I have been working on a brief essay entitled “Advancing Control in the Era of ML and AI”. The goal is to lay out the opportunity for the control systems community in the coming years as machine learning and artificial intelligence research progresses. This is a working paper and so comments and suggestions are welcome.
My student Scott Nivison, who just completed his Ph. D. under my guidance at the University of Florida, will present a paper entitled A Sparse Neural Network Approach to Model Reference Adaptive Control with Hypersonic Flight Applications at the 2018 AIAA Guidance, Navigation and Control and Intelligent Systems Conference of the 2018 SciTech (http://scitech.aiaa.org/) Conference in Kissimmee, FL in January 2018.
Abstract: Neural network-based model reference adaptive control (MRAC) is an effective architecture used in the flight control community to combat significant uncertainties where the structure of the uncertainty is unknown. In our previous work, a novel adaptive control architecture called sparse neural network (SNN) was developed in order to improve long-term learning and transient performance of flight vehicles with persistent uncertainties which lie in various regions throughout the operating regime. The SNN is designed to operate with small learning rates in order to avoid high-frequency oscillations and utilizes only a small number of active neurons in the adaptive controller in order to reduce the computational burden on the processor. In this paper, we enhance the SNN architecture by developing an innovative adaptive control term which is used to mitigate a control effectiveness matrix. Furthermore, we design a robust control term and a strict dwell time condition in order to ensure stability while switching between segments. We demonstrate the effectiveness of the SNN approach by controlling a sophisticated hypersonic vehicle with flexible body effects.
With my collaborators Roohallah Khatami and Masood Parvania from the University of Utah, our paper on scheduling and pricing of energy generation and storage has been accepted for publication in the IEEE Transactions on Power Systems (http://ieeexplore.ieee.org/document/8187664/.)
Here is the Abstract:
This paper proposes a fundamental model for continuous-time scheduling and marginal pricing of energy generation and storage in day-ahead power systems operation. The paper begins with formulating the economic operation problem of power systems with generating units and energy storage (ES) devices as a continuous-time optimal control problem, where the Lagrange multiplier trajectory associated with the continuous-time power balance constraint is proven to be the marginal price of energy generation and storage. The marginal price is calculated in closed-form, which reveals that in addition to the incremental cost rates of generating units, the marginal price embeds the financial ES charging offers and discharging bids that are defined as incremental charging utility and incremental discharging cost rates. This paper shows that the adjoint function associated with the ES state equation establishes a temporal dependence between the marginal prices during the ES charge and discharge states. A function space-based method is developed to solve the proposed model, which converts the continuous-time problem into a mixed-integer linear programming problem with finite dimensional decision space. The features of the proposed scheduling and pricing models are demonstrated using numerical studies conducted on the IEEE Reliability Test System.
Professor M. Vidyasagar’s 70th birthday and upcoming retirement were celebrated in a workshop entitled “Emerging Applications of Control and Systems Theory” organized by the University of Texas at Dallas. I had the honor and pleasure of making a presentation entitled “Grid Integration of Renewable Energy and Distributed Control” at this workshop and have posted slides in the Presentations section of this website. Sometime in early 2018, Springer will publish a book containing papers from this workshop. I had the privilege of writing a tribute to him as a Preface to this book. So watch for the book in early 2018 and read the tribute …
I had the honor to give a plenary presentation at the 1st IEEE Conference on Control Technology and Applications in Hawaii on August 28, 2017. I have posted the slides in the Presentations section of this website.
Ideas, comments and suggestions are most welcome.
The MForesight National Summit took place last week. The Summit theme was Factories of the Future: Digital, Distributed, Democratized Manufacturing. I participated in it in my role as a member of its Executive Committee and Leadership Council.
It was a very thought provoking and stimulating event. MForesight has now posted the speaker presentations. The presentations and discussions were centered around the key themes: digitization of manufacturing, distributed supply chains, and proliferation of manufacturing capabilities. A summary of the event has been posted as well.
Gary Pisano articulated the concept of The Great Convergence of manufacturing, services, and knowledge economy with information and data playing a central role in this convergence. Global value chain concept facilitates a deeper understanding of manufacturing in an increasingly connected global economy. The recent Global Value Chain Development Report 2017 contains very interesting information and data on these developments. [The short essay entitled “Global value chains shed new light on trade” by David Dollar is a very readable introduction to the ideas and findings of this report.] After a great deal of very insightful analysis of increasing role of services in production, authors state:
“Perhaps what really matters is not what a person makes but what the person does. For a long time, notions of economic performance have been closely tied to economic sectors— manufacturing, agriculture, and services. In a world of fragmented production these distinctions are hard to sustain and may not be economically meaningful. Instead, the focus could be on the implications of performing certain tasks. Do product design and marketing offer greater scope for innovation and learning-by-doing and thus for productivity growth than product assembly?” [Page 156]
Taken to its logical conclusion, the dis-aggregation of production and corresponding value chain analysis has deep implications. From the point of view an individual, the value added is not merely a function of her/his skills and work but also the the organization where the work is performed. For the organization, it is critical to have the best possible systems that harness the most value and long term competitiveness from the inherent skills and capacity of its workforce. In this connection, the recent HBR article by Nicholas Bloom entitled “Corporations in the Age of Inequality” offers very interesting and relevant insights.
Ultimately, technological and business innovations will shape the future as traditional boundaries between manufacturing, services, and knowledge work become more blurred and global economy evolves in the coming decades. It will also have significant impacts on how we educate the next generations in our schools and universities.
I will be going to the MForesight Annual Summit on July 24-26, 2017 in Washington, DC. I serve on the Executive Committee and the Leadership Council for MForesight. It is a think-and-do tank focusing on the next generation technologies that will strengthen U.S. manufacturing. It is funded jointly by the National Institute for Standards and Technology (NIST) and the National Science Foundation (NSF).
As I think about the discussions that will take place at this upcoming event, I decided to summarize my thoughts, some relevant data and critical questions.
First some data. I will focus on US data although global data is clearly relevant and important.
Here is the total US manufacturing real output (from St Louis Fed):
Manufacturing output grew strongly till 2007, fell during the financial crisis and has rebounded since and is now nearly as high as it was at its peak.
While manufacturing as a fraction of total GDP has hovered in the range 11-13% over the last 50 years, the fraction of manufacturing employment over total employment has reduced from 24% to about 8%.
The loss of employment in the manufacturing sector has ben quite dramatic and is an issue of great importance and much debate. This can be attributed to two main factors: (1) increasing productivity per worker and (2) loss of jobs to other nations. There has been much recent research on both of these factors. First, here is the productivity from St Louis Fed:
BLS estimates that productivity growth in manufacturing was only 0.2% and 0.3% in 2015 and 2016 respectively. Bailey and Bosworth in the paper, “US Manufacturing: Understanding Its Past and Its Potential Future,” Journal of Economic Perspectives, have analyzed the outsized role played by computers and electronics sector in manufacturing output and productivity. They conclude, “In summary, the computer and electronics industry has a large impact on one’s evaluation of the performance of manufacturing. This part of the sector has had tremendous quality-adjusted output and productivity growth, even allowing for data errors. In contrast, the noncomputer part of manufacturing has exhibited very slow output and multifactor productivity gains and only moderate labor productivity growth.”
Loss of jobs has been analyzed in the economics literature. The most recent results along this line are in “Import Competition and the Great US Employment Sag of the 2000s” by Acemoglu et al in the Journal of Labor Economics, 2016. Quoting from their paper: “Between 2000 and 2007, the economy gave back the considerable employment gains achieved during the 1990s, with a historic contraction in manufacturing employment being a prime contributor to the slump. We estimate that import competition from China, which surged after 2000, was a major force behind both recent reductions in US manufacturing employment and—through input-output linkages and other general equilibrium channels— weak overall US job growth. Our central estimates suggest job losses from rising Chinese import competition over 1999–2011 in the range of 2.0–2.4 million.”
With this background, we may begin to ask some key questions for the future of manufacturing in the US:
- How can traditional manufacturing industries become more productive? Certainly, part of the answer depends on specific manufacturing sectors. Some candidate technologies include: cloud computing, sensors and controls, big data analytics, energy efficiency, etc. Better understanding of the potential of technologies by manufacturing sector will be crucial.
- What emerging technologies will have outsized impacts on the future of manufacturing? There are two ways to think about this. One is to consider impact on existing industries and products. This is addressed in question 1 but can be expanded to include emerging technologies such as nanotechnology, biotechnology, machine learning and artificial intelligence, etc. The other is to think about new product categories that do not yet exist but could be enabled by emerging technologies and become large industry sectors over time. Such new products could meet current needs in new and disruptive ways or meet new needs. We need to have better understanding of both these aspects.
- What new academic R&D efforts are needed to in view of the above analysis? Academic R&D ranges from long-term basic research to translational and applied research. From the US manufacturing point of view, we must take into account the key insights gained by the above analysis to shape the academic R&D agenda and develop productive public-private partnerships. Engineering Research Centers, Industry-University Cooperative Research Centers, Manufacturing Innovation Institutes are examples of current models of federally funded research activities. How can these be guided from the manufacturing perspective for maximum success?
From my LinkedIn blog:
Much has been written about the impact of advances in information technology, artificial intelligence, automation, and robotics on work and jobs. These technological advances have eliminated certain categories of jobs, enabled off-shoring and distribution of work, and have also created jobs that did not exist before. The negative impact on those who lose their occupations and work has been devastating and has fueled intense political and social upheaval. Rising inequality, due to multiple causes ranging from technological disruptions, globalization, and economic and social policies, is a topic of great concern worldwide.
A recent (preliminary) report Information Technology and the U.S. Workforce from the National Research Council analyzes these issues, draws several interesting conclusions, and makes several recommendations. For example, it concludes, “These advances in technology will result in automation of some jobs, augmentation of workers’ abilities to perform others, and the creation of still others. The ultimate effects of information technology are determined not just by technical capabilities, but also by how the technology is used and how individuals, organizations, and policy makers prepare for or respond to associated shifts in the economic or social landscape.”
I found the recommendation around the nature of technology choices most stimulating. In particular, this report recommends research efforts that, “Target a deeper understanding of how choices about technology use or functionality can affect the workforce in order to inform the design of technologies and policies that will benefit workers, the economy, and society at large.”
This is a crucially important observation. While technological progress cannot be controlled, it is important that we understand the choices that influence its development. These choices are made by individuals, companies, government and society. And ultimate adoption and success is driven by many factors beyond control of any single entity. But by achieving a deeper and more transparent understanding of the possible choices, we can make have a better insight into their impacts. And for those of us in academic institutions, educating future engineers, scientists, and broader communities of students will be very important.
Let us consider the future of autonomous cars and trucks. This technology has its roots in prior academic research and is currently undergoing rapid progress in the private sector. It offers great potential in saving lives and making the transportation system more efficient and productive. At the same time, there is the possibility that many people might lose their jobs. What are the choices that are being made in developing this technology? Are there technology choices between those that assist human drivers versus those replacing them?
In a recent recode interview, the veteran New York Times technology reporter John Markoff stated, “I saw this dichotomy between machines that replace humans and machines that extend the power of humans. That’s been basically the dichotomy in our industry ever since, this was going back to the very dawn of interactive computing in the early 1960s. McCarthy on one side of the lab at Stanford and Englebart on the other. One wanted to replace the human, one wanted to extend the human. The problem is when you augment human, you need fewer humans. It’s not only a dichotomy but it’s a paradox. I don’t particularly see an easy way [out of it].”
Human history suggests that technological progress and advance has both beneficial and detrimental impacts. Certainly, industrial and agricultural revolutions had profound impacts on work and jobs. Therefore, current developments in automation, robotics, artificial intelligence are likely to have positive and negative impacts in the coming years and decades. By engaging in proactive thinking on the strategic technology research and development choices, we can aim to maximize the positive impacts while managing the negative outcomes. It is exciting to see that the National Science Foundation (NSF) has identified Work at the Human-Technology Frontier: Shaping the Future as one of its 10 Big Ideas.