Teraflops for the Masses: Killer Apps of Tom orrow Pradeep K. Dubey Senior Principal Engineer Corporate Technology Group EDGE UNC, Raleigh, May 23, 2006 Evolution continues … Evolution continues … Upcoming Transition Media Evolution Next Transition Modality-specific streaming Modality-aware transformation Graphics Evolution Multimodal recognition Scene complexity: moderate Local processing dominated Scene complexity: large Global processing dominated Mining Evolution Scene complexity: real-world Physical simulation dominated Dataset: static/structured Response: offline Dataset: dynamic, multimodal Response: interactive Dataset: massive+streaming Response: real-time Workload convergence: multimodal recognition and synthesis over complex datasets Workload convergence: multimodal recognition and synthesis over complex datasets 2 May 2 3 , 2 0 0 6 Pradeep K. Dubey pradeep.dubey@intel.com 1
Evolving tow ards m odel-based com puting Evolving tow ards m odel-based com puting Large dataset mining Semantic Web/Grid Mining Streaming Data Mining Mining Distributed Data Mining Content-based Retrieval Multimodal event/object Recognition Collaborative Filters Indexing Statistical Computing Multidimensional Indexing Machine Learning Streaming Dimensionality Reduction Clustering / Classification Dynamic Ontologies Efficient access to large, unstructured, sparse datasets Model-based: Recognition Stream Processing Bayesian network/Markov Model Photo-real Synthesis Neural network / Probability networks Real-world animation LP/IP/QP/Stochastic Optimization Synthesis Ray tracing Global Illumination Behavioral Synthesis Graphics Physical simulation Kinematics Emotion synthesis Audio synthesis 3 May 2 3 , 2 0 0 6 Pradeep K. Dubey pradeep.dubey@intel.com Video/Image synthesis Document synthesis Recognition Mining Synthesis What is a tumor? Is there a tumor here? What if the tumor progresses? It is all about dealing efficiently with complex multimodal datasets It is all about dealing efficiently with complex multimodal datasets Images courtesy: http://splweb.bwh.harvard.edu:8000/pages/images_movies.html 4 May 2 3 , 2 0 0 6 Pradeep K. Dubey pradeep.dubey@intel.com 2
RMS: Recognition Mining Synthesis RMS: Recognition Mining Synthesis Recognition Mining Synthesis Is it …? What is …? What if …? Find a model Create a model Model instance instance Today Model-less Real-time streaming and Very limited realism transactions on static – structured datasets Tomorrow Model-based Real-time analytics on Photo-realism and multimodal dynamic, unstructured, physics-based recognition multimodal datasets animation 5 May 2 3 , 2 0 0 6 Pradeep K. Dubey pradeep.dubey@intel.com I nteractive RMS ( iRMS) I nteractive RMS ( iRMS) Recognition Mining Synthesis What is …? Is it …? What if …? Create a new Find an existing Model model instance model instance Graphics Rendering + Physical Simulation Learning & Visual Input Synthesized Modeling Streams Visuals Computer Reality Vision Augmentation Most RMS apps are about enabling interactive (real-time) RMS Loop or iRMS Most RMS apps are about enabling interactive (real-time) RMS Loop or iRMS 6 May 2 3 , 2 0 0 6 Pradeep K. Dubey pradeep.dubey@intel.com 3
Next-Generation Entertainm ent Next-Generation Entertainm ent RMS Primitives: Shade/Bounce Model the ball Find the ball Replace the ball the ball Going beyond media-stream encode-decode-transcode! Going beyond media-stream encode-decode-transcode! 7 May 2 3 , 2 0 0 6 Pradeep K. Dubey pradeep.dubey@intel.com Real-tim e DCC Loop Real-tim e DCC Loop Mine/Track/Replace the ball What if … what if … RMS Closed Loop Rendering+Physics+Vision Model the ball Shade/Bounce the ball Going beyond ‘red-eye removal’ Going beyond ‘red-eye removal’ 8 May 2 3 , 2 0 0 6 Pradeep K. Dubey pradeep.dubey@intel.com 4
W here are w e headed … Machine learning Rendering Neural networks Simulation Probabilistic reasoning Fuzzy logic collision detection Belief networks force solver Evolutionary computing global illumination Chaos theory … … Physics Soft Soft Computing Physics? Constraint Dynamics Constraints Dynamics 9 May 2 3 , 2 0 0 6 Pradeep K. Dubey pradeep.dubey@intel.com RMS Com puting Core RMS Com puting Core Gaming Unstructured Information Analytics Vision/Tracking Management Benefiting Applications Real-time asset management, text mining camera stream mining, 3D graphics … Pool of Mathematical Techniques Conic optimization, Subspace projection … Pool of RMS Functions Interior-point, Spectral Bundle, SVM … 1 0 May 2 3 , 2 0 0 6 Pradeep K. Dubey pradeep.dubey@intel.com 5
Computer Vision Computer Vision Rendering Rendering Physical Simulation Physical Simulation (F (F inancial) Analytics inancial) Analytics Data Mining Data Mining Face Body Body Face Global Global Portfolio Portfolio Derivative Derivative CFD CFD Face Face Cloth Cloth Tracking Tracking Detection Detection Illumination Illumination Management Pricing Pricing Management Clustering Clustering Indexing Classification Indexing Classification PDE PDE NLP NLP FIMI FIMI SVM SVM SVM SVM Springs Springs Training Classification Classification Training IPM IPM K- -Means Means K (LP, QP) (LP, QP) Level Set Level Set Particle Particle Text Text Fast Marching Fast Marching Filtering Filtering Monte Carlo Monte Carlo Indexer Indexer Method Method Krylov Iterative Solvers Krylov Iterative Solvers Direct Solver Direct Solver Basic Iterative Solver Basic Iterative Solver Integrator Integrator (PCG) (PCG) (Cholesky) (Cholesky) (Jacobi, GS, SOR) (Jacobi, GS, SOR) Basic matrix primitives Basic matrix primitives (dense/sparse, structured/unstructured) (dense/sparse, structured/unstructured) Workload Convergence � � RMS Primitives Workload Convergence RMS Primitives 1 1 May 2 3 , 2 0 0 6 Pradeep K. Dubey pradeep.dubey@intel.com Hollyw ood to W all Street Hollyw ood to W all Street Common Core Entertainment Business/Finance Computing Kernel Tomorrow Stochastic Optimization Today (Sim Annealing, Genetic Alg, Bayes Learning) Object Recognition Portfolio Selection Vision Numerical Integration Asset Allocation Object Tracking (Monte Carlo, Quasi-MC, Gaussian) Multi-Look Computer Vision: Option Pricing Depth from Stereo Convex Optimization (LP, QP, SOCP, SDP, Network, SVM) Asset-Liability Mesh Refinement Management Combinatorial Optimization Rendering: (Integer Prog, Dynamic Prog) Risk Management Path Tracing Graphics Interest-Rate Differential Equations Solvers Computational Derivative Pricing (Parabolic, Elliptic, Hyperbolic, Fluid Dynamics Finite Element Method, Stochastic) Multi-Party Auctions AI for Games: Path planning Iterative Solvers (Conjugate Gradients, Gauss-Seidel, Jacobi, GMRES) 1 2 May 2 3 , 2 0 0 6 Pradeep K. Dubey pradeep.dubey@intel.com 6
RMS Com puting Core: Scaling to Next Generation Needs Map-based shading Global Illumination based SIMPLEX based linear IPM based LP/ QP/ NLP optimization optimization Mass-Spring based deformation FD/ FE/ FV based deformation Non-Linear Marker-based explicit surface Level Set based implicit surface And tracking tracking Generative Linear manifold based Non-linear manifolds computer recognition/ modeling vision Linear Complementarity problem Non-linear Complementarity Low dimension classifiers High dimension classifiers 1 3 May 2 3 , 2 0 0 6 Pradeep K. Dubey pradeep.dubey@intel.com Sum m ary There are mass applications that require significant increase in compute density • There is nothing as general-purpose as physics! • Visual computing is a proxy of this much larger class (RMS) These applications are not linear extensions of existing usage • Optimal platform for such apps should not be linear extension either There is a significant performance difference between a brute- force CMP Vs. a smart CMP targeted for this class • There is significant opportunity for silicon differentiation These apps will likely be the driver for most future technology vectors • Programming to processor to memory technology 1 4 May 2 3 , 2 0 0 6 Pradeep K. Dubey pradeep.dubey@intel.com 7
1 5 May 2 3 , 2 0 0 6 Pradeep K. Dubey pradeep.dubey@intel.com 8
Recommend
More recommend
