Data Science newsletter – June 8, 2017

GROUP CURATION: N/A

The company announced that it would establish an ad blocker for the Chrome web browser, which has become the most popular in America, employed by nearly half of the nation’s web users. The ad blocker — which Google is calling a “filter” — would roll out next year, and would be the default setting for Chrome when fully functional. In other words, the normal user sparking up their Chrome browser simply wouldn’t see the ads blocked by the system.

What ads would get blocked? The ones not sold by Google, for the most part.

The next frontier for AI in games is Starcraft 2, a space-war strategy game played in real time. Why? Its sheer complexity. Players harvest resources, build armies and fight on virtual terrain filled with pinch-points, alleys and strategic high ground. Gone are the neat grids and prescribed moves of chess or Go – Starcraft is chaos. The subtlety of Starcraft is enormous, as we explored back in 2010. That’s why, seven years later, if you ask the co-founder and CEO of DeepMind, Demis Hassabis, what his next target is after Go, he’ll tell you it’s Starcraft.

In the world’s wealthiest neighbourhoods, artificial intelligence (AI) systems are starting to steer self-driving cars down the streets, and homeowners are giving orders to their smart voice-controlled speakers. But the AI revolution has yet to offer much help to the 3 billion people globally who live in poverty.

That discrepancy lies at the heart of a meeting in Geneva, Switzerland, on 7–9 June, grandly titled the AI for Good Global Summit. The meeting of United Nations agencies, AI experts, policymakers and industrialists will discuss how AI and robotics might be guided to address humanity’s most enduring problems, such as poverty, malnutrition and inequality.

Development agencies are buzzing with ideas, although only a few have reached the stage of pilot experiments. But scientists caution that the rise of AI will also bring societal disruption that will be hard to foresee or manage, and that could harm the world’s most disadvantaged.

San Jose officials have confirmed longtime rumors that Google is eyeing a massive landing around the Diridon Station area in downtown, where it could take up to 6 million square feet in office and R&D space, making it the company’s largest campus in the country.

The Diridon Station area is envisioned by the city to be a huge, mixed-use development area surrounding several major transportation investments, including the addition of high-speed rail and electrifying the existing Caltrain route, an already popular transit mode that runs north and south along the Peninsula. Ultimately the area is expected to include publicly accessible plazas, ground-level retail, and a greenbelt and park along the Los Gatos Creek.

Apple’s iMac updates and new HomePod speaker drew most of the attention at the company’s World Wide Developers keynote. But tucked away in the middle were a short few minutes in which software chief Craig Federighi casually launched Apple into one of the tech industry’s fiercest competitions– the contest to help developers build the next generation of AI-powered applications.

Federighi had already spent some time boasting how Apple was using machine learning to do things like make Siri smarter, or help users rediscover old memories in the Moments photo app. Then he announced that Apple would also be gifting some AI superpowers to developers.

“We want to make powerful machine learning easy for you to incorporate in your apps,” he said. Federighi announced new APIs that help coders building apps for Apple devices do things like recognize faces or animals in photos, or parse the meaning of text.

The dichotomy between machines and living things is narrowing. Today, artificial intelligence (AI) is embedded in all kinds of technology, from robots to social networks. This affects the youngest among us as we see the emergence of an “Internet of Toys.” The trend is what prompted us to explore the impact of those “smart,” interconnected playthings on children. We’ll present our paper, “Hey Google, is it OK if I eat you?: Initial Explorations in Child-Agent Interaction,” at the Interaction Design and Children conference at Stanford University on June 27. This blog post provides a preview of our findings.

In the U.S., car crashes kill roughly 40,000 people a year. That’s a loss of life equivalent to four to five 737 airplanes crashing every week.

“In no other setting would this level of fatality go unnoticed,” said Franz Loewenherz, principal transportation planner for the city of Bellevue, Wash.

What makes it even worse is “a lot of these are preventable crashes,” he said. “If we have the data, we can make smart choices about how to deal with this.”

So Loewenherz and the city have teamed up with Microsoft and the University of Washington to form the Video Analytics Towards Vision Zero Partnership. Their plan is to generate that data and use it to improve roadways and prevent collisions between cars, trucks, bikes and pedestrians. Their ultimate goal is eliminating traffic deaths and serious injuries, a target referred to as “vision zero.”

The “worm” is back in at NASA — sort of.

Twenty-five years after the space agency unceremoniously retired the worm, its simple but iconic logotype, in favor of returning to its original 1958 graphic insignia, the retro logo has fallen back into NASA’s good graces, at least when it comes to its use on merchandise.

A Syrian-born human rights lawyer in Washington, D.C., is collecting those videos, hoping someday they will be used to build criminal cases against the perpetrators of the violence.

But he also faces a major problem: The volume of videos is staggering.

“We have 600,000 videos, and we’re in the process of downloading almost 2,000 videos a day,” says Mohammad Al Abdallah, executive director of the Syria Justice and Accountability Centre, a nonprofit supported by the State Department and a handful of European governments.

I don’t know how many hundreds of technology challenges I’ve heard from doctors over the years. Among those, having the sufficient data they need to tailor individualized treatment plans and to prevent delayed diagnosis are among pain points most frequently mentioned.

Last May, Health Tech Masters Students at Cornell Tech and Emergency Medicine residents at NY-Presbyterian aimed to bridge the gap between technologists and clinicians by hosting a first-ever conference of its kind: one that emphasizes the use of mobile devices to conduct medical research.

The ResearchSuite Conference, organized by chief emergency medicine resident Mark Shankar, took place at NY-Presbytarian Hospital in New York City and brought together over 50 physicians and engineers passionate about the emergence of mobile health for medical research studies. The gathering explored the use of ResearchSuite in this collaboration, referring to the suite of tools available to create applications for medical research.

The Ohio Supercomputer Center recently displayed the power of its new Owens Cluster by running the single-largest scale calculation in the Center’s history.

Scientel IT Corp used 16,800 cores of the Owens Cluster on May 24 to test database software optimized to run on supercomputer systems. The seamless run created 1.25 Terabytes of synthetic data.

The secret of advanced social insight is that there is no secret.

The data, analytics, solutions, and guidance you need are available and accessible, whatever your industry. Your challenge is to choose the right tools – the right methods and sources and of course the right software – for the job at hand.

That is no small caveat, however. Success is correlated with subject-matter expertise and analysis judgment and an open-to-exploration attitude.

Given the Brandwatch context for this article, you very likely have strong social analysis at your disposal – that’s my take-away from a look at
and the Audiences product at the recent – so we’ll focus on key elements that distinguish advanced social insight tech from foundational varieties: the ability to cope with the complexity of written (and spoken) language and to quantify and exploit sentiment, opinion, and emotion.

J.P. Morgan’s quantitative investing and derivatives strategy team, led Marko Kolanovic and Rajesh T. Krishnamachari, has just issued the most comprehensive report ever on big data and machine learning in financial services.

Titled, ‘Big Data and AI Strategies’ and subheaded, ‘Machine Learning and Alternative Data Approach to Investing’, the report says that machine learning will become crucial to the future functioning of markets. Analysts, portfolio managers, traders and chief investment officers all need to become familiar with machine learning techniques. If they don’t they’ll be left behind: traditional data sources like quarterly earnings and GDP figures will become increasingly irrelevant as managers using newer datasets and methods will be able to predict them in advance and to trade ahead of their release.

One of the U.S. Army’s biggest challenges is time. Transporting technical experts to and from combat zones is burdensome and can inhibit the overall success of the mission. At the same time, because technical experts face the same battlefield risks as soldiers, reducing the need for technical personnel on the battlefield puts fewer soldiers at risk.

CECOM has developed a suite of systems, named Virtual Logistics Assistance Representative (VLAR), to maximize self-reliance at combat outposts. The heart and underpinning of VLAR is a set of operations research (O.R.) methods and CECOM’s Causal Bayesian belief networks.

To catch a glimpse of the college course of the future, take a look at UC Berkeley’s Foundations of Data Science (Data 8), the pioneering course that introduces statistics, computer science and data analysis to lower-division students of all majors.

During lecture, students can write and edit code on their laptop computers in real time, working on the same data set and code that the professor is discussing in class. Learning in the class is active, and interactive: Students can try their own calculations, using large data sets that are available at the students’ fingertips, and follow up with questions to the professor.

Underlying these classroom innovations are Jupyter Notebooks (link is external), a web-based platform through which students can work on projects and submit homework assignments and labs.

A mother posting on Facebook about the way her son behaves while playing video games could provide a vital clue for the correct treatment for his epilepsy. This is but one type of social media chatter that is informing data scientists at Penn Medicine’s Health Language Processing Lab (HLP). One of the newest entities with the Penn Institute for Biomedical Informatics, HLP combines social media content with other sources of health information in a unique way aimed at understanding how people use language to communicate health needs.

All sorts of groups across Penn Medicine are harnessing data contained in electronic health records (EHRs) and social media to help improve outcomes. The Abramson Cancer Center, for instance, uses lab tests, radiology visits, and patient-reported symptoms to help lung cancer patients avoid the ER visits. Departments across the health system work with data scientists to use finely tuned algorithms to detect complications or underlying health conditions earlier in the continuum of care, and the Center for Digital Health focuses its efforts on how social media intersects with health care, working to determine how posts might help providers detect health problems before urgent care is needed, or even how Twitter might play a role in fighting HIV.

How can we ensure that artificial intelligence provides the greatest benefit to all of humanity?

By that, we don’t necessarily mean to ask how we create AIs with a sense of justice. That’s important, of course—but a lot of time is already spent weighing the ethical quandaries of artificial intelligence. How do we ensure that systems trained on existing data aren’t imbued with human ideological biases that discriminate against users? Can we trust AI doctors to correctly identify health problems in medical scans if they can’t explain what they see? And how should we teach driverless cars to behave in the event of an accident?

The thing is, all of those questions contain an implicit assumption: that artificial intelligence is already being put to use in, for instance, the workplaces, hospitals, and cars that we all use. While that might be increasingly true in the wealthy West, it’s certainly not the case for billions of people in poorer parts of the world. To that end, United Nations agencies, AI experts, policymakers and businesses have gathered in Geneva, Switzerland, for a three-day summit called AI for Good. The aim: “to evaluate the opportunities presented by AI, ensuring that AI benefits all of humanity.”

We have done a number of smaller builds including DeepLearning01 and DeepLearning02 that we published. While those builds were focused on an introductory, getting your feet wet with frameworks, today we have a different monster, DeepLearning10. This is an 8x NVIDIA GTX 1080 Ti GPU beast and is relatively affordable. We will note that as we are writing the DeepLearning10 piece, DeepLearning11 (a second 8x GPU system) is already up and running half populated with GPUs. As we have seen these systems deployed in Silicon Valley data centers, they are rarely one-offs and are more often clustered.

Rather than remote-controlling drivers all over the country, Boland’s venture, FiveAI, based in Cambridge and Bristol, will be completely driverless. It is the Holy Grail of motoring that every car maker and technology company including Uber is racing to crack. He expects to be testing on the streets of London in the second half of 2019.

And which part of the capital stands ready to become this new frontier land? “Certainly Croydon is a possibility, and so is Bromley.” He simply couldn’t resist the challenge. “It is probably the most exciting thing happening in engineering and computer science at the moment. It is the application of artificial intelligence and machine learning to solve real world problems,” he says.

It isn’t a secret that distributed robotic systems are starting to revolutionize many applications from targeted material delivery (e.g., Amazon Robotics) to precision farming. Assisted by technological advancements such as cloud computing, novel hardware design, and manufacturing techniques, nowadays distributed robot systems are starting to become an important part of industrial activities including warehouse logistics or autonomous transportation.

However, as many engineers and scientists in this field know, several of the heterogeneous characteristics of these systems that make them ideal for certain future applications — robot autonomy, decentralized control, collective emergent behavior, collective learning, knowledge sharing, etc. — hinder the evolution of the technology from academic institutions to the public sphere.

For the first time, astronomers have measured the mass of a star by observing the way its mass deforms light passing by it.

It’s an observation that Einstein predicted but thought could never actually happen, due to the incredibly precise alignment between distant astronomical objects it entails. But using modern observing tools, researchers recently found and tracked two distant stars as they lined up almost perfectly.

We are beginning to hear more companies forming teams of machine learning engineers. This isn’t really a new specialty, as such; as machine learning (and, in particular, deep learning) became the rage in data science circles, data engineers were bound to look to the next step. But what distinguishes a machine learning engineer from a data engineer?

To some extent, machine learning engineers do what software engineers (and good data engineers) have done all along.

I was curious how his plans have taken shape in the couple of months since the announcement, so I caught up with him at his office at the Gates Computer Science Building at Stanford University. Given how influential his career has been to date, I was curious where he would focus his attention from this point forward. We also covered his recommendations for companies that are nearer to the beginning of the journey of implementing artificial intelligence, the emergence of roles like the chief artificial intelligence officer, and the industries that are most likely to be impacted by AI, as well as his comparison between the business cultures in the United States and China, among a variety of other topics.

Zurich, Switzerland November 27-28, produced by Pinnacle Sports & Events [$$$]

If that bicyclist whizzing by seems a little happier than the average gridlock-bound car commuter, you’re not imagining it: A recent study found that two-wheeled commuters were happier than their gas pedal-stomping, car-caged peers.

Becoming a bike commuter might seem daunting, but the benefits can be worth it: exercising regularly, saving money, decreasing your carbon footprint, absolving yourself of guilt over that break-room doughnut.

Benefits aside, that happiness bikers had over their driving counterparts? It’s all about control.

“This is a curated list of medical data for machine learning. This list is provided for informational purposes only, please make sure you respect any and all usage restrictions for any of the data listed here.”

“Python support has been tested, and is now in Beta.”

Going from a research idea to experiments is fundamental. But this step is typically glossed over with little explicit advice. In academia, the graduate student is often left toiling away—fragmented code, various notes and LaTeX write-ups scattered around. New projects often result in entirely new code bases, and if they do rely on past code, are difficult to properly extend to these new projects.

Motivated by this, I thought it’d be useful to outline the steps I personally take in going from research idea to experimentation, and how that then improves my research understanding so I can revise the idea.

For over a year we surveyed thousands of companies from all types of industries and data science advancement on how they managed to overcome these difficulties and analyzed the results. Here are the key things to keep in mind when you’re working on your design-to-production pipeline.

This is a basic example of train and use a basic Keras neural network model (XOR) on iPhone using Apple’s coremltools on iOS11. Note that showing the integration starting from a Keras model to having it running in the iOS app is the main point and not the particular choice of model, in principle a similar approach could be used for any kind of Deep Learning model, e.g. generator part of Generative Adversarial Networks, a Recurrent Neural Network (or LSTM) or a Convolutional Neural Network.

“In two new papers, we explore the ability for deep neural networks to perform complicated relational reasoning with unstructured data. In the first paper – A simple neural network module for relational reasoning – we describe a Relation Network (RN) and show that it can perform at superhuman levels on a challenging task. While in the second paper – Visual Interaction Networks – we describe a general purpose model that can predict the future state of a physical object based purely on visual observations.”

Core ML supports many machine learning models (neural networks, tree ensembles, support vector machines, and generalized linear models). The model should be in the Core ML model format (models with a .mlmodel file extension)1.

In our example, we will use a famous model, the VGG16, which is used to classify images, and luckily for us it is one of the pre-trained models that are available in Core ML.

Streaming methods have been adopted in a number of domains, starting with telecommunications but spreading to search engines, social networks, finance, and time-series analysis. These ideas are also finding application in areas using traditional approaches, but where the rough-and-ready sketching approach is more cost effective. Successful applications of sketching involve a mixture of algorithmic tricks, systems know-how, and mathematical insight, and have led to new research contributions in each of these areas.

This article introduces the ideas behind sketching, with a focus on algorithmic innovations. It describes some algorithmic developments in the abstract, followed by the steps needed to put them into practice, with examples. The article also looks at four novel algorithmic ideas and discusses some emerging areas.