Snapshots of AI methods and applications Agnar Aamodt and Lester Solbakken (with thanks to Keith Downing) Institutt for datateknikk og informasjonsvitenskap Seksjon for Intelligente Systemer NTNU
Hva er Kunstig intelligens 1 AI = Things that make you go WOW! or?? Well, somewhat more sober although more dull: AI enables systems to perform tasks in ways that woud be called intelligent if done by humans. AI enables complex problem solving and interaction beyond what other (non-ai) methods do.
Example applications Software: Pro-aktive beslutningsstøttesystemer Automatisk data-analyse Lærende systemer, f.eks.: Anbefalingssystemer AI i spill Ansiktsgjenkjenning Naturlig språk Robotnavigering, syn, planlegging Adapterende GUI... Embedded systems Intelligente komponenter i totalsystemer (hardware + software) Annen hardware: Autonome roboter Online bildefortolking Samarbeid Planleggingssystemer Hjernesimulering Kognisjonsvitenskap Selvorganiserende systemer
Hva er Kunstig intelligens 2 INFORMATIKK STUDIET AV INTELLIGENTE SYSTEMER RELATERT TIL KOMPUTASJONELLE PROSESSER er delfelt av REALISERING AV DATASYSTEMER SOM KAN SIES Å OPPVISE INTELLIGENT ADFERD - DVS. ' SMARTERE ' SYSTEMER har vitenskapelig perspektiv er koblet via empirisk vitenskapelig metode KUNSTIG INTELLIGENS (AI) har teknologisk perspektiv MATEMATIKK FILOSOFI bygger bl.a. på har metoder har metoder SYMBOLORIENTERTE METODER (KUNNSKAPSBASERTE METODER) KOGNITIV PSYKOLOGI BIOLOGI SUBSYMBOLSKE METODER (BIO-INSPIRERTE METODER)
Core idea AI = Representation + Search The concept of search plays an important role in science and engineering Any problem whatsoever can be seen as a search for the right answer or at least a good answer This search can be viewed to take place in a problem space, which constrains the search through its representation
SØKING I TILSTANDSROM (PROBLEMROM) starttilstand mellomtilstander måltilstander traverserte søkeveier mislykkede noder aktive noder node der testing pågår Sentralt i enhver AI-metode er en eller flere søkestrategier for traversering av tilstandsrommet (søkerommet) fra en starttilstand til en egnet måltilstand.
KUNNSKAPSBASERTE (SYMBOLORIENTERTE) METODER - UTVIKLINGSTRENDER Heuristiske regler Regelbaserte systemer (f.eks.: MYCIN)
KUNNSKAPSBASERTE METODER - UTVIKLINGSTRENDER Kontroll-kunnskap Heuristiske regler Eksplisitt kontrollkunnskap (f.eks. NEOMYCIN) - kunnskap om typer regler for typer tilstander
KUNNSKAPSBASERTE METODER - UTVIKLINGSTRENDER Kontroll-kunnskap Heuristiske regler Dyp kunnskap Dypere modeller, lærebok-kunnskap - flere relasjoner, semantiske nett, rammer (f.eks. CASNET)
KUNNSKAPSBASERTE METODER - UTVIKLINGSTRENDER Kontroll-kunnskap Heuristiske regler Spesifikke case Dyp kunnskap Fra generell kunnskap til situasjons-spesifikke case (f.eks. CYRUS, PROTOS) - case-basert resonnering
The Case-Based Reasoning (CBR) Cycle (Aamodt&Plaza 1994)
KUNNSKAPSBASERTE METODER - UTVIKLINGSTRENDER Kontroll-kunnskap Heuristiske regler Spesifikke case Dyp kunnskap Integrerte systemer (f.eks. SOAR, CREEK, META-AQUA) - totalarkitekturer for intelligent problemløsning
Herb Simon Push
Hypen rundt A.I. in three to eight years we will have a machine with the general intelligence of an average human being. I mean a machine that will be able to read Shakespeare, grease a car, play office politics, tell a joke, and have a fight. At that point, the machine will start to educate itself with fantastic speed. In a few months it will be at genius level and a few months after that its powers will be incalculable. Marvin Minsky i 1970!
A computer chess success http://www.youtube.com/watch?v=njarxpyyofi
AI state of the art examples Google's Driverless Car: http://www.youtube.com/watch?v=4zofntkxmvq IBM Watson: http://www.youtube.com/watch?v=fc3irywr4c8
Push
Subsymbolic / Bio-inspired AI Methods
Emergence The signal feature of life is not the carbon-based substrate...(but)...that the local dynamics of a set of interacting entities (e.g. molecules, cells, etc.) supports an emergent set of global dynamical structures which stabilize themselves by setting the boundary conditions within which the local dynamics operates (Charles Taylor, biologist, UCLA)
Swarm Intelligence Follow Trail Find Food Make Trail
Termite Arch-Building (Stigmergy) Turtles, Termites and Traffic Jams: Explorations in Massively Parallel Microworlds (Resnick, 1994) pheremone
Columns to Arches Positive Feedback: Pheromone Concentration in middle gets higher and higher as more dirt balls are added.
Emergence examples http://www.red3d.com/cwr/boids/ http://www.youtube.com/watch?v=sqrtn1urczi
Ubiquity of Emergence
Emergence & Intelligence Emergence Spectrum How does intelligent behavior arise from the interactions of 100 billion neurons, without central control? How has the brain evolved?
Evolutionary Progressions along the Intelligence Spectrum Living organisms Computers Sense & Act: 10,000,000+ years. 15+ years Reason: 100,000+ years. 30+ years Calculate: 1,000+ years 50+ years Evolution of reasoning was tightly constrained and influenced by sensorimotor capabilities. Else extinction! GOFAI systems are often in their own little worlds, making unreasonable assumptions about independent sensorimotor apparatus. To achieve AI s scientific goal of understanding human intelligence, the road from sense-and-act to reasoning via simulated evolution may be the only way.
Cognitive Incrementalism Tacit assumption of SEAI research. Cognition (and hence common sense) is an extension of sensorimotor behavior. This is the idea that you do indeed get full-blown, human cognition by gradually adding bells and whistles to basic (embodied, embedded) strategies of relating to the present at hand Mindware, pg. 135 (Andy Clark, 2001). I am, therefore I think. Brooks, Steels, Pfeifer, Scheier, Beer, Thelens, Nolfi, Floreano
Darwinian Evolution Physiological, Behavioral Phenotypes Natural Selection Ptypes Morphogenesis Reproduction Sex Genotypes Recombination & Mutation Gtypes Genetic
Evolutionary Algorithms Parameters, Code, Neural Nets, Rules Semantic Performance Test P,C,N,R Translate R &M Generate Bit Strings Syntactic Recombination & Mutation Bits
Artificial Neural Networks
World Model Behav Gen Body GOFAI World Brain Connectionism World Model Behav Gen Body World SEAI The world is its own best model Rodney Brooks World Model Behav Gen Body World Brain
GOFAI -vs- SEAI Brittle Nerds -vs- Well-Rounded Insects Knowledge Selection Pressure GOFAI SEAI Knowledge Cramming -vs- Adaptive Systems
Integrated methods: Cognitive architectures (eks: LIDA)
IDIs Gruppe for Intelligente Systemer - Organisering i 3 hovedområder Kunnskapsbaserte systemer Case-basert resonnering Kunnskapsmodellering Intelligente agenter Adaptive brukergrensesnitt Usikkerhetsbehandling/grafiske modeller Bildebehandling/kunstig syn Maskinlæring/datamining. Selvorganiserende systemer Evolusjonære metoder Konneksjonisme Nevrovitenskap Kunstig liv Maskinlæring Språkteknologi Naturlig språklig fortåelse Beregnbar logikk Tekstmining BusTuc 31 ansatte: 11 heltidsstillinger 4 Deltid 3 Forskere 13 PhD studenter 30 40 MSc studenter per år
NTNU NTNU
Intelligent Systems Group - Scientific Staff
Intelligent Systems Group PhD Candidates
Fagplan - DIS Basisfag Høst Logikk og resonnerende systemer (AI-1) Videregående fag, Høst Statistisk bildeanalyse og læring Kunnskapsrepresentasjon Maskinlæring og case-basert resonnering Kunstig intelligens programmering Vår Metoder i kunstig intelligens (AI-2) Bildeteknikk Vår Datasyn Sub-symbolske AI-metoder Naturlig språk grensesnitt Distribuert AI og intelligente agenter Intelligente brukergrensesnitt Ca. 10-12 teoriemner, fordypning (3.75 Bt) 3-4 dr.gradsemner
IDI, AI-gruppa: Forskningsområder Kunnskapsbaserte systemer Kunnskapsmodellering, maskinlæring, case-basert resonnering, usikkerhetsbehandling, intelligente agenter, adaptive brukergrensesnitt, bildetolkning, kunstig syn, rådgiviningssystemer. Selvorganiserende systemer Evolusjonære metoder, konneksjonisme, nevrovitenskap, kunstig liv, maskinlæring, intelligent hardware. Språkteknologi Naturlig språklig fortåelse, beregnbar logikk, grensesnitt mot databaser, tekstmining, maskinell oversettelse.
A master thesis in AI at IDI a few examples
Eksempler på master-oppgaver Improved game AI through case-based and statistical reasoning
Eksempler på master-oppgaver
Eksempler på master-oppgaver
Eksempler på master-oppgaver Bilde- og/eller Video-analyse (Her: Segmentere bilder av karbonfiberarmert epoxy)
Eksempler på master-oppgaver Bilde- og/eller Video-analyse (Her: Segmentere bilder av fisk i Mauritius)
Eksempler på master-oppgaver Robots (pictured) that interact with either a real or simulated other robot. Within our PUCKER system, researchers and students can easily test their AI control strategies on this type of robot (epucks).
Eksempler på master-oppgaver Intelligent Hardware Today s hardware technologies, especially Field programmable Gate Arrays (FPGAs), provide many possibilities for the creation of intelligent Hardware - that is AI techniques embedded in hardware. Such embedding may be for the purpose of speed-up of a given AI technique for perhaps real-time application requirements or for the purpose of creating hardware circuits, applying bioinspired techniques as the design technique. The latter is known as the field of Evolvable Hardware and includes applications in today s technology and approaches to achieve computation in tomorrow s technology. Application areas range from Vision, art to electronic circuits.
Eksempler på master-oppgaver Språkteknologi - maskinoversetting
Eksempler på master-oppgaver
Eksempler på master-oppgaver Textual CBR. Discovery of causal relations in incident reports An incident report (i.e., a 'textual case') describes how a problem unfolds. That is, the story starts with less important 'symptoms'/evidence which, in turn, triggers/causes more serious ones, and this chain of evidence ends up with an undesired, anomalous event. It is important to identify the events when they are small, and discover the causal mechanisms underlying the chain of events. Use of eye-tracking in the selection of important features in a text and determining how important they are - the latter is called 'weighting. This in cooperation with people at Dragvoll.
Eksempler på master-oppgaver Computer Assisted Assessment and Treatment of Pain Probabilistic networks, Rules, CBR, meta-level reasoning
Eksempler på master-oppgaver Data mining and Decision support in Fish Farming
Eksempler på master-oppgaver Evolving Populations of Social Insects to Perform Annular Sorting Vegard Hartmann Acting Sensing P = Pick up F = Forward L = Left D = Deposit B = Backward R = Right Andre Hei Vik
Eksempler på master-oppgaver Fitness Evaluation
Eksempler på master-oppgaver Three-object annular structure
Eksempler på master-oppgaver Reducing unwanted down.me in oil drilling One day of unwanted downtime on this rig means increased cost of 1,6 MNOK for the ongoing drilling operation. Providing the relevant experience and getting the right information precisely when needed will reduce unwanted operational downtime. The result is a more reliable drilling process, reduced drilling costs, and increased productivity.
Eksempler på master-oppgaver Improved decision support through experience capture and reuse pa9ern analysis case based reasoning
IDIs AI-gruppe har deltatt i etablering av tre spin-off selskaper: - LingIT AS - naturlig språk tolkning og dialogsystemer - Trollhetta AS - bildeanalyse og beslutningsstøtte - Verdande Technology AS - erfarings-lagring og aktiv gjenbruk, primært innen oljeboring
Eksempler på master-oppgaver
AI s 10 to Watch (IEEE Jour. Int. Syst. 2008) AI and natural language AI for autonomous robotic cars Image statistics in computational photography Lightning up the semantic web Learning representations for visual scenes Multimodal perception of human nonverbal behaviours AI and ontology technologies Human computation Combining logic and probability Logics and statistics for complex networks
AI - covers a lot of methods and application areas - is interesting, useful, and fun So, learn your - basic AI formalisms, such as - logics - representations - state-space search methods Link to videos shown (and more!): http://videolectures.net/aaai07/ http://videolectures.net/aaai08/ http://videolectures.net/ijcai09_video_competition/ A useful link to all of AI: http://www.aaai.org/aitopics
Evolutionary Computation