Deep Planning and Beyond

Brought to my attention,   from ICLR 2020

Deep Imitative Models for Flexible Inference, Planning, and Control

Nicholas Rhinehart, Rowan McAllister, Sergey Levine

Keywords: autonomous driving, imitation learning, planning ... '

AI for Materials Discovery

New, lucrative  and very interesting space

Artificial Intelligence for Materials Discovery

By Don Monroe

Communications of the ACM, April 2023, Vol. 66 No. 4, Pages 9-11   10.1145/3583080

3D chemical compounds floating in space, illustration  ... 

The software-driven successes of deep learning have been profound, but the real world is made of materials. Researchers are turning to artificial intelligence (AI) to help find new materials to provide better electronics and transportation, and the energy to run them.

Despite its undeniable power, however, "Machine learning, especially the deep learning revolution, relies heavily on large amounts of data," said Carla Gomes, a computer scientist at Cornell University. "This is not how science works. [Scientists] don't just memorize things."

"Machine learning as we know it is not enough for scientific discovery," she said. "We still have a long way to go."

Nevertheless, researchers are off to a promising start in addressing materials science.

Combinatorial Explosion

One of the challenges in materials discovery is the astronomical number of compositions that might have interesting properties. "High-entropy alloys" (HEA), for example, combine four or more metals. "If you consider all the elements in the periodic table and you will find that you have many combinations, then infinite combinations of the different elements, so that makes prediction very difficult," explained Ziyuan Rao, a postdoc at the Max Planck Institute for Iron Research in Düsseldorf, Germany.

Nonetheless, Rao and his colleagues created a multistage analysis to search for alloys with low thermal expansion, which are important for cryogenic storage of liquified natural gas and for other purposes. The analysis draws on extensive materials datasets, but the available compositions are a tiny, sparse subset of the universe of perhaps 1050 possibilities.

After training a machine-learning model with this data, the researchers used it to select promising candidates, often completely novel. They then used computationally intensive density-functional theory (DFT) calculations to get more precise estimates of each compound's properties. DFT is a widely used shortcut around full quantum mechanical theory. In fact, researchers at DeepMind recently used deep learning to let DFT determine how electron charge is distributed between competing atoms, a longstanding challenge.

A key feature of the HEA search is active learning, which suggests new compositions to examine that will be most informative. "it's a little different from traditional machine learning," Rao said, which typically aims to increase the accuracy of the model. "We also want to use this model to predict new materials with very good properties."

Indeed, Rao and his colleagues further refined their search by experimentally making and measuring some of the best candidates. "You need real-world data," he said, because "Simulated data sometimes is inaccurate." The experimental results are folded back into the modeling, and the loop is repeated six times. The study successfully identified two new alloy compositions with a tiny thermal expansion coefficient, less than two parts per million per degree.  ... ' 

COVID Modelers Expand their Missions

COVID Modelers Expand their Missions, By Gregory Goth

Commissioned by CACM Staff,March 9, 2023

Last August, the University of Texas COVID-19 Model Consortium posted its model of the expected number of COVID-19 cases that would be arriving at school in September in the form of an interactive map.

In the initial days of the COVID-19 pandemic, a new kind of computational modeling theory began to emerge: instead of relying on traditional public health data as used in "tried and true" SIR (Susceptible, Infected, Recovered) models, new types of data, such as aggregated mobility data gleaned from smartphone GPS signals, helped policymakers and scientists alike begin to make sense of the dangerous and deadly effects of the novel virus.

The fast spread of the virus and the drastic effects it had on daily life were, it turned out, a tailor-made field of possibility for academic computer scientists and public health researchers, who ramped up complex models very quickly.

"In the beginning of the pandemic, there were a lot of national-level data models, and in our virtual water cooler talks we said that was ridiculous," David Rubin, M.D., director of the Children's Hospital of Philadelphia (CHOP) PolicyLab, which undertook localized data modeling on a national scale during the pandemic. "The fact that New York City was surging didn't mean that Montgomery, AL, was surging, so we developed a model that grew to about 800 counties. With lots of local variables incorporated every week, we started to calibrate a model that was more like a weather forecast for a local area, and I think that was the real value of the approach we took."

Other leading-edge modeling efforts also emerged, such as the University of Texas (UT) COVID-19 Modeling Consortium, and the University of Virginia (UVA) Biocomplexity Institute. Among the results of the UT consortium was a staged alert system that treaded a careful line between restrictive measures and keeping society open, forecasting hospital demand on that fine line. At UVA, Biocomplexity Institute researchers began modeling the spread of the disease in January 2020 and had already produced predictions of the effects of social distancing mandates within weeks of the near-national shutdown that March.  .. ' 

Thoughts on Modeling Influencers

Fascinating piece,  Have thought about who 'influencers' are and if they could be reasonably modeled.  Twinned?  Virtual?   Is there also a kind  of influencer model akin to a Language model, which show what comes next to influence most?    Then construct the right thing to make it happen?   Recently became more connected to Youtube, so seeing the specifics.  Thoughts?   Thinking it further.  Someone want to examine?

Virtual Influencers in the Real World ,  By Logan Kugler

Communications of the ACM, March 2023, Vol. 66 No. 3, Pages 23-25     10.1145/3579635

The next time you buy a flashy new outfit after browsing Instagram, or tap the heart button on a particularly compelling TikTok video, you might discover that the person who posted it isn't real—and you might not care at all.

That is, if virtual influencers (and their creators) get their way.

A virtual influencer is a digital personality that posts on social media to build an audience of passionate fans, just like a human influencer; at least, that's how it seems. In reality, a team of humans uses computer-generated imagery (CGI), motion capture, and marketing magic to give a digital avatar a voice, a life, and a brand.

The result makes virtual influencers seem like, well, real people. Just like human influencers, virtual ones share behind-the-scenes posts about their 'lives', as well as promoting their favorite products and brands. Virtual influencers usually sound and/or look like humans—or cartoonish representations of them—though they don't try too hard to hide the fact they are artificial.

Not that audiences seem to care. Top virtual influencers like Lil Miquela, Lu, Noonoouri, and Hatsune Miku have millions of social media followers and routinely post about their lives, feelings, and views. Their followers appear as invested in their lives as they are in the lives of human influencers, if the tens of thousands of likes and comments on virtual influencer posts are any indication.

Virtual influencers even turn digital clout into real-world cash. It's common for virtual influencers to work hand-in-hand with brands to star in their advertisements and promote their products. At least one has even been signed by a talent agency that usually works with human actors and artists. ... '   (more at link)

3D Heart Replica

Uses include individual study, for diagnosis and planning care.

3D-Printed Heart Replicas Look, Pump Like the Real Thing, By MIT News, February 24, 2023

A doctor holds a custom-made robotic heart.

MIT engineers are hoping to help doctors tailor treatments to patients’ specific heart form and function, with a custom robotic heart. Physicists at the Massachusetts Institute of Technology (MIT) led a team that developed a procedure to enable the creation of three-dimensionally (3D) printed replicas of patients’ hearts.

After converting medical images of a patient's heart into a 3D computer model, the researchers used a polymer-based ink to 3D-print a flexible shell identical to the patient's heart.

The researchers developed sleeves that may be wrapped around a 3D-printed heart and aorta to replicate a patient's blood-pumping ability.  Said graduate student Luca Rosalia, "The advantage of our system is that we can recreate not just the form of a patient's heart, but also its function in both physiology and disease."

From MIT News

View Full Article    

Uses of Digital Twins in the Warehouse

Thoughtful look  at Twin based modeling.  5 Cases.

Uses of Digital Twins in the Warehouse   in SupplyChainBrain

The digital twin of a warehouse — a real-time, 3D virtual representation of an actual facility down to the space, people, equipment and inventory — lets operators monitor activity, respond instantly to disruption and model future scenarios to improve performance. Here are five top use cases for a digital twin.

1  Manage volatility and complexity with a static workforce.

Chronic shortages in warehouse labor continue, leaving operators to navigate more complex order flow and heightened customer expectations for service reliability with the same or fewer workers, all as wages and operating costs rise.

The latest U.S. Department of Labor numbers for October 2022 tell the story: 482,000 job openings in transportation and warehousing, with 322,000 hires minus 294,000 quits, layoffs and discharges. Industrywide churn hovers around 50% annually. Worker-retention efforts focus on flexible hours, more varied work, less travel time and less repetitive physical stress, as much as on pay.

Automation is key to filling workforce gaps — not just robotics but also process automation to optimize workflow as companies begin many days uncertain about who will show up to do which jobs. Prioritizing, allocating and dynamically adjusting work to match demand requires control tower visibility across the entire facility.

“If I’m a warehouse VP, director or manager, I want to know what’s going on in my warehouse,” explains Bill Denbigh, vice president, product marketing for cloud-based supply chain software provider Tecsys Inc. “How busy am I? How full am I? Where are the most common pick locations? Where am I running out of space?”

A digital twin provides that baseline operating visualization. Underlying artificial intelligence and machine learning monitor operations in real time, map high-activity areas, flag disruptions and orchestrate movement of people and assets against defined objectives like order priority, throughput, travel time or cost per move. As a virtual copy, it can model “what-if” scenarios without disrupting workflow.

2  Help machines run smarter. 

As DCs become more capital-intensive with adoption of mechanization, automation and robotics, they become more like factories, with a growing share of ROI reliant on maintaining, orchestrating and fully utilizing capital assets.

A digital twin can manage both equipment and processes. It optimizes equipment performance and interaction — say, robotic arms loading a conveyor belt — but also monitors systems for predictive maintenance to minimize downtime. Twins also manage process flow, the movement of people, machines and product in the warehouse space. Near real-time simulation is a key differentiator for the technology. “Say you’re running a digital simulation using AGVs and realize you need to adjust dynamically to the reality of what’s happening on the floor to make them more efficient,” says Joe Vernon, principal business consultant with enterprise software and consulting firm EPAM Systems. “You can now do that over and over again, in very fast cycles.”

The result is shorter simulation times for “smarter” equipment that can be pivoted quickly to adapt to changing conditions, in both simulations and actual operations. Systems can also generate real-time feedback from the floor on performance, constraints from narrow aisles or tight corners, or speed adjustments needed to align with human activity.     ... '

Predictive Reverse Aging for Archaeology

New three dimensional models predicts past faces from Skeletal Structure.

Faces from Ancient Egypt Coming Back to Life in Extraordinary Detail

By Newsweek, January 23, 2023

Said the Face Lab's Caroline Wilkinson, "We are pretty confident in our ability to predict face shape from skeletal structure."

Researchers at the U.K.'s Liverpool John Moores University (LJMU) and Egypt's Cairo University (CU) used software and a "reverse aging" process to replicate ancient Egyptian pharaoh Ramesses II's face.

CU's Sahar Saleem used a computed tomography (CT) scanner to produce a three-dimensional model of Ramesses' head and skull, which formed the basis of the facial reconstruction.

LJMU's Caroline Wilkinson said, "We have tested our methods using CT [scans] from living donors and we have evaluated the facial reconstruction using geometric comparison that shows approximately 70% [of the] surface of the facial reconstruction with less than 2 millimeters of error."

Wilkinson said ancient Egyptian mummies also preserve features like ear shape, creases, or hair pattern, which "should increase the level of accuracy [of the reconstruction]."

From Newsweek   

Synthetic Populations Revisited

Akin to building 'Digital Twins', but here constructing a population, say of your current or targeted customers.    We did similar things to feed simulations being tested, say as to their reaction to new products.  Led to examination of:

CACMmag (@Communications of the ACM) Tweeted: .@Shibaura_it and @UnivKansai #researchers in #Japan have developed a way to assign workplaces to individuals in computer-generated 'synthetic populations'.    See:  http://bit.ly/3J69gzJ   https://t.co/Kuex1rOA4B   and  https://twitter.com/cacmmag/status/1617639341156884480?s=51&t=m6BeqGbF_-LmLEIpixKnFg

See above references, taking a closer look at this and will report back  if useful.    Comments?

We did some Critical Path Work in the Enterprise

 We did some Critical Path Work in the Enterprise, so this was of Interest,  but we did not include many  latency aspects.  Technical,  but worth a look. 

 Distributed Latency Profiling through Critical Path Tracing   By Brian Eaton, Jeff Stewart, Jon Tedesco, N. Cihan Tas  in CACM

Communications of the ACM, January 2023, Vol. 66 No. 1, Pages 44-51  10.1145/3570522

For complex distributed systems that include services that constantly evolve in functionality and data, keeping overall latency to a minimum is a challenging task. Critical path tracing (CPT) is a new applied mechanism for gathering critical path latency profiles in large-scale distributed applications. It is currently enabled in hundreds of different Google services, which provides valuable day-to-day data for latency analysis.

Fast turnaround time is an essential feature for any online service. In determining the root causes of high latency in a distributed system, the goal is to answer a key optimization question: Given a distributed system and workload, which subcomponents can be optimized to reduce latency?

Low latency is an important feature for many Google applications, such as Search,4 and latency-analysis tools play a critical role in sustaining low latency at scale. The systems evolve constantly because of code and deployment changes, as well as shifting traffic patterns. Parallel execution is essential, both across service boundaries and within individual services. Different slices of traffic have different latency characteristics.

CPT provides detailed and actionable information about which subcomponents of a distributed system are contributing to overall latency. This article presents results and experiences as observed in using CPT in a particular application: Google Search.

Critical path describes the ordered list of steps that directly contribute to the slowest path of request processing through a distributed system so optimizing these steps reduces overall latency. Individual services have many subcomponents, and CPT relies on software frameworks17 to identify which subcomponents are on the critical path. When one service calls another, RPC (remote procedure call) metadata propagate critical path information from the callee back to the caller. The caller then merges critical paths from its dependencies into a unified critical path for the entire request.

The unified critical path is logged with other request metadata. Log analysis is used to select requests of interest, and then critical paths from those requests are aggregated to create critical path profiles. The tracing process is efficient, allowing large numbers of requests to be sampled. The resulting profiles give detailed and precise information about the root causes of latency in distributed systems.

An example system. Consider the distributed system in Figure 1, which consists of three services, each with two subcomponents. The purpose of the system is to receive a request from the user, perform some processing, and return a response. Arrows show the direction of requests where responses are sent back in the opposite direction.

Figure 1. A simple distributed system.

The system divides work across many subcomponents. Requests arrive at Service A, which hands request processing off to subcomponent A1. A1 in turn relies on subcomponents B1, B2, and A2, which have their own dependencies. Some subcomponents can be invoked multiple times during request processing (for example, A2, B2, and C2 all call into C1).

Even though the system architecture is apparent from the figure, the actual latency characteristics of the system are hard to predict. For example, is A1 able to invoke A2 and B1 in parallel, or is there a data dependency so the call to B1 must complete before the call to A2 can proceed? How much internal processing does A2 perform before calling into B2? What about after receiving the response from B2? Are any of these requests repeated? What is the latency distribution of each processing step? And how do the answers to all of these questions change depending on the incoming request?

Without good answers to these questions, efforts to improve overall system latency will be poorly targeted and might go to waste. For example, in Figure 1, to reduce the overall latency of A1 and its downstream subcomponents, you must know which of these subcomponents actually impact the end-to-end system latency. Before deciding to optimize, you need to know whether A1 → A2 actually matters. ... ' 

Building a Digital City

Much to consider in making a city and running it in simulation to improve its design and operation.

How Do You Build a (Digital) City?

Bloomberg

Immanual John Milton, August 27, 2022

Technology companies, governments, and retailers are generating three-dimensional worlds to create metaverse spaces, while some cities are developing virtual environments for their citizens. For example, Virtual Helsinki is a digital twin of the center of the Swedish capital, designed to be toured in virtual reality (VR). Although the metaverse-as-city concept is in its infancy, consultants foresee massive profits as VR, augmented reality, and mixed reality technologies overcome clunky interfaces. Consultancy McKinsey & Co. cites shopping, social events, fitness, dating, and education as the activities consumers most prefer in the metaverse. The University of California, Irvine's Ian Larson suggests the principles behind SimCity and other videogames, like socialization and digital community design, can serve as a model for digital city spaces.  ... '

Quantum Algorithm Simulates Evolving State of Quantum Particles

 Scratching to understand this beyond the broadest concept.  Materials I can see.  

Quantum Algorithm Simulates Evolving State of Quantum Particles

By City College of New York

July 28, 2022, Comments

A quantum algorithm that simulates the evolving state of many interacting quantum particles over time has been developed by a multi-institutional team of scientists headed by the City College of New York's Pouyan Ghaemi.

"Our quantum-computing algorithm opens a new avenue to study the properties of materials resulting from strong electron-electron interactions," said Ghaemi. "As a result, it can potentially guide the search for useful materials, such as high-temperature superconductors." Ghaemi also believes the research could enable researchers to consider using quantum computers to explore phenomena that result from the interaction of electrons in solids.

From City College of New York

View Full Article    

Monte Carlo Simulation

 Good intro piece to a method we used for many purposes in the enterprise.   Even creating usable models for key processes that were used for decades.  Consider its similarities to Digital Twins.  

 Monte Carlo Simulation

Darío Weitz   in Towards Data Science, Engineer, Ms. Sc., Former Associate Professor at Ing. en Sistemas de Información, Fac. Reg. Rosario, Univ. Tecnológica Nacional, Argentina. Data Viz Consultant.

Part 1: The News Vendor Problem

In the first article of this series, we defined simulation as a numerical technique consisting of building a mathematical and/or logical model of the system under study and then experimenting with it, collecting data that allows us to obtain an estimator to help solve a complex decision problem.

In the same article, we defined a model as a simplified but valid representation of a real process or system, intending to gain some understanding of its behavior.

We also made a classification of models, distinguishing in particular between continuous models, those in which their behavior (state variables) changes continuously over time, and discrete models, those in which the state variables only change at separate points in time. Another important classification involves static models, those that are a representation of the system at a particular time, and dynamic models, those that evolve over time.

Related to the above classification there are three different types of simulations: continuous event simulation, discrete event simulation, and Monte Carlo simulation.

Principles and concepts about Discrete Event Simulation (DES) were provided in the previously indicated series. We coded several examples with SimPy, an object-oriented, process-based, discrete-event simulation framework based on pure Python. In future articles, we will develop concepts and principles related to continuous event simulation.

In this article (and probably in a couple of others) we are dealing with Monte Carlo Simulations.

Monte Carlo Methods

Monte Carlo Methods (MCM) is a collection of numerical methods for the solution of mathematical problems, where the use of random samples differentiates them from equivalent methods.

The term was coined by the Greek-American physicist Nicholas Metropolis when he was working at Los Alamos National Laboratory with John von Neumann and Stanislaw Ulam in the development of the first atomic bomb. The term gets its origin from the famous casino located in the Principality of Monaco.

The conceptual idea of the MCM consists in the estimation of certain quantities through repeated sampling from models represented in a computer. Two classes of mathematical problems are usually solved with these techniques: integration and optimization.

Concerning the contents described in this series of articles, when we refer to Monte Carlo Simulation models we are talking about static, discrete, stochastic models trying to solve an optimization problem.

From a methodological point of view, a Monte Carlo simulation is a sampling experiment whose aim is to estimate the distribution of a quantity of interest that depends on one or more stochastic input variables. We are particularly interested in calculating point estimates and confidence intervals for that quantities. Inevitably, our estimator will have a sampling error and our first task will be to determine the number of replications to improve the degree of certainty in the value of the estimator.  .... ' 

Another Good Intro to Digital Twins

Well done, overview that is only minimally technical.

Virtual Duplicates, By Neil Savage,   Communications of the ACM, February 2022, Vol. 65 No. 2, Pages 14-16  10.1145/3503798

Back in 1970 during the seventh crewed mission of the Apollo space program (the third intended to land on the Moon), the three astronauts aboard Apollo 13 were calmly going about their duties when an explosion in an oxygen tank rocked the spacecraft, spilling precious air into space and damaging the main engine. Personnel in Mission Control suddenly had to devise a plan to get the crew home, and to do that they had to understand what condition the damaged ship was in and the materials available for repairs, and then test what the astronauts might be able to accomplish.

To figure it out, they turned to the flight simulators used to plan and rehearse the mission. They updated the simulators with current information about the physical state of Apollo 13 and tried various scenarios, eventually coming up with the plan that safely returned the astronauts to Earth. This was, some argue, the first use of a digital twin, a model that simulated the state of a physical system with real-time data and made predictions about its performance under varied conditions.

Digital twins are growing in popularity, especially as the Internet of Things provides data from sensors in all sorts of places. The concept is being applied in a range of areas, from buildings to bridges, from wind turbines to aircraft, from weather systems to the human heart.

A digital twin is more than just a simulation of some arbitrary object or system. "It's not a generic model of an airplane or a car or wind turbine or a generic person," says Karen Willcox, director of the Oden Institute for Computational Engineering and Sciences at the University of Texas at Austin (UT Austin). "It's a personalized model of one particular aircraft or one particular person."

To qualify as a digital twin, Willcox says, the model needs to take into account current information about the state of the system and evolve over time as it is updated with new data about the system. Another distinguishing feature is that the model and the data help people make decisions about the system, which in turn can change the data and require the model to be updated again.

Perhaps the most obvious use of digital twins is to monitor the long-term health of expensive or complex equipment, such as engines, manufacturing equipment, or industrial heating, ventilation, and air conditioning (HVAC) systems. That sort of use is increasingly being touted as part of Industry 4.0, which incorporates digital technology, machine learning, and big data to improve industrial processes.

IBM, for instance, is combining sensors with its Watson artificial intelligence technology to help large companies make decisions about what maintenance to perform and when, to extend the lifetime of equipment and cut costs. In one example, the company created a digital twin of an engine blade in a 777 aircraft to monitor when it begins to degrade and requires upgrade or replacement. Similarly, GE created digital twins of its wind turbines to predict when the equipment will need maintenance and builds that into a schedule, so wind farm operators can address issues before a turbine breaks, avoiding costly downtime. Researchers at Siemens are applying a similar approach to the human body, developing digital twins of individual human hearts they hope could predict the effectiveness of a specific therapy for a particular patient, instead of merely relying on statistics about hearts in general. .... '

Game Theory Modeling Vaccine Rollout

Excerpt,   Found Game Theory to be useful, as the article suggests, when you are modeling problems that include potentially competing and cooperative players.   Most difficult were establishing varying payoffs.    Full article at link.  

How Game Theory Could Solve the COVID-19 Vaccine Rollout Puzzle

Rand, Fortune

"  ..... Game theory is a field of mathematics that models competitive and cooperative human interactions, where a “game” is composed of players, their actions, and the resulting payoffs. Often applied to competitive economic and political contexts, game theory can be valuable for predicting behavior and incentivizing decisions that improve a broader system. In the context of health care, it models individuals' decisionmaking criteria when accessing health services.

In the COVID-19 vaccine rollout, the “players” would be individuals seeking care; the actions would be the individuals' selection of a facility; and the payoffs would be measured in terms of how individuals perceive the risk of vaccination, distance traveled, and level of service available at a chosen facility. The level of service might be captured in terms of the congestion of facilities or a supply-demand ratio.

In past research, game theory predicted whether or not individuals would vaccinate, if herd immunity could be achieved, potential vaccine accessibility, and how individuals select vaccination centers. Among other things, these analyses can help calculate how many vaccines need to be sent to each vaccination center. This approach has been proven valuable in after-the-fact analyses of the H1N1 vaccination campaign in 2009 and the response to Haiti's cholera epidemic in 2010. In those scenarios it enabled the identification of “equilibrium solutions,” which represent how individuals may select vaccination centers when given the freedom to choose. .... " 

Broader Strategies for Cyberattack

 Being more precise about detection data gathering and retaliation.   Wondering what the context of the existing modeling included. 

A Better Kind of Cybersecurity Strategy

MIT News   By Peter Dizikes

Researchers at the Massachusetts Institute of Technology (MIT), Northwestern University, and the University of Chicago contend Russia's use of North Korean IP addresses for a cyberattack during the opening ceremonies of the 2018 Winter Olympics underscored the need for a new cybersecurity strategy involving selective retaliation. Said MIT's Alexander Wolitzky, "If after every cyberattack my first instinct is to retaliate against Russia and China, this gives North Korea and Iran impunity to engage in cyberattacks." After extensive modeling of scenarios in which countries are aware of cyberattacks against them but have imperfect information about the attacks and attackers, the researchers found a successful strategy involves simultaneously improving attack detection and gathering more information about the attackers' identity before retaliating. Wolitsky added, "If you blindly commit yourself more to retaliate after every attack, you increase the risk you're going to be retaliating after false alarms."