Main Page: Difference between revisions
| Line 29: | Line 29: | ||
CDS 231 Robust Control is not offered W 2021. Instead, I'll teach CDS 270 Robust Learning. This will be maximally accessible to grad students and advanced undergrads. P/F only, with any units 3-12 by arrangement based on effort, from watching lecture videos to working on team projects. If at all possible, register and don’t just “call in” or audit so we have a clearer picture of who is involved. You can take just 3 units and watch the videos and discussions and not do any project. | CDS 231 Robust Control is not offered W 2021. Instead, I'll teach CDS 270 Robust Learning. This will be maximally accessible to grad students and advanced undergrads. P/F only, with any units 3-12 by arrangement based on effort, from watching lecture videos to working on team projects. If at all possible, register and don’t just “call in” or audit so we have a clearer picture of who is involved. You can take just 3 units and watch the videos and discussions and not do any project. | ||
Anyone with minimal coursework background in either machine learning or control will be welcome, but hopefully grad students working at the intersection will take the course as well and lead team projects. The course details will depend on who signs up, but will focus on what theory and tools are needed for a "Robust AI" particularly for mission-critical control functions in layered architectures for systems such as future smartgrid, autonomous vehicles, autonomy for space missions (with JPL). The latest in robust control, System Level Synthesis (SLS), is ideal not only for handling SLSDNQD (sparse, local, saturating, delayed, noisy, quantized, distributed) constraints on sense\comms | Anyone with minimal coursework background in either machine learning or control will be welcome, but hopefully grad students working at the intersection will take the course as well and lead team projects. The course details will depend on who signs up, but will focus on what theory and tools are needed for a "Robust AI" particularly for mission-critical control functions in layered architectures for systems such as future smartgrid, autonomous vehicles, autonomy for space missions (with JPL). The latest in robust control, System Level Synthesis (SLS), is ideal not only for handling SLSDNQD (sparse, local, saturating, delayed, noisy, quantized, distributed) constraints on sense\comms|compute/actuate but also including learning and layered architectures. | ||
We'll survey the current state of the art and one focus of projects will be to find counterexamples to the most popular “results.” | We'll survey the current state of the art and one focus of projects will be to find counterexamples to the most popular “results.” | ||
Revision as of 16:03, 2 December 2020
 |
John C. Doyle 道耀 |
|
 |
|
|
|
|
Winter 2021 Courses
CDS 231 Robust Control is not offered W 2021. Instead, I'll teach CDS 270 Robust Learning. This will be maximally accessible to grad students and advanced undergrads. P/F only, with any units 3-12 by arrangement based on effort, from watching lecture videos to working on team projects. If at all possible, register and don’t just “call in” or audit so we have a clearer picture of who is involved. You can take just 3 units and watch the videos and discussions and not do any project.
Anyone with minimal coursework background in either machine learning or control will be welcome, but hopefully grad students working at the intersection will take the course as well and lead team projects. The course details will depend on who signs up, but will focus on what theory and tools are needed for a "Robust AI" particularly for mission-critical control functions in layered architectures for systems such as future smartgrid, autonomous vehicles, autonomy for space missions (with JPL). The latest in robust control, System Level Synthesis (SLS), is ideal not only for handling SLSDNQD (sparse, local, saturating, delayed, noisy, quantized, distributed) constraints on sense\comms|compute/actuate but also including learning and layered architectures.
We'll survey the current state of the art and one focus of projects will be to find counterexamples to the most popular “results.”
Will leverage CDS alums, colleagues, friends and others at Caltech and JPL who are the emerging experts in "robust learning." For example, Nikolai Matni taught a course at Penn in F 2019 that will be a source of inspiration and material. Lots of Caltech faculty are working on important aspects of this, and we'll aim to get them for guest videos and discussions. (But easy on the counterexamples to Nik's or Caltech faculty results...) JPL has interest in this area and some engineers plan to participate and provide their motivation for greater autonomy in space missions, and also the challenges of operating deep space (delay tolerant) communication networks.
An optional direction will use the course ideas to study evolution, adaptation, and learning in bio/neuro systems, from viruses to microbes, microbiomes, immune systems, cancer, brains, groups, and societies, and their interactions and “arms races.” Lots of material there but can be a parallel path as the theory is the same but there are no necessary dependencies between these applications and the ones in tech that will be our main emphasis.
The class meeting "time" won't be used, so don't worry about time conflicts. We will have asynchronous video lectures and no live lectures, and we'll hopefully find convenient times for everyone for discussions... also, I expect we'll somewhat fragment into a few groups based on background and interests for some detailed discussions, but all of that will depend on who signs up...Stay tuned for more details.
Videos with overview of research
Aimed to be accessible to a general audience with an emphasis on neuroscience, biology, and medicine. Not much math. Almost.
It's not very well organized but there are a variety of subfolders with videos, slides, and papers. Download the videos or they will run in preview mode and terminate early.
My latest movie files are in the folder 0.2.NewBE are:
0.Readme.NeuroExpmnts.docx (not a movie)… a document that describes some simple experiments that will be leveraged to explain layering, virtualization, tradeoffs, etc…
1.FragileWebinarUpdate.mp4 (An updated version of part of a webinar on fragility with Martin Rees, Terry Sejnowski, and Pat Churchland at UCSD and that I remade for Caltech’s BE167 class this term.) This talks broadly about laws and layers in cells, internet, brains, language, and the (in)justice system, with an “architectural” analysis of the 14th amendment and infectious hijacking by viruses, bad memes, bad law, etc… this is hopefully fun but motivates the big ideas…
2.BE167Neuro.mp4 (another short movie on recent work on sensorimotor control for BE167 …this is a bit more technical)
3.BEFacNeuro.mp4 (some overlap but more on a bunch of simple neuro experiments that can be done easily at home with minimal equipment but illustrate important general ideas…)
In the subfolder Rants, these experiments are described in 1.NeuroExpmnts.docx and there is a file called 2.RantOnSustain.docx that is as named… it talks about the problems of sustainable infrastructure and motivates SLS… 3.RantOnAI4Ctrl.docx is about how AI and control interact though it is out of date….
A broader laws/layers/levels story is started in the folder 0.1.NewIntro and the rest of the folders have hopefully descriptive names. Overall, neuro and in particular sensorimotor control is a most recent case study but there is lots of (some older) biology and medicine… the newest topic is social architecture which is in various places … (and eager to get more feedback on all of this.)
Brief Bio
John Doyle is the Jean-Lou Chameau Professor of Control and Dynamical Systems, Electrical Engineer, and BioEngineering at Caltech, and received the BS&MS in EE, MIT (1977), and PhD in Math, UC Berkeley (1984)). He was a consultant at Honeywell Systems and Research Center from 1976 to 1990.
Research is on mathematical foundations for complex networks with applications in biology, technology, medicine, ecology, neuroscience, and multiscale physics that integrates theory from control, computation, communication, optimization, statistics (e.g. Machine Learning). An emphasis on universal laws and architectures, robustness/efficiency and speed/accuracy tradeoffs, adaptability, and evolvability and large scale systems with sparse, saturating, delayed, quantized, uncertain sensing, communications, computing, and actuation. Early work was on robustness of feedback control systems with applications to aerospace and process control. His students and research group developed software packages like the Matlab Robust Control Toolbox and the Systems Biology Markup Language (SBML).
Prizes, awards, records, championships include the 1990 IEEE Baker Prize (for all IEEE publications), also listed in the world top 10 “most important" papers in mathematics 1981-1993, IEEE Automatic Control Transactions Award (twice 1998, 1999), 1994 AACC American Control Conference Schuck Award, 2004 ACM Sigcomm Paper Prize and 2016 “test of time” award, and inclusion in Best Writing on Mathematics 2010. Individual awards include 1977 IEEE Power Hickernell, 1983 AACC Eckman, 1984 UC Berkeley Friedman, 1984 IEEE Centennial Outstanding Young Engineer (a one-time award for IEEE 100th anniversary), and 2004 IEEE Control Systems Field Award. Best known for fabulous friends, partner, colleagues, and students. Has held world and national records and championships in various sports, but is otherwise quite fragile.
Old Application Papers
For recent theory papers see Nikolai Matni
For fairly complete list of references see Google Scholar
Neuroscience and Machine Learning : Interpretation of the Precision Matrix and Its Application in Estimating Sparse Brain Connectivity during Sleep Spindles from Human Electrocorticography Recordings Das, Sampson, Lainscsek, Muller, Lin, Doyle, Cash, Halgren, Sejnowski, Neural Computation, 2017
Education and Neuroscience: Tutorial on education for Conference on Decision and Control, 2016
Medicine: Robust efficiency and actuator saturation explain healthy heart rate control and variability Li, Cruz, Chien, Sojoudi, Recht, Stone, Csete, Bahmiller, Doyle (2014), P Natl Acad Sci USA 111 (33)
Medicine: Sepsis: Something Old, Something New, and a Systems View J Crit Care. (2012)
Universal architectures: Architecture, constraints, and behavior, JC Doyle, MC Csete, P Natl Acad Sci USA, vol. 108, Sup 3 15624-15630
Biology: Gycolytic oscillations and limits on robust efficiency, FA Chandra, G Buzi, JC Doyle Science 333(6039):187-192, July 2011
Turbulence: Amplification and nonlinear mechanisms in plane Couette flow., D Gayme, B McKeon, B Bamieh, A Papachristodolou, and J Doyle. Physics of Fluids v23:6:065108 (2011)
Biology: Analysis of autocatalytic networks in biology, G Buzi, U Topcu, J Doyle, Automatica 47:1123-1130 (2011)
Earthquakes: The magnitude distribution of earthquakes near Southern California faults Page, Alderson, and Doyle JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 116, (2011)
Physics: On Lossless Approximations, the Fluctuation-Dissipation Theorem, and Limitations of Measurements, H Sandberg, JC Delvenne, JC Doyle, IEEE Trans Auto Control, v56:2, 293-308 (2011)
Wireless: Cross-layer design in multihop wireless networks, L Chen, SH Low, and JC Doyle, Computer Networks 55:480–496 (2011)
Circuits: Solving Large-Scale Hybrid Circuit-Antenna Problems Lavaei, Babakhani, Hajimiri and Doyle, IEEE Transactions on Circuits and Systems I, vol. 58, no. 2, pp. 374-387, Feb. 2011.
Complexity: Contrasting Views of Complexity and Their Implications For Network-Centric Infrastructures Alderson and Doyle, IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART A: SYSTEMS AND HUMANS, VOL. 40, NO. 4, JULY 2010
Internet: Mathematics and the Internet: A Source of Enormous Confusion and Great Potential Willinger, Alderson, and Doyle, Notices of the AMS Volume 56, Number 5 (2009)
Fire: Fire in the Earth System, Science 324, 481 (2009)
Biology: Robustness of Cellular Functions, Stelling, Sauer, Szallasi, Doyle, and Doyle, Cell, 2004
Biology: Reverse Engineering of Biological Complexity, Csete and Doyle, Science, (2002)
News
- Dennice Gayme (Hopkins) named Carol Linde Croft Faculty Scholar.
- Na (Lina) Li (Harvard) gets NSF CAREER and AFOSR YI awards.
- Javad Lavaei (Berkeley) gets SIAM Control and Systems Theory Prize and AACC Eckman, and too many other awards to list.
- Old: Discover magazine "This man wants to control the internet" by Carl Zimmer, Discover magazine, 2008.
- Newer: Blog and new videos Follow link to dropbox folder with accessible introductory videos and case studies in neuroscience, cell biology, and medical physiology. Our you can go directly to the dropbox folder or see above video lists.
Please download the .mp4 files from the dropbox, otherwise they will run in preview mode, which limits the time.
Old talk slides
U Wisc Madison CS Sept 2012 pdf
UCSB Sage lectures, May 2012. (These are pdf files. Ask me for the ppt if you want to steal anything. I would be very flattered.)
Summary: Universal laws and architectures (maybe start here)
Old Teaching Material
- CDS 213, Robust Control (Spring 2012)
- CDS 212, Feedback Control Theory (Fall 2010)
- The Architecture of Robust, Evolvable Networks (Wi10)
Contact
|
Mailing Address John Doyle |
Contact information E-mail: doyle AT caltech dot edu Admin Assistant: Monica Nolasco |
Other Caltech links |