RobotTaxis to Thrive in Vegas

 Where Else?  Good place for live tests, proof of use in complex use, Uber complexity ...  On to LA they say ... 

Uber and Motional's robotaxis arrive in Las Vegas

It marks the first time public riders and access the autonomous vehicles using Uber.

Steve Dent|@stevetdent|December 7, 2022 6:00 AM

Uber has launched public robotaxi rides in Las Vegas using Motional's Hyundai Ioniq 5 autonomous EVs with the aim of offering a full driverless service to the public in 2023. It will eventually expand to Los Angeles, where the two companies have been testing autonomous Uber Eats deliveries since May 2022. It's all part of a 10-year agreement between Uber and Motional to offer autonomous ride-hailing and deliveries. 

The taxi rides will be monitored by safety drivers, with the goal of launching a fully driverless service to the public in 2023. "Today, Motional becomes the first AV company to conduct all-electric autonomous rides on the Uber network for public passengers," said Motional VP Akshay Jaising. (Uber offered "autonomous" taxi rides using its own self-driving tech back in 2016, but riders were accompanied by engineers ready to take the wheel.) 

Lyft also partners with Motional, a joint venture between Hyundai and Aptiv. It beat Uber to the punch by launching Motional rides in Las Vegas in August this year with safety drivers on board. Lyft has also said it will ditch those drivers and offer true autonomous service by next year. 

Don't be surprised if that timeline changes, though. Apart from Motional, only Alphabet division Waymo and GM's Cruise are offering true driverless services at a reasonably large scale. The Waymo One service is operating in Phoenix and San Francisco, while Cruise rides are currently limited to San Francisco. Both operate only in specific areas of cities (which can be mapped out in great detail) and some vehicles still use safety riders.

Uber co-founder Travis Kalanick once said that the his company had to be "tied for first, at least" in the race to offer true driverless rides. Since both Uber and Lyft rely on Motional for self-driving tech, that's exactly what might happen. ... ' 

Experts Making Complex Decisions

 Note also my mention of 'process mining' which can lead to and simplify useful decision making. 

How Experts Make Complex Decisions

By studying 200 million chess moves, researchers shed light on what gives players an advantage—and what trips them up.

BASED ON THE RESEARCH OF: Yuval Salant, Yevgenia Nayberg  in https://insight.kellogg.northwestern.edu 

Making a simple decision is akin to ordering off a restaurant menu: you evaluate the available options one by one and choose whichever alternative promises to make you happiest or deliver the greatest payoff. But when it comes to more complicated choices—say, shopping for a house, devising a business plan, or evaluating insurance policies—identifying the objective “best” option is impractical, and often impossible.

Choosing a health-insurance plan, for example, requires estimating the likelihood that you’ll need a biopsy or an appendectomy—a multilayered guessing game sure to be fraught with error. Selecting a marketing strategy can be similarly knotty, as every potential move opens the door to myriad reactions from customers and competitors, leading to millions of possible scenarios, any of which the decision-maker can only imperfectly foresee.

“There’s a different dimension to decision-making when the available alternatives are so complex that you can’t even figure out what a given option is worth to you,” says Jörg L. Spenkuch, an associate professor of managerial economics and decision sciences at the Kellogg School.

So what does it take to make a good choice when facing this kind of complexity? Does slowing down or having more experience help—or do these convoluted decisions simply leave everyone grasping at straws, regardless of their expertise or how long they spend pondering their options?

These questions have received limited attention from social scientists, says Yuval Salant, a professor of managerial economics and decision sciences. So he and Spenkuch recently teamed up to shed new light on the dynamics of decision-making in complicated scenarios. In a new study, they derive first-of-their-kind predictions about how people behave when making complex choices by using an unusual laboratory: the chess board. “Complexity and chess go hand-in-hand,” says Salant.

Using an immense dataset of more than 200 million moves from an online chess platform, the researchers draw novel conclusions about how chess players find their way through the fog of complexity. They find that slowing down helps everyone, but that masters of the game benefit considerably more from extra decision time than less-expert players. And counterintuitively, they show that adding a mediocre option into the mix can actually be worse than adding a bad one.

How Chess Players Make Decisions

Chess has several features that make it perfect for studying byzantine decisions.

First, the quality of each move can be objectively ranked. Unlike picking an insurance plan or a marketing strategy, where accurately measuring and ranking alternatives is possible only with a well-functioning crystal ball, certain chess moves can be clearly identified as part of a winning strategy: these moves will (if followed up by subsequent optimal play) guarantee a win, no matter what one’s opponent does. Other moves can similarly guarantee a draw or would result in a loss.

(It may seem bizarre that the famously cerebral game can be boiled down to predefined winning and losing moves. But this was proved more than a century ago by the German mathematician Ernst Zermelo. “He basically said, ‘Chess isn’t an interesting game,’” Salant explains, “‘because either white has a winning strategy no matter what black is doing, or black has a winning strategy no matter what white is doing, or both of them can force a draw.’”)  ... ' 

Complexity of a Knot

Saw this measure come up in addressing complexity.

 How Complex Is a Knot? New Proof Reveals Ranking System That Works.

“Ribbon concordance” will let mathematicians compare knots by linking them across four-dimensional space.

Leila Sloman

Back in 1981, Cameron Gordon introduced a new way to relate two knots — mathematical constructs modeled after the knots that appear in a single thread or string. In his paper, he conjectured that this new relationship could be used to arrange groups of knots according to how complicated they are.

This winter, Ian Agol, a mathematician at the University of California, Berkeley, posted a six-page paper that proved Gordon’s conjecture, giving mathematicians a new way to order knots by complexity. “What was really surprising about this paper is, one, that it’s super short,” said Arunima Ray, a researcher at the Max Planck Institute for Mathematics. “And secondly, that it’s using some tools that are, let’s say, unusual to this particular question.”

Abstractions navigates promising ideas in science and mathematics. Journey with us and join the conversation.

Progress on the question was slow until 2019, when Ian Zemke, a mathematician at Princeton University, showed how to apply powerful new methods to the problem that weren’t around in the 1980s. Agol learned about the conjecture a little over a year ago, and he started working on it in earnest late last fall.

“That was nice, other people were kind of thinking about this problem,” said Gordon, who is a professor at the University of Texas, Austin. “And then, somewhat out of the blue, along comes Ian Agol with his beautifully short, beautifully elegant proof.”

Gordon’s conjecture is one of many in knot theory that attempt to organize the infinitely tangled universe of knots. At the heart of this project is the observation that you can drastically alter a knot’s appearance by twisting the strands or sliding them around. (To prevent mathematicians from simply unraveling the string and retying it however they like, the ends of the string are merged to form a closed loop, like a rubber band.) Given drawings of knots, knot theorists try to figure out which ones are truly distinct, and which are different depictions of the same object.  .... ' 

Algorithms with Prediction

Useful thoughts, technical.   Considering the worst case. 

Algorithms with Predictions  By Michael Mitzenmacher, Sergei Vassilvitskii

Communications of the ACM, July 2022, Vol. 65 No. 7, Pages 33-35   10.1145/3528087

The theoretical study of algorithms and data structures has been bolstered by worst-case analysis, where we prove bounds on the running time, space, approximation ratio, competitive ratio, or other measure that holds even in the worst case. Worst-case analysis has proven invaluable for understanding aspects of both the complexity and practicality of algorithms, providing useful features like the ability to use algorithms as building blocks and subroutines with a clear picture of the worst-case performance. More and more, however, the limitations of worst-case analysis become apparent and create new challenges. In practice, we often do not face worst-case scenarios, and the question arises of how we can tune our algorithms to work even better on the kinds of instances we are likely to see, while ideally keeping a rigorous formal framework similar to what we have developed through worst-case analysis.

A key issue is how we can define the subset of "instances we are likely to see." Here we look at a recent trend in research that draws on machine learning to answer this question. Machine learning is fundamentally about generalizing and predicting from small sets of examples, and so we model additional information about our algorithm's input as a "prediction" about our problem instance to guide and hopefully improve our algorithm. Of course, while ML performance has made tremendous strides in a short amount of time, ML predictions can be error-prone, with unexpected results, so we must take care in how much our algorithms trust their predictors. Also, while we suggest ML-based predictors, predictions really can come from anywhere, and simple predictors may not need sophisticated machine learning techniques. For example, just as yesterday's weather may be a good predictor of today's weather, if we are given a sequence of similar problems to solve, the solution from the last instance may be a good guide for the next.

What we want, then, is merely the best of both worlds. We seek algorithms augmented with predictions that are:

Consistent: when the predictions are good, they are near-optimal on a per instance basis;

Robust: when the predictions are bad, they are near-optimal on a worst-case basis;

Smooth: the algorithm interpolates gracefully between the robust and consistent settings; and

Learnable: we can learn whatever we are trying to predict with sufficiently few examples.

Our goal is a new approach that goes beyond worst-case analysis.14 We identify the part of the problem space that a deployed algorithm is seeing and automatically tune its performance accordingly.

As a natural starting example, let us consider binary search with the addition of predictions. When looking for an element in a large sorted array, classical binary search compares the target with the middle element and then re-curses on the appropriate half (see Figure 1). Consider, however, how we find a book in a bookstore or library. If we are looking for a novel by Isaac Asimov, we start searching near the beginning of the shelf, and then look around, iteratively doubling our search radius if our initial guess was far off (see Figure 2). We can make this precise to show that there is an algorithm with running time logarithmic in the error of our initial guess (measured by how far off we are from the correct location), as opposed to being logarithmic in the number of elements in the array, which is the standard result for binary search. Since the error is no larger than the size of the array, we obtain an algorithm that is consistent (small errors allow us to find the element in constant time) and robust (large errors recover the classical O(log n) result, albeit with a larger constant factor).  ... ' 

Automating Data Science: Prospects and Challenges

 Long, insightful piece, good insights,  somewhat technical, but useful.   A favorite topic we experimented with since very early days.

Home/Magazine Archive/March 2022 (Vol. 65, No. 3)/Automating Data Science/Full Text

 By Tijl De Bie, Luc De Raedt, José Hernández-Orallo, Holger H. Hoos, Padhraic Smyth, Christopher K. I. Williams

Communications of the ACM, March 2022, Vol. 65 No. 3, Pages 76-87 10.1145/3495256

--> Given the complexity of typical data science projects and the associated demand for human expertise, automation has the potential to transform the data science process.

Key insights: 

• Automation in data science aims to facilitate and transform the work of data scientists, not to replace them.

• Important parts of data science are already being automated, especially in the modeling stages, where techniques such as automated machine learning (AutoML) are gaining traction.

• Other aspects are harder to automate, not only because of technological challenges, but because open ended and context-dependent tasks require human interaction.

Introduction

Data science covers the full spectrum of deriving insight from data, from initial data gathering and interpretation, via processing and engineering of data, and exploration and modeling, to eventually producing novel insights and decision support systems. Data science can be viewed as overlapping or broader in scope than other data-analytic methodological disciplines, such as statistics, machine learning, databases, or visualization

To illustrate the breadth of data science, consider, for example, the problem of recommending items (movies, books or other products) to customers. While the core of these applications can consist of algorithmic techniques such as matrix factorization, a deployed system will involve a much wider range of technological and human considerations. These range from scalable back-end transaction systems that retrieve customer and product data in real time, experimental design for evaluating system changes, causal analysis for understanding the effect of interventions, to the human factors and psychology that underliehow customers react to visual information displays and make decisions.

As another example, in areas such as astronomy, particle physics, and climate science, there is a rich tradition of building computational pipelines to support data-driven discovery and hypothesis testing. For instance, geoscientists use monthly global landcover maps based on satellite imagery at sub-kilometer resolutions to better understand how the earth’s surface is changing over time [50]. These maps are interactive and browsable, and they are the result of a complex data-processing pipeline, in which terabytes to petabytes of raw sensor and image data are transformed into databases of automatically detected and annotated objects and information. This type of pipeline involves many steps, in which human decisions and insight are critical, such as instrument calibration, removal of outliers, and classification of pixels.

The breadth and complexity of these and many other data science scenarios means that the modern data scientist requires broad knowledge and experience across a multitude of topics. Together with an increasing demand for data analysis skills, this has led to a shortage of trained data scientists with appropriate background and experience, and significant market competition for limited expertise. Considering this bottleneck, it is not surprising that there is increasing interest in automating parts, if not all, of the data science process. This desire and potential for automation is the focus of this article.

As illustrated in the examples above, data science is a complex process, driven by the character of the data being analyzed and by the questions being asked, and is often highly exploratory and iterative in nature. Domain context can play a key role in these exploratory steps, even in relatively well-defined processes such as predictive modeling (e.g., as characterized by CRISP-DM [5]) where, for example, human expertise in defining relevant predictor variables can be critical.  .... '

AI Turns Rembrandt into Data

Interesting look into the information in a complex image.

AI Turned a Rembrandt Masterpiece into 5.6 Terabytes of Data

Popular Science, Purbita Saha, January 11, 2022

The Rijksmuseum in the Netherlands has posted an ultra-high-resolution image of Rembrandt's "The Night Watch," constructed from 8,439 photos taken with a 100-megapixel HD camera that were digitized, color-corrected, and stitched together by algorithms. The Rijksmuseum said it is the largest digital image of an art piece ever created, at 717 billion pixels and 5.6 terabytes of data. Visitors to the museum's Website can zoom into every 0.0005-millimeter square at fine resolution. The digital version of "The Night Watch" was created following a two-year restoration process in which artificial intelligence was used to restore missing elements from the original painting ...'

P vs NP: Complexity, The Problem today

 This was a big deal when I was in grad school, we did many tests to scope how easy it was to solve problems in evolving contexts.   Had the impression this was still unsolved.    Below a good scoping of the current day views.  Includes good intro video:

 Fifty Years of P vs. NP and the Possibility of the Impossible   By Lance Fortnow

Communications of the ACM, January 2022, Vol. 65 No. 1, Pages 76-85   10.1145/3460351

On May 4, 1971, computer scientist/mathematician Steve Cook introduced the P vs. NP problem to the world in his paper, "The Complexity of Theorem Proving Procedures." More than 50 years later, the world is still trying to solve it. In fact, I addressed the subject 12 years ago in a Communications article, "The Status of the P versus NP Problem."13

The P vs. NP problem, and the theory behind it, has not changed dramatically since that 2009 article, but the world of computing most certainly has. The growth of cloud computing has helped to empower social networks, smartphones, the gig economy, fintech, spatial computing, online education, and, perhaps most importantly, the rise of data science and machine learning. In 2009, the top 10 companies by market cap included a single Big Tech company: Microsoft. As of September 2020, the first seven are Apple, Microsoft, Amazon, Alphabet (Google), Alibaba, Facebook, and Tencent.38 The number of computer science (CS) graduates in the U.S. more than tripled8 and does not come close to meeting demand.

Rather than simply revise or update the 2009 survey, I have chosen to view advances in computing, optimization, and machine learning through a P vs. NP lens. I look at how these advances bring us closer to a world in which P = NP, the limitations still presented by P vs. NP, and the new opportunities of study which have been created. In particular, I look at how we are heading toward a world I call "Optiland," where we can almost miraculously gain many of the advantages of P = NP while avoiding some of the disadvantages, such as breaking cryptography.

As an open mathematical problem, P vs. NP remains one of the most important; it is listed on the Clay Mathematical Institute's Millennium Problems21 (the organization offers a million-dollar bounty for the solution). I close the article by describing some new theoretical computer science results that, while not getting us closer to solving the P vs. NP question, show us that thinking about P vs. NP still drives much of the important research in the area

The P vs. NP Problem

Are there 300 Facebook users who are all friends with each other? How would you go about answering that question? Let's assume you work at Facebook. You have access to the entire Facebook graph and can see which users are friends. You now need to write an algorithm to find that large clique of friends. You could try all groups of 300, but there are far too many to search them all. You could try something smarter, perhaps starting with small groups and merging them into bigger groups, but nothing you do seems to work. In fact, nobody knows of a significantly faster solution than to try all the groups, but neither do we know that no such solution exists.

This is basically the P vs. NP question. NP represents problems that have solutions you can check efficiently. If I tell you which 300 people might form a clique, you can check relatively quickly that the 44,850 pairs of users are all friends. Clique is an NP problem. P represents problems where you can find those solutions efficiently. We don't know whether the clique problem is in P. Perhaps, surprisingly, Clique has a property called NP-complete—that is, we can efficiently solve the Clique problem quickly if and only if P = NP. Many other problems have this property, including 3-Coloring (can a map be colored using only three colors so that no two neighboring countries have the same color?), Traveling Salesman (find the shortest route through a list of cities, visiting every city and returning to the starting place), and tens to hundreds of thousands of others.

Formally, P stands for "polynomial time," the class of problems that one can solve in time bounded by a fixed polynomial in the length of the input. NP stands for "nondeterministic polynomial time," where one can use a nondeterministic machine that can magically choose the best answer. For the purposes of this survey, it is best to think of P and NP simply as efficiently computable and efficiently checkable.

For those who want a longer informal discussion on the importance of the P vs. NP problem, see the 2009 survey13 or the popular science book based on that survey.14 For a more technical introduction, the 1979 book by Michael Garey and David Johnson16 has held up surprisingly well and remains an invaluable reference for those who need to understand which problems are NP-complete.

Why Talk About It Now?

On that Tuesday afternoon in 1971, when Cook presented his paper to ACM Symposium on the Theory of Computing attendees at the Stouffer's Somerset Inn in Shaker Heights, OH, he proved that Satisfiability is NP-complete and Tautology is NP-hard.10 The theorems suggest that Tautology is a good candidate for an interesting set not in [P], and I feel it is worth spending considerable effort trying to prove this conjecture. Such a proof would represent a major breakthrough in complexity theory.

Dating a mathematical concept is almost always a challenge, and there are many other possible times where we can start the P vs. NP clock. The basic notions of algorithms and proofs date back to at least the ancient Greeks, but as far as we know they never considered a general problem such as P vs. NP. The basics of efficient computation and nondeterminism were developed in the 1960s. The P vs. NP question was formulated earlier than that, we just didn't know it.  .... ( more at link including intro video and additional resources) ..... ' 

Reducing Complexity of Big Data

Feature detection and extraction to simplify data use in complex data.  Intriguing approach is worth a look,  Simplifying is always a good idea if it provides insight. 

Algorithm Could Reduce Complexity of Big Data

Texas A&M Engineering News, By Stephanie Jones, March 8, 2021

Researchers at Texas A&M University, the University of Texas at Austin, and Princeton University have developed an algorithm that can be applied to large datasets, with the ability to extract and directly order features from most to least salient. Texas A&M's Reza Oftadeh said, "There are many ad hoc ways to extract these features using machine learning algorithms, but we now have a fully rigorous theoretical proof that our model can find and extract these prominent features from the data simultaneously, doing so in one pass of the algorithm." The algorithm adds a new cost function to an artificial neural network to provide the exact location of features directly ordered by their relative performance, allowing it to perform classic data analysis on larger datasets more efficiently.  ... ' 

Computational Complexity and Quantum

Computational complexity provides measures of how hard things are to compute.   Taught in many as part of analytics curriculum..   Which define classes of complexity.   Though we would like to define these for things like 'AI', turns out that is still quite difficult to say until we can appropriately define define it. .   So how well will quantum computing be able to solve different levels of complexity?  Another hard question.   Here is a starting look at that problem.   Technical.

The Power of Quantum Complexity   By Don Monroe

Communications of the ACM, March 2021, Vol. 64 No. 3, Pages 15-17  10.1145/3446875

For decades, computer scientists have compared the fundamental difficulty of solving various tasks, such as factoring a number or finding the most efficient route for a traveling salesperson. Along this way, they have described an alphabet soup of computational complexity classes and formal techniques for showing how various classes relate to each other.

The advent of quantum computers has introduced new flavors into such classification. It also has given urgency to understanding the potential of these still-limited machines, including the role of mysterious correlations of distant particles, known as entanglement. A recent manuscript concludes that incorporating entanglement into a well-known framework could allow verification of a staggering range of proofs, no matter how long they are.

The new result was first posted on a preprint server in January 2020, and immediately stimulated chatter in the vibrant computational-complexity blogosphere, but it has not yet been peer reviewed. Indeed, the authors already have identified a flaw in an earlier paper they built upon, although they later devised an alternate argument that left their conclusions intact.

If it stands up, the proof also will disprove a longstanding mathematical conjecture, with profound implications for both pure mathematics and physics. As a result, "Quite a few people take this to be true and are trying to follow up on it without really fully understanding it," said Vern Paulsen, a mathematician at the University of Waterloo, Ontario, who was not involved in the work. "It just shows how mainstream to the world of science computational complexity is."

"At some point, people who are working far away from computer science will find another proof of this result in their own language," said Henry Yuen of the University of Toronto. For now, however, "There's a lot of computer science concepts that lend themselves very naturally to putting the pieces together." Yuen co-authored the new paper with fellow computer scientists Zhengfeng Ji of the University of Technology, Sydney; Anand Natarajan and Thomas Vidick of the California Institute of Technology, and John Wright of the University of Texas, Austin. ... '

Considering the Graph Database

A useful challenge, in general graphic databases are more indicative of what they show.    But they also introduce levels of complexity that has to be understood.

Graph databases are proven architectures for storing data with complex relationships. Why aren't more companies using them?
   
 By Isaac Sacolick in InfoWorld

Twenty years ago, my development team built a natural language processing engine that scanned employment, auto, and real estate advertisements for searchable categories. I knew that we had a difficult data management challenge. The data in some ad types were relatively straightforward, like identifying car makes and models, but others required more inference, such as identifying a job category based on a list of skills.

We developed a metadata model that captured all the searchable terms, but the natural language processing engine required the model to expose significant metadata relationships. We knew designing a metadata model with arbitrary connections between data points in a relational database was complex, so we explored using object databases to manage the model.

What we were trying to accomplish back then with object databases can be done better today with graph databases. Graph databases store information as nodes and data specifying their relationships with other nodes. They are proven architectures for storing data with complex relationships.

Graph database usage has certainly grown during the past decade as companies considered other NoSQL and big data technologies. The global graph database market was estimated at $651 million in 2018 and forecasted to grow to $3.73 billion by 2026. But many other big data management technologies, including Hadoop, Spark, and others, have seen much more significant growth in popularity, skill adoption, and production use cases compared to graph databases. By comparison, the big data technology market size was estimated at $36.8 billion in 2018 and forecasted to grow to $104.3 billion by 2026.

I wanted to understand why more organizations aren’t considering graph databases. Developers think in objects and use hierarchical data representations in XML and JSON regularly. Technologists and business stakeholders intrinsically understand graphs since the Internet is an interconnected graph through hyperlinks and concepts like friends and friends of friends from social networks. Then why haven’t more development teams used graph databases in their applications?  .... 

The Certainty of Having to Deal with Uncertainty

Late to this article, nicely done.

Home/Magazine Archive/December 2019 (Vol. 62, No. 12)/Uncertainty/Full Text

Uncertainty    By Peter J. Denning, Ted G. Lewis
Communications of the ACM, December 2019, Vol. 62 No. 12, Pages 26-28
10.1145/3368093

In a famous episode in the "I Love Lucy" television series—"Job Switching," better known as the chocolate factory episode—Lucy and her best-friend coworker Ethel are tasked to wrap chocolates flowing by on a conveyor belt in front of them. Each time they get better at the task, the conveyor belt speeds up. Eventually they cannot keep up and the whole scene collapses into chaos.

The threshold between order and chaos seems thin. A small perturbation—such as a slight increase in the speed of Lucy's conveyor belt—can either do nothing or it can trigger an avalanche of disorder. The speed of events within an avalanche overwhelms us, sweeps away structures that preserve order, and robs our ability to function. Quite a number of disasters, natural or human-made, have an avalanche character—earthquakes, snow cascades, infrastructure collapse during a hurricane, or building collapse in a terror attack. Disaster-recovery planners would dearly love to predict the onset of these events so that people can safely flee and first responders can restore order with recovery resources standing in reserve.

Disruptive innovation is also a form of avalanche. Businesses hope their new products will "go viral" and sweep away competitors. Competitors want to anticipate market avalanches and side-step them. Leaders and planners would love to predict when an avalanche might occur and how extensive it might be.

In recent years complexity theory has given us a mathematics to deal with systems where avalanches are possible. Can this theory make the needed predictions where classical statistics cannot? Sadly, complexity theory cannot do this. The theory is very good at explaining avalanches after they have happened, but generally useless for predicting when they will occur.  .... " 

Learning, Reliability in Complex Systems

Useful piece in Infoq .... learning is hard because we need to continually put it into context for ourselves, goals. Plus the ultimate implications of reliability after they have learned something new.

Jason Hand explores the challenges with learning in complex systems, the relationship between high and low stakes learning opportunities as well as the cost associated.  In InfoQ

Bio: Jason Hand is Senior Cloud Advocate at Microsoft. He writes, presents, and coaches on the principles & nuance of DevOps, Site Reliability Engineering, and modern incident management practices. Named “DevOps Evangelist of the Year” in 2016, he recently authored a book on the topic of Site Reliability Engineering. He is a co-host on the podcast “Community Pulse”- a show on building community in tech

Hand: One of the things I want to talk about today, or the main theme for today is that I hope that I challenge some of the thinking that you've maybe been resting in for a long time in how we build, and operate, and maintain our systems. A lot of the ideas I want to share today came from the travels that I've been doing over the past six or seven months. Back in September, I joined Microsoft, previously I was at a company called VictorOps, which, if you're not familiar with them, they do incident management, on-call management, similar to PagerDuty if you're familiar with that service. I joined Microsoft, and one of the very first things that I was pulled into was what they call Microsoft's Ignite The Tour, which is a global tour. All around the world, there are 17 different cities.

I was pulled into this project, which was actually very interesting, very eye-opening. Over the course of the past 6 months or so, I've traveled almost 92,000 miles, which ended up being a little bit over 3 and a half times around the world, which I never would have thought I would have done in my career. It blows my mind. I've spent a total of over seven days, just in the air alone during that time.

These stats aren't that important, but what is important is that everywhere I go and everybody I talk to, I find that I'm using language like complex systems, and there's not a common understanding of what that actually means. I think - I'm guilty of this - we've fallen into this trap of using words, using terminology that not everyone's on the same page of what it actually means. When I say complex systems, I think sometimes people just accept that as face value and don't actually dig into what does that actually mean when he says complex.

That's a big part of what I want to talk about today, and especially what I want to start with is what do we mean when we're talking about complex systems? Not only that but why - when we're trying to focus on learning so much, we've heard about how it's important to be a learning organization - why are we struggling to actually learn? If you were here for Ryan's [Kitchens] talk previously, he touched on a lot of things that I'm going to also try to amplify a little bit more. It's very difficult for us to really find good methods to learn about our systems, to learn new ways to improve them, and continuously build them and make them better for the world, for everybody that we're trying to serve and trying to make better.  .....

Measuring Document and Chart Complexity

Measuring Document Complexity, I assume to provide feedback to the authoring and generation of the same?   Can it also provide direct suggestions as to how they should be updated for differing goals?

New Data Science Method Makes Charts Easier to Read at a Glance   By Columbia University 

Medical doctors reading EEGs in emergency rooms, first responders looking at multiple screens showing live data feeds from sensors in a disaster zone, brokers buying and selling financial instruments all need to make informed decisions very quickly. Visualization complexity can complicate decision-making when one is looking at data on a chart. When timing is critical, it is essential that a chart be easy to read and interpret.

To help decision-makers in scenarios like these, computer scientists at Columbia Engineering and Tufts University have developed a new method—"Pixel Approximate Entropy"—that measures the complexity of a data visualization and can be used to develop easier to read visualizations. Eugene Wu, assistant professor of computer science, and Gabriel Ryan, a Ph.D. student at Columbia, will describe their findings in "At a Glance: Pixel Approximate Entropy as a Measure of Line Chart Complexity," to be presented at the IEEE VIS 2018 conference in Berlin, Germany.  .... "

Conversation with J Doyne Farmer

I recall following the work of J Doyne Farmer when we were interested in economic models that included arguments about complexity, at the Santa Fe Institute.  Here is a conversation with him that is a good update on what he is working on.  Note the mention of the Prediction Company, which I recall hearing about.  And the link to BiosGroup, which received $5 Million from P&G for Supply Chain modeling. 

Collective Awareness
A Conversation with J. Doyne Farmer [7.17.18] in the Edge

J. DOYNE FARMER is director of the Complexity Economics Programme at the Institute for New Economic Thinking at the Oxford Martin School, professor in the Mathematical Institute at the University of Oxford, and an external professor at the Santa Fe Institute. He was a co-founder of Prediction Company, a quantitative automated trading firm that was sold to the United Bank of Switzerland in 2006. 

Complexity of Contracts

This came up in a conversation about the complexity of agreements,  and in particular how these might be embedded in smart contracts, and how disclosure terms might be part of that.  Has a link to full doc on the study.  What is the measure of complexity?

Complex Disclosure  in HBS Working Knowledge
by Ginger Zhe Jin, Michael Luca, and Daniel Martin

OVERVIEW — This study shows that companies looking to hide unfavorable information might strategically be making contract terms unnecessarily complex, harming consumers and undermining the effectiveness of disclosure. These results highlight a role for regulation that would encourage simpler forms of disclosure. ... " 

When to Hold Em, When to Compete

Nash Equilibrium's use in Competitive Situations is re-examined, with hope for its use in competitive behavior situations.  We looked at it for that and found no golden egg, but it doesn't mean others couldn't find it.   A reexamination.   Complexity technical.  Just because optimum solutions are known to be probably impossibly hard, very good solutions are probably better than what we are doing today.

When to Hold 'Em    By CACM Staff 
Communications of the ACM, Vol. 61 No. 6, Pages 6-7
10.1145/3210585

Neil Savage deserves praise for his informative overview of recent computational results related to Nash equilibrium in his news story "Always Out of Balance" (Apr. 2018). I fully agree that the notion of Nash equilibrium does not always reflect how competitors behave in competitive situations, and that the fact that Nash equilibrium is provably computationally intractable makes it less useful than John Nash himself might have envisioned when he developed it. However, Savage also overstated (somewhat) the effect of intractability by claiming the intractability of computing Nash equilibrium necessitates researchers abandon this notion in favor of other competition-related ideas.

While looking for Nash equilibrium yields additional computational complexity, the decision-making problem is, in general, already computationally intractable (NP-hard) for non-competitive situations (such as when a company makes internal planning decisions). In doing so, a company would be looking for an optimal solution (such as one that would aim to help produce maximum profit), but computational optimization is, in general, NP-hard. Such computational intractability does not mean researchers have to abandon the idea of optimization and look for other ideas. Many real-life problems are NP-hard (such as robotic movement) and what makes working on them such an intellectual and computational challenge.

Indeed, there is no general feasible algorithm (unless P = NP), so computer scientists need to be creative when designing algorithms for specific practical problems.  .... " 

Vladik Kreinovich, EL Paso, TX, USA

New Kinds of Brain Maps

Unique tagging and looking for patterns to determine interaction among the brain cells. Other ways to use this method?

New Brain Maps With Unmatched Detail May Change Neuroscience
in Quanta Magazine by Monique Brouillette 

A technique based on genetic bar codes can easily map the connections of individual brain cells in unprecedented numbers. Unexpected complexity in the visual system is only the first secret it has revealed....."

Iterative Random Forests

New Learning Method:  Sees the Forest and the Trees

".... Researchers at the Department of Energy's Lawrence Berkeley National Laboratory (Berkeley Lab) and University of California, Berkeley have created a novel machine learning method that enables scientists to derive insights from highly complex systems in record time.

In a paper published recently in the Proceedings of the National Academy of Sciences, the researchers describe a technique called "iterative Random Forests," which they say could have a transformative effect on any area of science or engineering with complex systems, such as biology. ... "

Complexity

A means of understanding how systems work and interact with our world, and ultimately what can be solved or not solved by machine learning systems.  Technical.

Chasing Complexity

MIT News  Larry Hardesty

Massachusetts Institute of Technology faculty member Ryan Williams in the Computer Science and Artificial Intelligence Laboratory has made a key contribution toward solving the problem of P versus NP. Much of that contribution stems from his attempts as an IBM research fellow to bridge a divide within theoretical computer science, between researchers who work on computational complexity and those who design algorithms. Williams considered exploiting the symmetry between these two camps, as establishing an algorithm's minimum running time requires generalizing nearly all possible instances of a specific problem that it will ever have to face. His focus was on adapting algorithmic design methods to establish lower bounds. Williams first proved the hypothetical link between algorithms and lower bounds, and then demonstrated its application to a very particular class of logic circuits. "This is essentially the circuit class where progress on P not equal to NP stopped in the mid-'80s," Williams notes.  .... " 

Debt in Machine Learning

The term was new to me, but O'Reilly discusses how this complexity measure changes and how to address it.  This reminds me of complexity measures in retail.  Could it be that autonomous AI systems will ultimately solve this?

What is machine learning debt?
Overcome three types of debt to ship quality machine learning code.  By Matthew Kirk 

" .... Debt is a term many software engineers use to describe a general growth of complexity over time. In general, it’s something that is an obligation to revisit something in the future. We’ll get into the definition of each type of debt, but know that debt generally increases as a project matures.

In this article, we’ll talk about how you will encounter these issues as you deploy your code as well as ways of mitigating the risk.  .... "