Gary Rost Bradski                                                 2/16/2006

Consulting Faculty, Stanford University AI Lab

Manager/Principle Engineer: Machine Learning Group

I

Wk: 408_765_8906

 Research Aim: 

Enable AI; Birthing robots.

Key Accomplishments:

Intel, Santa Clara, CA                                                                          June 1996-Present

At Intel I have both managed and coded, researched and designed, published and taught. At various times I worked on or managed teams in: Audio Visual Speech Recognition (AVSR), Open Source Computer Vision Library (OpenCV); Light Field Rendering; Probabilistic Graphical Models Library (PNL); Convex optimization; Code trace compression; Bio informatics; Machine Learning Library (MLL), and Test reduction in manufacturing.  

I accelerated the use and focus on machine learning in Intel resulting in the establishment of a machine learning tools group in manufacturing, architecture analysis teams analyzing machine learning based code, Intel Research setting up a senior director in the area and the performance library group creating commercial products out of the libraries I directed.  Externally I have worked on and directed popular open source software libraries.  These and other accomplishments are listed below:

Stanley:

I pulled together an Intel computer vision sub-team for the Stanford entry in the DARPA Grand Challenge robot race.

Applying Machine Learning to Manufacturing

Sort Test Reduction/autoGFA                                                                      2004-2005

What: Working with Fab 18 and 14 we automated Sort Test time reduction by predicting the Gross Failure Area (GFA) map on a wafer and focusing testing there, reducing it elsewhere.  

Impact: A detailed ROI  (Return On Investment) analysis and projection was done for ICH6, 1263 process for Fab 22 resulting in a test time savings of 14% with only a 0.03% miss rate (caught at final class test) .  When fully rolled out, this implies a savings of 14%  on a testing costs on order of $1B

 resulting in a division award and a promotion.

Statistical Tools  DNut – IDEAL                                                                  2001-2002

What: Working with a tools group in TMG, (Technology Manufacturing Group) we developed a statistical prediction tool based on boosting and decision trees called Dependency Network Utility (DNut).  This was taken over by TMG and was developed into the Interactive Data Exploration And Learning (IDEAL) tool now in wide use.

Impact: This tool has been instrumental in maintaining uptime by its ability to rapidly find multiple causes of failure. 

Diagnostic Trouble Shooting                                                             2003-Present

What:  Working with Fab 17 I directed the creation of the Bayesian Network Diagnostic Troubleshooting model built on top of the PNL library (below) that diagnoses problems with manufacturing machines and directs users in the best way to fix it in order to shorten or avoid machine down time.  Current model is for deposition microwave subsystem HDP MW.

?         Published by team members in ITJ Q4’04: http://developer.intel.com/technology/itj/2004/volume08issue04/art10_bayesnetwork/p01_abstract.htm

Impact: In test now and is expected to show a shortening or avoiding of down time verses the troubleshooting flow now in use.  Downtime is a substantial cost on a multi-billion capital base. This type of modeling can apply to any machine in manufacturing.

Library products + Enabling/Accelerating the market for MIPS

Computer Vision Library: OpenCV                          1999-2001,      2004 in SSG   

http://www.intel.com/technology/computing/opencv/index.htm

http://www.intel.com/software/products/ipp/index.htm

What: This is a computer vision library I started, coded, built a team around, and released to source forge 4 years ago.  User group is now over 15,000 members, and OpenCV itself has well over 1,000,000 downloads.  It is now far and away the most used computer vision library around the world.

Impact: This was one of libraries that SSG’s Performance Libraries Group used to develop their “Intel Performance Primitives” concept, now spanning their product line.  SSG created 200 performance primitives out of OpenCV adding to their IPP product.  A large part of the vision world now uses it in products, research and classroom. Runs on Windows or Linux, MSVC++, Borland, gcc, Proton.

Machine Learning library                                                                             2001-2004

What:  This library I started spans the function mapping based statistical machine learning that we and others have used at Intel.  It contains CART, MART, boosting, SVM, K-NN, neural net, Random Forests, EM, Stochastic Discrimination, feature selection, cross validation and many clustering methods.

Impact: The code is the engine behind both the IDEAL and autoGFA tools used in manufacturing described above and the code has been transferred to SSG/Performance Library team.  It will appear in external product form in early 2006.

Probabilistic Network Library: PNL                                                 2002-2004

http://www.intel.com/research/mrl/pnl/

What:  I directed the creation of  this library spanning a large portion of the techniques in Probabilistic Graphical Models (Bayesian Networks, Markov Random Fields and mixtures of both).  One interface to it will be U. Pitt’s Genie interface.

Impact: PNL is the engine behind the Diagnostic Troubleshooting project above.  Hope for release this December at NIPS 2005.  14,000+ downloads, 26 external developers.

 -------------------------------------------------------------------------------------------

Work (Pre-Intel):

Quantitative Researcher/Analyst                                                     Feb. 1994-June 1996

First Union National Bank, Charlotte, NC

Helped design and implement First Union’s derivative trading system in C++.  Adapted the Ho and Lee model to price and hedge swaptions, caps and floors.  Researched and developed a Heath, Jarrow, Morton Lattice for pricing longer range interest rate options.  Developed a machine learning trading system for Treasurey futures in C++.  Unix, shell programming, C++ and Matlab.

Research Fellow                                                                                Sept. 1989-Sept. 1993

Boston University

Performed ARPA sponsored research in visual pattern recognition. UNIX, shell programming, X Windows, C and Mathematica.

Independent Consultant                                                                    Oct. 1987-Aug. 1989

Adept Computer Consulting

(1) Application programming for NeuroSoft: User interface for computerized EEG system. Real time signal processing and display, statistical software development. C and Assembly on 8086, 80286.

(2) Subcontract through Automated Vision Systems to IBM: User interface to hard disk drive testing equipment, IBM data network interface, robotic stage controller, VB, C.

Software Engineer                                                                             Sept 1985-Sept. 1987

NeuroScience

Wrote FFT, statistical software and an MS-DOS compatible filing system for a parallel processing diagnostic EEG brain imaging system (1 68000 with 7 6809 slaves) in C and assembly.  Development on VAX/VMS and logic simulators.

Hardware Engineer                                                                            Sept. 1984-Aug. 1985

Schlage Electronics

Designed, built, debugged a digital electronics for a distributed security/access control system making use of radio keys. Linked to an IBM PC user interface and database. C and 6802/6809 Assembly.

Internships:

NASA/AMES Research Center                                                         June 1980-Aug.1980

Developed test circuits to digitize holographic interferogram video of wind tunnel ballistic tests.

EECS Department, UC. Berkeley, Prof. Otto Smith                       June 1979-Aug. 1979

Helped develop a power generating wind turbine and performance measuring circuits.

Hewlett Packard                                                                                June 1978-Aug. 1978

Automated the stress testing and statistical tracking of electronic calculator components.

Education:

Ph.D: Machine Intelligence and Computer Vision              Sept. 1989-Aug.1993

Boston University                                                                              

Thesis:  Dual parts: One, a mathematical modeling of biological working memory; Two, visual pattern recognition via temporal sequences.

BS: Electrical Engineering and Computer Science              Sept. 1978-May 1981

University of California, Berkeley

Physics Major,                                                                                   Sept. 1976-May 1978

University of Oregon, Eugene (Transferred to Berkeley)

Publications: 

1. Cheng-Tao Chu, Yi-An Lin, YuanYuan Yu, Gary Bradski, and Andrew Ng, "Multi-core Machine Learning", submitted to ICML 2006
   • We show an effective parallel API for machine learning that works on multi-core, multi-processor and clusters.
2. Hendrik Dahlkamp, Adrian Kaehler, David Stavens, Sebastian Thrun and Gary Bradski, "Self-supervised Monocular Road Detection in Desert Terrain", submitted to RSS 2006.
   • The actual visual road segmentation technique used in Stanley, winning entry to the Darpa Grand Challenge.
3. Ara Nefian and Gary Bradski "Detection of Drivable Corridors for Off-Road Autonomous Navigation", Submitted to ICIP 2006.
   • Use of nested HMMs to find road corridors.
4. Brad Schumitch, Sebastian Thrun, Gary Bradski and Kunle Olukotun, “The Information-Form Data Association Filter”, to appear, NIPS 2005, Dec. 2005.
   • A solution to the data association problem when tracking 100's of objects.
5. Gary Bradski, Adrian Kaehler and Vadim Pisarevsky, "Learning-Based Computer Vision with OpenCV." Intel Technology Journal. http://developer.intel.com/technology/itj/2005/volume09issue02/art03_learning_vision/p01_abstract.htm  May 2005.
   • Computer Vision techniques supported by OpenCV.
6. Gary Bradski, “Machine Learning in Manufacturing”, Global Semiconductor, 2005.
7. Tao Wang, Qian Diao, Yimin Zhang, Gang Song, Chunrong Lai, Gray Bradski, “A Dynamic Bayesian Network Approach to Multi-cue based Visual Tracking”, International Conference on Pattern Recognition (ICPR), 2004.
8. Gary Bradski, “Open Source Computer Vision Library”, Ch 11 pp 521-582 in Emerging Topics in Compute Vision, Medioni and Kang Eds, Prentice Hall, 2004.
9. K. Grauman, M. Betke, J. Lombardi, J. Gips, G.R. Bradski, “Communication via eye blinks and eyebrow raises: video-based human-computer interfaces”, International Journal Universal Access in the Information Society, Vol 2, #4, pp 331-341, Nov. 2003.
10. Qian Diao, Jianye Lu, Wei Hu, Yimin Zhang, Gary Bradski, “DBN Models and a Prediction Method for Visual Tracking”, 19th Conference on Uncertainty in Artificial Intelligence (UAI), Bayesian Modeling Applications Workshop, Acapulco, Mexico, Aug. 2003
11. Gary Bradski, “Review of Graphical Models: Foundations of Neural Computation”, Neural Networks, 2003.
12. Victor Eruhimov, Kevin Murphy, Gary Bradski, “Intel’s Open-Source Probabilistic Network Library (PNL)”, UAI, August 2003.
13. Gary Bradski, Vadim Pisarevsky, Wayne W. Cheng, Qingcang Yu, Harry H. Cheng, “Ch and OpenCV: an open relationship with vision; from computer vision to optimized production-ready code--with no need to re-code. : An article from: Advanced Imaging”, Advanced Imaging, May 2003.
14. Gary Bradski and James W. Davies, “Motion Segmentation and Pose Recognition with Motion History Gradients”, Machine Vision and Applications, Vol 13, #3, pp 174-184, July 2002.
15. Gary Bradski, “OpenCV: Examples of use and new applications in stereo, recognition and tracking”, Proc. Intern. Conf. on Vision Interface. 2002.
16. G. Bradski and T. Boult eds. “Stereo and Multi-Baseline Vision”, International Journal of Computer Vision 47(1): 5; Apr 2002.
17. Grauman, K, Betke, M, Gips, J. and Bradski G.R., “Communication via Eye Blinks – Detection and Duration Analysis in Real Time”, CVPR  pp I-1010-I-1017 Vol 1, 2001.
18. Gary Bradski, Victor Eruhimov, Sergey Molinov, Valery Mosyagin, and Vadim Pisarevsky, “A Video Joystick from a Toy”,  PUI 2001.
19. Radek Grzeszczuk, Gary Bradski, Michael H. Chu and Jean-Yves Bouguet. "Stereo Based Gesture Recognition Invariant to 3D Pose and Lighting". Proc. IEEE Comput. Soc. Conf. Comput. Vision and Pattern Recogn., Hilton Head, South Carolina, USA, June 2000.
20. Gary Bradski, “The OpenCV Library”, Dr. Dobb’s Journal, Nov. 2000, pp 120-126.
21. Davis, J. and Bradski, G. "Real-time Motion Template Gradients using Intel CVLib", IEEE ICCV'99 FRAME-RATE WORKSHOP, 1999.
22. Grossberg, S. and Bradski, G. "A Self-Organizing Architecture for Invariant 3-D Object Learning and Recognition from Multiple 2-D Views", In Jain, L.C., adn Vemuri, R.V. (Eds.), Industrial Applications of Neural Networks. CRC Press, ISBN 0-8493-9802-9 pp 115-157. 1999.
23. Bradski, G., Yeo, B-L. and M. Yeung. Gesture for video content navigation. In SPIE’99, 3656-24 S6, 1999.
24. Bradski, G. Computer Vision Face Tracking For Use in a Perceptual User Interface. In Intel Technology Journal, http://developer.intel.com/technology/itj/q21998/articles/art_2.htm, Q2 1998.
25. Bradski, G., Yeo, B-L., and Yeung, M. "Gesture and speech for video content navigation", PUI'98, pp75-78, 1998.
26. Bradski, G. "Real Time Face and Object Tracking as a Component of a Perceptual User Interface", IEEE WACV, pp 214-219, 1998.
27. Bradski, G. and Holler, M. "Computer Vision Modules for Game Interfaces and in Virtual Holography", PUI 1997 pp 10-14, 1997.
28. Bradski, G. and Grossberg, S. "Recognition of 3-D objects from multiple 2-D views by a self-organizing neural      architecture," In V. Cherkassky, J.H. Friedman, and H. Wechsler (Eds.), From statistics to neural networks: Theory and pattern  recognition. New York: Springer-Verlag, 1995.
29. Bradski, G. "Neural Networks for Trading," Invited Speaker at Risk Magazine: Advanced Mathematics for Derivatives  Conference, NYC. Sept. 1995.
30. Bradski, G. and Grossberg, S. "VIEWNET: A neural architecture for learning to recognize 3-D objects from multiple 2-D  views," In Proceedings of the conference on intelligent robots and computer vision XIII: Algorithms and computer vision. Bellingham,  WA: SPIE, 1995.
31. Grossberg, S. and Bradski, G. "VIEWNET architectures for invariant 3-D object learning and recognition from multiple  views," In B. Bouchon-meunier, R.R. Yager, and L.A. Zadeh (Eds.), Fuzzy logic and soft computing. Singapore: World Scientific  Publishing, 1995.
32. Bradski, G., and Grossberg, S. "Fast Learning VIEWNET Architectures for Recognizing 3-D Objects from Multiple 2-D   Views," Neural Networks Special Issue on ATR. 1995.
33. Bradski, G. and Grossberg, S., "Fast learning VIEWNET architectures for recognizing 3-D objects from multiple 2-D  views," Neural Networks (Special Issue on Automatic Target Recognition), in press, (1995). Boston, MA: Boston University Technical Report CAS/CNS-TR-93-053. 
34. Bradski, G., Carpenter, G.A., and Grossberg, S. "STORE Working Memory Networks for Storage and Recall of Arbitrary  Temporal Sequences," Biological Cybernetics, 71, 469-480, 1994.
35. Bradski, G. and Grossberg, S. "A neural architecture for 3-D object recognition from multiple 2-D views," In Proceedings of  the world congress on neural networks, San Diego, IV 211-219. Hillsdale, NJ: Erlbaum Associates, 1994.
36. Bradski, G. "Neural Network Architectures for Temporal Pattern Learning, Memory, and Recognition with Application to  3-D Vision," Boston University Thesis. August 1993.
37. Bradski, G., and Cohen, M. "A Fast Learning Architecture for Speaker Independent Speech Recognition," Proceedings of  WCNN-93 Portland, OR., 1993.
38. Bradski, G., and Grossberg, S. "Visual Recognition of 3D Objects from 2D View Sequences," Technical Report CAS/CNS-TR-93-053, 1993.
39. Bradski, G., Carpenter, G.A., and Grossberg, S. " Working Memory Networks for Learning Temporal Order with Application  to 3-D Visual Object Recognition," Neural Computation, 4:2 March 1992.
40. Bradski, G. "Dynamic Programming for Optimal Control of Set-Up Scheduling with Neural Network Modifications,"    Proceedings of IJCNN-92 Baltimore, MD., I, 281-287, 1992. 
41. Bradski, G., Carpenter, G.A., and Grossberg, S. "Working Memories for Storage and Recall of Arbitrary Temporal   Sequences," Proceedings of IJCNN-92 Baltimore, MD., II, 57-63, 1992. 
42. Bradski, G. "Dynamic Programming for Optimal Control of Set-Up Scheduling with Neural Network Modifications,"   Technical Report CAS/CNS-TR-92-002, 1992a.
43. Bradski, G., Carpenter, G.A., and Grossberg, S.,"Working Memory Networks for Learning Multiple Groupings of Temporally Ordered Events: Applications to 3-D Visual Object Recognition," Proceedings of IJCNN-91 Seattle, WA., I, 723-728, 1991. CAS/CNS-TR-91-007 1991
44. Bradski, G., Carpenter, G.A., and Grossberg, S. "Working Memories for Storage and Recall of Arbitrary Temporal   Sequences," Technical Report, CAS/CNS-TR-92-003, 1992b. 

Patents: 24 Total 

9 Issued :

12 Pending: