Organic ComputingCS PhD Seminar Mar 3, 2003
Christoph von der Malsburg
Computer Science and Biology DepartmentsUniversity of Southern California
and
Institut für NeuroinformatikRuhr-Universität Bochum
Moore‘s Law
Chip complexity doubles every 18 months
Expectations
More Complex Functions
Flexibility, Robustness
Adaptivity, Evolability
Autonomy
User Friendliness
Situation Awareness
We expect our systems to become intelligent!
SW: Complexity
SW: Time
SW: FailureNIST study 02: yearly US losses due to SW failure: $ 60 Billion
Life as Model
• Living Cell: as complex as PC, but flexible, robust, autonomous, adaptive, evolvable, situation aware
• Organism: more complex than all existing software
• Human Brain: intelligent, conscious, creativeIt is the source of all algorithms!!Estimated computing power: 1015 OPS PC today 109 OPS, will equal brain in 30 years according to Moore‘s Law
• But: Life is not digital, not deterministic, not algorithmic
Davidson 1
Davidson 2
Neuron
A new computing paradigm –
• From Algorithms …Arithmetic, Accounting, Differential Equations
• … To SystemsCoordination of Sub-ProcessesCommunicationPerceptionAutonomous Action
Organic Computing:
Organisms are Computers!
Computers should be Organisms!!
Organic Computing is not
Molecular computing,
about faster computers
but
being fault-tolerant and self-organizing,it will lay the foundation for molecular and massively parallel computers
IBM‘s Autonomic Computing Campaign http://www.ibm.com/research/autonomic
Human:
Detailed Communication
Machine:
Creative Infrastructure: Goals, Methodology, Interpretation, Diagnostics
Algorithms: deterministic, fast, clue-less
Algorithmic Division of LaborAlgorithmic
DOL
Debugging
Comparison of actual result with original goalAutonomous debugging: Goals must be represented in the machine
Human:
Loose Communication
Machine:
Goals
Creative Infrastructure:Methodology, Interpretation, Diagnostics, Debugging, Goals,
Data, „Algorithms“
Organic ComputersOrganic
Computers
Algorithmic Machines...are programmed contain no infrastructuremay be simplehave to be simple
Electronic Organisms...growcontain infrastructurehave to be complex may be complex
Electronic Organisms
In-out vs. out-in
Relevant Methodologies
Neural NetworksFuzzy LogicGenetic AlgorithmsArtificial LifeAutonomous AgentsAmorphous ComputingBelief Propagation
First Application Domains
• Artificial Vision• Autonomous Robots
Autonomous Vehicles Toy Robots Service Robots
• User Interfaces• Natural Language Understanding • Computer Security
van Essen Anatomy
van Essen Wiring
Joachim Triesch
Triesch-cue
Triesch-confidences
Triesch-results
One-Click Learning
Hartmut Loos
Bottles found
One person found
Hartmut Loos
More persons
Hartmut Loos
Face Finding