Cognitive Systems High Performance Embedded Computing Workshop Robert Graybill DARPA IPTO Sept 28, 2004

1. Cognitive Systems High Performance Embedded Computing Workshop Robert Graybill DARPA IPTO Sept 28, 2004

2. The Challenge • Computer systems are the backbone of key national infrastructure and critical DoD systems • Virtually all important transactions involve massive amounts of software and multiple computer networks • DoD future vision is “network-centric warfare” • While computational performance is increasing, productivity and effectiveness are not keeping up • Cost of building and maintaining systems is growing out of control • Systems have short lifespans with decreasing ROI • Demands on expertise of users are constantly increasing • Users have to adapt to system interfaces, rather than vice versa • As a result, systems have grown more complex, more fragile, and more difficult to develop We need to change the game completely

3. Capability Provided by Software in DoD Systems is Increasing but so are the Challenges… 90 90 F-22 80 80 Multi-year delays associated with software and system stability 70 70 B-2 60 60 50 50 F-16 F-16 Software and testing delays push costs up Percent of Specification Requirements Involving Software Control 40 40 F-15 F-15 30 30 F-111 F-111 20 20 A-7 A-7 F-4 F-4 10 10 0 0 1960 1960 1964 1964 1970 1970 1975 1975 1982 1982 1990 1990 2000 2000 Ref: Defense Systems Management College

4. IPTO’s Approach Developing Cognitive Systems: Systems that know what they’re doing • A cognitive system is one that • can reason, using substantial amounts of appropriately represented knowledge • can learn from its experience so that it performs better tomorrow than it did today • can explain itself and be told what to do • can be aware of its own capabilities and reflect on its own behavior • can respond robustly to surprise

5. Computing Systems that know what they’re doing can… • …reflect on what goes wrong when an anomaly occurs and anticipate its occurrence in the future • …respond to naturally-expressed user directives to change behavior or increase functionality • …be configured and maintained by non-experts • …reconfigure themselves in response to environmental changesand mission events • …reduce the effort to develop and maintain software • …thwart adversarial systems that don’t know what they’re doing • …preserve “corporate memory” to ease transitions for rotational personnel

6. Protocols Mission Micro Architectures Vdd Scaling Clock Gating Compilers/OS Algorithms Four Tiers of Agile Processing Systems That Know What They’re Doing What’s Next? + Cognitive Processing Hardware Elements SBIRs • Intelligent Systems • - Architectures for Cognitive Information Processing (ACIP) • High-End Application Responsive Computing • High Productivity Computing Systems Program (HPCS) • Mission Responsive Architectures • Polymorphous Computing Architectures Program (PCA) • Power Management • Power Aware Computing and Communications Program (PAC/C) + HECURA + OneSAF Objective System + XPCA??

7. ACIP Program Vision Biological Clues Cognitive Algorithm Clues DoD Mission Challenge Clues MTO 3-D Interconnects, Optical, Nano IPTO Cognitive Efforts, PAL, Real, SRS… ACIP Phase 1 Early Architecture Concepts & In-Context Evaluation - 2 Years - Functional Demonstrations & Algorithm Developments Physical Devices, Interconnect, and Packaging ACIP Phase 2 Full Scale Implementation and Demonstration - 4 Years - ACIP Phase 3 Cognitive Technology DOD System Transitions - 2 Years -

8. Metric Evaluation MTO Technology IPTO Projects ACIP Phase I Study Considerations Deliverables Cognitive Processing Requirements System architecture concepts, models, & evaluations-to-SDR Cognitive Reasoning, Learning & Knowledge Technologies In-Context Evaluations Analyze Cognitive Techniques, Computing, Memory and Control Requirements Early Architecture Concepts Technology Assessments Concept device specification and technology roadmap DoD Applications Requirements & Real-time Constraints Cognitive computing requirements, analysis, specifications and runtime characterization In-Context Cognitive DoD Applications Living Framework Draft Innovative Architecture Concepts Composable runtime concepts HW/SW Architectures & Technologies Phase II Challenge Problem, Metrics, & Go No/Go Definition (2 Year Study) Multiple Disciplinary Teams

9. ACIP Phase I Response Broad Multi-Disciplinary Coverage required for System Innovation • Fantastic Response!!! • Participation Mix (Including Subs) • 9 Defense Contractors • 11 Research Laboratories • 51 Universities • 30 Commercial Companies Large Diverse and Robust Teams Resulted in the Best Concepts Study Technical Framework Concept has Emerged

10. COGENT Funded ACIP Efforts • COGnitive ENGine Technology (COGENT) • Raytheon Company - Network Centric Systems • Polymorphous Cognitive Agent Architecture (PCAA) • Lockheed Martin Advanced Technology Labs • CEARCH: Cognition Enabled Architecture • University of Southern California/ISI

11. ACIP Related SBIRs • Reservoir Labs Inc – Cognitive Processing Hardware and Software elements • Intelligent Automation Inc. – Hardware Architectures for Flexible Component Based Hybrid Cognitive Systems • Hoplite Systems LLC – Cognitive Processing Hardware Elements • Cardinal Research LLC – Cognitive Processing Hardware Elements • Saffron Technologies – Associative Memory Hardware Elements for Cognitive Systems (Funded by AFRL)

12. Cognitive Technology Classification

13. Future Future Future Future Future Future Future Cognitive Services Cognitive_Services Reasoning_Strategy Learning Control_Regime Production_System Semantic Forward_Chaining Analogy ILP JTMS Episodic Backward_Chaining Deduction S_R_Rules Resolution Subsymbolic Reactive_Execution Abduction Preference_Based Bayesian Deliberation Induction Utility_Based Future Compilation HTN Planning EBG Future UC_POP Reflection Future Future Knowledge_Representation Knowledge_Domains First_Order_Logic Space Logical Description_Logic Times Probabilistic Resource Episodic_Trace Semantic_Network Preferences Associative Hopfield_Network Future Fuzzy Future Lockheed Martin

14. Cognitive Computing Requires Innovation Classical Computing Cognitive Processing • History of prior results guides next: “learning” • Markovian –current state only • Memory-oriented; unpredictable access patterns, with metadata guiding access • Processor-oriented; favors regular addressing • Goal oriented – with multiple, possibly incompatible objectives, • Process oriented – history + new perceptions => new knowledge • Context oriented – computation based on metadata from prior results • Procedural, results oriented – apply this function next • Key operations: wide spectrum including complex pattern matching • Key operations: arithmetic & simple scalar decision making • Often multiple “acceptable” results • Single deterministic result • Speculation, futures a first class activity • Parallelism difficult to extract • Functional composition determined at run-time • Functional composition determined at compile time • Dynamic resource management (Reasoning vs Learning Balance) • Largely static resource management

15. Embedded DoD Mission Requirements Ground-Based Goals for Mission Manage to Goals Meta Cognitive Resource Manager ACIP Strawman Framework Cognitive Functional Architecture Reactive……………………….…..………..……..Deliberative Tolerant……..…….………………..……………..……Intolerant Transparent …………………………………………………Opaque Manage to Goals Meta Cognitive Resource Manager Cognitive Algorithms Learning Knowledge Reasoning Probabilistic Symbolic Hybrid Common Set of Services Run-Time High Level Compiler Low Level Resource Allocation & Meta Data Management - Agents Run-Time Low Level Compiler Micro Hardware Architecture GISC GISC CVM SVM TVM PCA (SVM+TVM) + CVM = ACIP???

16. Potential New Research Ideas! Leveraging Embedded Computing Workshop Ideas Chaired by MIT LL and ISI Future Role of Embedded Computing Devices: GP, DSP, GPU, NIC, FPGA, ASIC

17. Physical (COTS) PCA Systems The Solution The Problem War Fighter Mission Application Software Mission Software E D High Level Software Development Environment Software Development C Stable Architecture Abstraction Layer (SAAL) PCA Morphware Technology Computation Density B SVM TVM DTVM X A Low Level Device Vendor Specific Game Chips PCA GPU & NICs PCA FPGA Computational Efficiency Manual Low Visibility Stove Pipe SW Development Environments Developed under PCA program Physical (COTS) PCA Systems Concept

18. Software Developer’s Assistant Embedded Computing Complexity Challenge War Fighter Mission Software High Level Software Development Environment Embedded Software Developer Cognitive SW Development & Runtime Assistant Stable Architecture Abstraction Layer (SAAL) The Solution: Cognitive Software Developer’s Assistant SVM TVM DTVM X Low Level Device Vendor Specific Game Chips PCA GPU & NICs PCA FPGA Developed under PCA program

19. The Future is Yours Become an DARPA Program Manager!!