the highest total proclivity.

oliviergeorgeon.com26/2926Class InteractionClass Interaction//attributes: Experience experience;Result result;int value;String label;Interaction preInteraction;Interaction postInteraction;int weight;oliviergeorgeon.comPostInteractoinComposite Interaction weightpreInteraction27/2927Decision mechanismoliviergeorgeon.comList propositions = new ArrayList(); for (Interaction activatedInteraction : getActivatedInteractions()){Proposition proposition = new Proposition(activatedInteraction.getPostInteraction().getExperience(), activatedInteraction.getWeight() * activatedInteraction.getPostInteraction().getValence());int index = propositions.indexOf(proposition);if (index < 0)propositions.add(proposition);elsepropositions.get(index).addProclivity(activatedInteraction.getWeight() * activatedInteraction.getPostInteraction().getValence());}Collections.sort(propositions);If (propositions.size()> 0)proposedExperience = propositions.get(0).getExperience();28/2928Class PropositionClass Proposition implements Comparable// attributs:Experience experienceInt proclivity// constructorProposition(Experience experience, int Proclivity)// methodsint compareTo(Proposition proposition)return new Integer(proposition.getProclivity()).compareTo(proclivity);boolean equals(Object otherProposition)return ((Proposition)otherProposition).getExperience() == this.experience;void addProclivity(int proclivity)this.proclivity += proclivity;oliviergeorgeon.com29/2929