Big Data Archives - Business Intelligence Info

Category Archives: Big Data

Self-driving car companies complain California test data may mislead

February 23, 2020 Big Data

(Reuters) — Developers of self-driving cars are amping up criticism of a California reporting requirement on test data, saying the data could mislead, as the state prepares to release the latest results for 2019.

Companies such as General Motors’ Cruise and startup Aurora have said the metric, called disengagements, is not an accurate or relevant way to measure their technical progress, even though it is widely used to do just that. The debate is taking on more importance amid delays in the rollout of self-driving vehicles and concerns over a lack of regulation and the prospects for profitability for the companies that make such vehicles. The focus on disengagements — when a human driver must take manual control from a self-driving system — and the backlash from self-driving companies have been growing since the California Department of Motor Vehicles began releasing annual disengagement reports five years ago.

California requires all companies testing self-driving vehicles on public roads to submit an annual report on disengagements and what caused them, “written in plain language.” In 2018, the companies with the most miles between disengagements were Alphabet’s Waymo and Cruise. Companies with the greatest number of disengagements were Apple and Uber. In a February 2019 blog post, Waymo wrote that “the key to self-driving technology safely improving and scaling is through a robust breadth of experience and scenario testing, represented by a wider array of data points beyond disengagement alone.” In the same post, Waymo noted its disengagement rate in 2018 dropped 50% from the previous year, while miles between disengagements nearly doubled.

Self-driving companies say the disengagement data can draw unfair comparisons between companies and their self-driving technology. Aurora co-founder Chris Urmson, who previously headed Waymo’s self-driving program, wrote last month that “these numbers mean little when there’s no clear definition of what constitutes a disengagement.”

Cruise co-founder Kyle Vogt expressed similar concerns in a blog post last month, adding: “The general public and regulators deserve hard, empirical evidence that an (autonomous vehicle) has performance that is super-human” — that is, better than the average human driver — if the deployment of that technology is to have “a positive overall impact on automotive safety and public health.”

Let’s block ads! (Why?)

Big Data – VentureBeat

Google launches TensorFlow library for optimizing fairness constraints

February 22, 2020 Big Data

Google AI today released TensorFlow Constrained Optimization (TFCO), a supervised machine learning library built for training machine learning models on multiple metrics and “optimizing inequality-constrained problems.”

The library is designed to help address issues like fairness constraints and predictive parity and help machine learning practitioners better understand things like true positive rates on residents of certain countries, or recall illness diagnoses depending on age and gender.

In tests with a Wikipedia data set, the library achieved lower false-positive rates when predicting whether a comment on a Wiki is toxic based on race, religion, gender identity, or sexuality, while maintaining similar accuracy rates.

TFCO is made to “take into account the societal and cultural factors necessary to satisfy real-world requirements,” said Andrew Zaldivar on behalf of the TFCO team today in a Google AI blog post.

“The ability to express many fairness goals as rate constraints can help drive progress in the responsible development of machine learning, but it also requires developers to carefully consider the problem they are trying to address,” he said. “A ‘safer’ alternative is to constrain each group to independently satisfy some absolute metric, for example by requiring each group to achieve at least 75% accuracy. Using such absolute constraints rather than relative constraints will generally keep the groups from dragging each other down.”

The library, which includes an optional “two-dataset” approach to improving generalization, is built on a trio of research papers published last year, according to the library GitHub page.

The release of the TFCO library comes a day after Google removed gendered labels from the Cloud Vision API.

Let’s block ads! (Why?)

Big Data – VentureBeat

Google Cloud AI removes gender labels from Cloud Vision API to avoid bias

February 21, 2020 Big Data

Google Cloud AI is removing the ability to label people in images as “man” or “woman” with its Cloud Vision API, the company told VentureBeat today. Labeling is used to classify images and train machine learning models, but Google is removing gendered labels because it violates Google’s AI principle to avoid creating biased systems.

“Given that a person’s gender cannot be inferred by appearance, we have decided to remove these labels in order to align with the artificial intelligence principles at Google, specifically Principle #2: avoid creating or reinforcing unfair bias. After today, a non-gendered label such as ‘person’ will be returned by Cloud Vision API,” a Google spokesperson told VentureBeat in an email.

VentureBeat reached out to Microsoft’s Azure and Amazon’s AWS to ask if they intend to remove gender labels from their cloud AI services. We will update this story with more information when we hear back.

The Google Cloud Vision API provides computer vision for customers to detect objects and faces. Google previously blocked the use of gender-based pronouns in an AI tool in 2018.

Many facial analysis and facial recognition systems on the market today predict gender but have challenges identifying people who do not conform to gender norms, people who are transgender, and women of color.

A study last fall by University of Colorado, Boulder researchers found that AI from Amazon, Clarifai, Microsoft, and others maintained accuracy rates above 95% for cisgender men and women but misidentified trans men as women 38% of the time. People with no gender identity were misidentified 100% of the time.

Lead author Morgan Klaus Scheuerman told VentureBeat he believes Google is attempting to set itself apart from competitors. The analysis found that Google did not provide gender classification services but did allow “man” and “woman” labels. Systems from companies like Microsoft can label people like waitresses, air woman, or military woman.

“We basically discussed [in work last fall] how the decisions that are being made in all systems are inherently political. And in the cases of where you’re kind of classifying things about human beings, it becomes more. I think we should assess more what the political notions of that are. And so I’m very excited that Google is kind of taking that seriously,” he told VentureBeat in a phone interview.

In recent years, researchers like Joy Boulamwini performing system audits found major facial recognition providers tend to work best on white men and worse on women of color.

A lack of high performance for all people is a primary reason lawmakers in state legislatures, cities like San Francisco, and the U.S. Senate have proposed bans or moratoriums on the use of facial recognition systems.

Let’s block ads! (Why?)

Big Data – VentureBeat

Researchers propose data set to measure few-shot learning performance

February 20, 2020 Big Data

A plethora of AI models have been tailored to tackle few-shot classification, which refers to learning a classifier for new classes given only a few examples (an ability humans naturally possess). Improving on it could lead to more efficient algorithms capable of expanding their knowledge without requiring large labeled data sets, but to date, many of the procedures and corpora used to assess progress here are lacking.

That’s why researchers at Google AI, the University of California, Berkeley, and the University of Toronto propose in a preprint paper a benchmark for training and evaluating large-scale, diverse, and more “realistic” few-shot classification models. They say it improves upon previous approaches by incorporating multiple data sets of “diverse” distributions and by introducing realistic class imbalance, which they say allows the testing of robustness across a spectrum from low-shot learning onward.

The work was published in May 2019, but it was recently accepted to the International Conference on Learning Representations (ICLR) that’ll be held in Addis Ababa, Ethiopia in April.

As the team explains, as opposed to synthetic environments, real-life learning experiences are heterogeneous in that they vary by the number of classes and the examples per class. They also measure only within-corpus generalization, and they ignore the relationships between classes when forming episodes — i.e., the coarse-grained classification of dogs and chairs may present different difficulties than the fine-grained classification of dog breeds. (An “episode” encompasses states that come between an initial-state and a terminal-state, such as a game of Chess.)

By contrast, the researchers’ data set — the Meta-Dataset — leverages data from 10 different corpora, which span a variety of visual concepts natural and human-made and vary in the specificity of the class definition. Two are reserved for evaluation, meaning that no two classes from them participate in the training set, while the remaining ones contribute some classes to each of the training, validation, and test splits of classes.

Meta-Dataset separately employs an algorithm for sampling episodes, which aims to yield imbalanced episodes of variable shots (class precision) and ways (accuracy). A prescribed number of examples of each chosen class are chosen uniformly at random to populate the support and query sets.

In experiments, the team trained meta-learning models via training episodes sampled using the same algorithm as used for Meta-Dataset’s evaluation episodes. They say that, tested against Meta-Dataset, the models generally didn’t improve when provided multiple data sources and that they didn’t benefit from meta-learning across the data sets. Moreover, they report that the models weren’t robust to the amount of data in test episodes; rather, each excelled in a different part of the spectrum.

“We believe that our exploration of various models on Meta-Dataset has uncovered interesting directions for future work pertaining to meta-learning across heterogeneous data,” wrote the coauthors, who added that addressing the uncovered shortcomings constitutes an important research goal. “[I]t remains unclear what is the best strategy for creating training episodes, the most appropriate validation creation and the most appropriate initialization.”

Let’s block ads! (Why?)

Big Data – VentureBeat

Untrue News indexes and flags false and misleading news articles

February 19, 2020 Big Data

According to Statista, as of March 2018, around 52% of Americans felt that online news websites regularly reported fake stories in the U.S., with another 34% stating they believed news websites occasionally reported fake stories. Those sentiments aside, sussing out the sources of empirically untrue news online isn’t easy. But a team of researchers from the Berlin Institute of Technology, the University of Valencia, the Federal University of Rio Grande do Sul, and the German Research Centre for Artificial Intelligence recently sought to tackle the problem with a search engine dubbed Untrue News.

It complements work published by University of Indiana researchers concerning Hoaxy, a public search engine for visualizing the spread of claims and fact-checking. In a related study by MIT Computer Science and Artificial Intelligence Laboratory (CSAIL), the coauthors demonstrated AI that could determine whether a source is accurate or politically prejudiced. More recently, CSAIL developed an automated system that can detect false statements of fact; elsewhere, a team from George Washington University, Amazon AWS AI, and startup AdVerifai investigated a machine learning approach to classifying misleading speech.

Untrue News is open source and uses both automated and semi-automated natural language processing techniques to crawl and identify false and misleading news articles. Its AI performs text translation for multiple languages, classification, and the linking of entities like people, organizations, locations, and more, indexing over 64,000 fake stories in Portuguese, German, English, and other languages.

Untrue News combines fake stories around the same subject or person, providing results in users’ own languages. A web crawler taps a template to navigate through websites and extract only publicly available metadata, while a data enrichment module comprising several algorithms — including one that performs entity recognition and document linking to open semantic databases — enriches the content. And for comparison with false stories, Untrue News draws on sources considered trustworthy by the International Fact-Checking Network, the unit of the nonprofit Poynter Institute dedicated to bringing together fact-checkers worldwide.

For each query, Untrue News displays the number of news items found in total, as well as the title of articles by fact-checking agencies, the date of publication, the country in which the news was checked, a link to the original fact, an icon related to the category of trustworthiness, and an excerpt. For instance, news that a Brazilian politician posted about the transfer of 8 million dollars to the Ministry of Education shows a symbol that indicates a partial truth, because the funds were transferred to other ministries as well.

“Building a collaborative, reliable and open-source platform for collective use is the main focus of Untrue News,” wrote the researchers in a paper describing their work. “In a time when rumors and false stories are decisive in building the collective thinking of society, Untrue News presents itself as a platform for fighting disinformation.”

Let’s block ads! (Why?)

Big Data – VentureBeat

Industry chief tells Facebook to adapt to EU

February 18, 2020 Big Data

(Reuters) — EU industry commissioner Thierry Breton said on Monday it was for Facebook to adapt to Europe’s standards, not the other way round, as he criticized the U.S. social media giant’s proposed internet rules as insufficient.

The blunt comments came after a short meeting with Facebook CEO Mark Zuckerberg and two days before Breton is due to present the first of a raft of rules to rein in U.S. tech giants and state-aided Chinese companies.

“It’s not for us to adapt to this company, it’s for this company to adapt to us,” Breton, a former CEO at French telecoms provider Orange and French technology company Atos, told reporters after the meeting.

Zuckerberg had earlier told reporters he had a good, wide-ranging conversation with Breton.

Breton also said he would decide by the end of the year whether to adopt tough rules as part of the digital services act to regulate online platforms and set out their responsibilities.

He dismissed a discussion paper issued by Facebook on Monday that rejects what it calls intrusive regulations and suggests looser rules whereby companies would periodicially report content and publish enforcement data.

“It’s not enough,” Breton said, adding that Facebook had omitted any mention of its market dominance and also failed to spell out its responsibilities.

Force for Good Or Bad?

EU justice chief Vera Jourova, who also met Zuckerberg, was equally adamant on Facebook’s role in the fight against online hate speech, disinformation and election manipulation.

“Facebook cannot push away all the responsibility. Facebook and Mr Zuckerberg have to answer themselves a question ‘who do they want to be’ as a company and what values they want to promote,” she said in a statement.

“It will not be up to governments or regulators to ensure that Facebook wants to be a force of good or bad.”

Breton will announce proposals on Wednesday aimed at exploiting the EU’s trove of industrial data and challenging the dominance of Facebook, Google and Amazon. It will announce rules to govern the use of artificial intelligence too, which will also affect companies such as Facebook.

Referring to the possibility that the EU may hold internet companies responsible for hate speech and other illegal speech published on their platforms, Facebook in its discussion document said this ignored the nature of the internet.

It urged regulators to understand the capabilities and limitations of technology in assessing content and allow internet companies the flexibility to innovate.

Zuckerberg’s visit came on the heels of visits by Alphabet Chief Executive Sundar Pichai and Microsoft President Brad Smith to Brussels last month.

(Reporting by Foo Yun Chee)

Let’s block ads! (Why?)

Big Data – VentureBeat

California’s data privacy rules get clearer

February 16, 2020 Big Data

On Friday, February 7, the California Office of the Attorney General (CAG) published a “notice of modifications” to the California Consumer Privacy Act (CCPA), followed by an update on Monday, February 10. Although the CCPA is now law, the rulemaking process is still ongoing, with a final draft of the law expected sometime before the anticipated enforcement date of July 1, 2020. The CAG is now accepting public comments on these proposed modifications until Tuesday, February 25.

While the latest update doesn’t provide us with the final regulations, it offers much needed clarity in several key areas.

1. The scope of data & businesses subject to CCPA processes is clearer

One of the critical lessons from December’s CCPA hearings was that the law required further clarification on terms essential to the operationalization of the CCPA. This month’s updates do a decent job of alleviating some of the uncertainty by providing definitions, examples, and additional clarifying language. Some highlights include:

Clarification on the definition of “personal information.” A new section titled “Guidance Regarding the Interpretation of CCPA Definitions” (§ 999.302) has been created. Currently, there’s only one subsection (a), which defines what qualifies as personal information (PI) under the CCPA using IP addresses as an illustration. The key takeaway is that whether data is classified as PI depends on if it is — or can be — linked to a consumer or household. Given the title of the section, other terms may be clarified in this fashion at a later point.

New communication methods for accepting data requests are specified. Section 999.312, “Methods for Submitting Requests to Know and Requests to Delete,” now clarifies that businesses should consider making consumer requests for data available through “the methods by which it primarily interacts with consumers.” Subsection (a) states that online-only businesses need only provide an email for customers to submit requests to know. The language around how to accept delete requests, however, remains largely the same.

Exclusions now exist for fulfilling consumer requests to know. New language in subsection (c) of § 999.313, “Responding to Requests to Know and Requests to Delete,” excludes businesses from having to search for PI to fulfill a consumer request for data if several conditions are met. The business must not maintain the PI in a searchable or reasonably accessible format, and the PI must only be maintained for legal or compliance purposes. Finally, the business cannot sell the PI or use it for commercial purposes. If a business informs consumers of these reasons, then it can be exempt from having to include PI meeting these conditions within a consumer request for data.

Explicit details now exist for how service providers can use PI. Section § 999.314 (Service Providers) goes into greater detail about what any entity defined as a service provider can and cannot do with PI. Specifically, subsection (c) has been completely rewritten to list five exceptions where service providers are permitted to retain, use, or disclose personal information. One of the exceptions allows service providers to use data to improve the quality of their services or clean and augment data.

In addition to these highlights, the proposed changes also elaborate on the scope of the CCPA as it applies to entities like authorized agents, who can make requests on a consumer’s behalf, as well as data brokers and other third parties.

2. We now have more details on how opt-out requests and do not track will work

New language in § 999.315, “Requests to Opt-Out” suggests that regulators intend for consumer opt-out requests to be as painless as possible. Subsection (c) seems to be worded explicitly to address the problem of UX “dark patterns” within privacy controls, stating “… a business shall not utilize a method that is designed with the purpose or substantial effect of subverting or impairing a consumer’s decision to opt-out.” Given that dark patterns are suspected of helping companies circumvent parts of the GDPR, the new CCPA subsection makes sense, though it’s not clear how it’ll be enforced.

Additionally, subsections (d)(1) and (d)(2) discuss the role that global privacy controls, such as browser settings like do not track, will play in opt-out requests. Privacy controls that function in accordance with the CCPA are to be treated as opt-out requests, even in the instance they conflict with a consumer’s business-specific settings. Businesses, however, may notify consumers of the conflict and how it might impact their service.

3. The rules on how to provide consumer notices have new detail

The CCPA requires that companies inform consumers about company practices as well as customer’s rights at specific points in the customer’s interaction. The new modifications have specified that online CCPA-required notices should follow industry-recognized accessibility standards like the Web Content Accessibility Guidelines, version 2.1.

Sections for specific notices, like the notice at collection of personal information (§ 999.305) and the notice of right to opt-out of sale (§ 999.306), now include details about where notices should be displayed. For example, the modifications in § 999.305 (4) state that if PI collection happens in a mobile application for a purpose not reasonably expected by a consumer, a “just-in-time” notice with a summary of the collected PI should be provided. Modifications in § 999.306 say that opt-out notices within mobile applications may be provided through a link in the application’s settings menu. For a more thorough understanding of how notice requirements have changed, organizations should take a deeper look at these sections.

What’s next for privacy compliance?

From now until February 25, the CAG will be accepting comments on the current round of CCPA modifications via email or mail. From there, we’ll likely see the process for the final rulemaking record begin. Once the AG prepares the final rulemaking record and the Final Statement of Reasons, these will be submitted to the Office of Administrative Law (OAL) for approval. After 30 working days, the OAL will decide whether to approve the record. If approved, the final record will go to the California Secretary of State. All of this will likely take place sometime before July 1, leaving any stragglers with little time to make significant changes.

Although the CCPA is currently on everyone’s mind, the California law is merely a bellwether of an emerging change taking place within the compliance landscape. Beyond the CCPA, organizations should watch for The California Privacy Rights Act of 2020 (CalPRA), dubbed “CCPA 2.0.” The group Californians for Consumer Privacy is hoping to get the act on November’s ballot. Nebraska, New York, and a handful of other states also seem intent on joining California in implementing privacy legislation. Finally, developments in other countries — India, for example — illustrate how the demand for privacy legislation is growing abroad.

Privacy compliance does seem to be a trend that’s here to stay. Organizations that take the time to thoroughly ensure CCPA compliance today will likely have the systems in place needed to ensure compliance with future legislation.

Michael Osakwe is a tech writer and Content Marketing Manager at Nightfall AI.

Let’s block ads! (Why?)

Big Data – VentureBeat

Roboflow: Popular autonomous vehicle data set contains critical flaws

February 16, 2020 Big Data

A machine learning model’s performance is only as good as the quality of the data set on which it’s trained, and in the domain of self-driving vehicles, it’s critical this performance isn’t adversely impacted by errors. A troubling report from computer vision startup Roboflow alleges that exactly this scenario occurred — according to founder Brad Dwyer, crucial bits of data were omitted from a corpus used to train self-driving car models.

Dwyer writes that Udacity Dataset 2, which contains 15,000 images captured while driving in Mountain View and neighboring cities during daylight, has omissions. Thousands of unlabeled vehicles, hundreds of unlabeled pedestrians, and dozens of unlabeled cyclists are present in roughly 5,000 of the samples, or 33% (217 lack any annotations at all but actually contain cars, trucks, street lights, or pedestrians). Worse are the instances of phantom annotations and duplicated bounding boxes (where “bounding box” refers to objects of interest), in addition to “drastically” oversized bounding boxes.

It’s problematic considering that labels are what allow an AI system to understand the implications of patterns (like when a person steps in front of a car) and evaluate future events based on that knowledge. Mislabeled or unlabeled items could lead to low accuracy and poor decision-making in turn, which in a self-driving car could be a recipe for disaster.

Roboflow: Popular autonomous vehicle data set contains critical flaws

Above: Several example images containing pedestrians that didn’t contain any annotations in the original dataset.

Image Credit: Roboflow

“Open source datasets are great, but if the public is going to trust our community with their safety we need to do a better job of ensuring the data we’re sharing is complete and accurate,” wrote Dwyer, who noted that thousands of students in Udacity’s self-driving engineering course use Udacity Dataset 2 in conjunction with an open-source self-driving car project. “If you’re using public datasets in your projects, please do your due diligence and check their integrity before using them in the wild.”

It’s well understood that AI is prone to bias problems stemming from incomplete or skewed data sets. For instance, word embedding, a common algorithmic training technique that involves linking words to vectors, unavoidably picks up — and at worst amplifies — prejudices implicit in source text and dialogue. Many facial recognition systems misidentify people of color more often than white people. And Google Photos once infamously labeled pictures of darker-skinned people as “gorillas.”

But underperforming AI could inflict far more harm if it’s put behind the wheel of a vehicle, so to speak. There hasn’t been a documented instance of a self-driving car causing a collision, but they’re on public roads only in small numbers. That’s likely to change — as many as 8 million driverless cars will be added to the road in 2025, according to marketing firm ABI, and Research and Markets anticipates there will be some 20 million autonomous cars in operation in the U.S. by 2030.

Above: Examples of errors (red-highlighted annotations were missing in the original dataset).

Image Credit: Roboflow

If those millions of cars run flawed AI models, the impact could be devastating, which would make a public already wary of driverless vehicles more skeptical. Two studies — one published by the Brookings Institution and another by the Advocates for Highway and Auto Safety (AHAS) — found that a majority of Americans aren’t convinced of driverless cars’ safety. More than 60% of respondents to the Brookings poll said that they weren’t inclined to ride in self-driving cars, and almost 70% of those surveyed by the AHAS expressed concerns about sharing the road with them.

A solution to the data set problem might lie in better labeling practices. According to the Udacity Dataset 2’s GitHub page, crowd-sourced corpus annotation firm Autti handled the labeling, using a combination of machine learning and human taskmasters. It’s unclear whether this approach might have contributed to the errors — we’ve reached out to Autti for more information — but a stringent validation step might’ve helped to spotlight them.

For its part, Roboflow tells Sophos’ Naked Security that it plans to run experiments with the original data set and the company’s fixed version of the data set, which it’s made available in open source, to see how much of a problem it would have been for training various model architectures. “Of the datasets I’ve looked at in other domains (e.g. medicine, animals, games), this one stood out as being of particularly poor quality,” Dwyer told the publication. “I would hope that the big companies who are actually putting cars on the road are being much more rigorous with their data labeling, cleaning, and verification processes.”

Let’s block ads! (Why?)

Big Data – VentureBeat

AI Weekly: Machine learning could lead cybersecurity into uncharted territory

February 15, 2020 Big Data

Once a quarter, VentureBeat publishes a special issue to take an in-depth look at trends of great importance. This week, we launched issue two, examining AI and security. Across a spectrum of stories, the VentureBeat editorial team took a close look at some of the most important ways AI and security are colliding today. It’s a shift with high costs for individuals, businesses, cities, and critical infrastructure targets — data breaches alone are expected to cost more than $ 5 trillion by 2024 — and high stakes.

Throughout the stories, you may find a theme that AI does not appear to be used much in cyberattacks today. However, cybersecurity companies increasingly rely on AI to identify threats and sift through data to defend targets.

Security threats are evolving to include adversarial attacks against AI systems; more expensive ransomware targeting cities, hospitals, and public-facing institutions; misinformation and spear phishing attacks that can be spread by bots in social media; and deepfakes and synthetic media have the potential to become security vulnerabilities.

In the cover story, European correspondent Chris O’Brien dove into how the spread of AI in security can lead to less human agency in the decision-making process, with malware evolving to adapt and adjust to security firm defense tactics in real time. Should costs and consequences of security vulnerabilities increase, ceding autonomy to intelligent machines could begin to seem like the only right choice.

We also heard from security experts like McAfee CTO Steve Grobman, F-Secure’s Mikko Hypponen, and Malwarebytes Lab director Adam Kujawa, who talked about the difference between phishing and spear phishing, addressed an anticipated rise in personalized spear phishing attacks ahead, and spoke generally to the fears — unfounded and not — around AI in cybersecurity.

VentureBeat staff writer Paul Sawers took a look at how AI could be used to reduce the massive job shortage in the cybersecurity sector, while Jeremy Horwitz explored how cameras in cars and home security systems equipped with AI will impact the future of surveillance and privacy.

AI editor Seth Colaner examines how security and AI can seem heartless and inhuman but still relies heavily on people, who are still a critical factor in security, both as defenders and targets. Human susceptibility is still a big part of why organizations become soft targets, and education around how to properly guard against attacks can lead to better protection.

We don’t know yet the extent to which those carrying out attacks will come to rely on AI systems. And we don’t know yet if open source AI opened Pandora’s box, or to what extent AI might increase threat levels. One thing we do know is that cybercriminals don’t appear to need AI to be successful today.

I’ll leave it to you to read the special issue and draw your own conclusions, but one quote worth remembering comes from Shuman Ghosemajumder, formerly known as the “click fraud czar” at Google and now CTO at Shape Security, in Sawers’ article. “[Good actors and bad actors] are both automating as much as they can, building up DevOps infrastructure and utilizing AI techniques to try to outsmart the other,” he said. “It’s an endless cat-and-mouse game, and it’s only going to incorporate more AI approaches on both sides over time.”

For AI coverage, send news tips to Khari Johnson and Kyle Wiggers and AI editor Seth Colaner — and be sure to subscribe to the AI Weekly newsletter and bookmark our AI Channel.

Thanks for reading,

Khari Johnson

Senior AI Staff Writer

Let’s block ads! (Why?)

Big Data – VentureBeat

PlaySight trained AI on thousands of hours of videos to understand sports

February 14, 2020 Big Data

Sports analytics, which refers to the use of data and statistics to measure the performance of players (or teams) and make informed coaching decisions, is an enormous market. Grand View Research pegs its worth at $ 4.6 billion by 2025, expanding at a compound annual growth rate of 31.2% from 2019.

Perhaps unsurprisingly, startups are pursuing it with gusto, and one of the pack leaders is PlaySight. The Tel Aviv-based company keeps a low profile, but it’s raised $ 26 million in capital since its founding in 2010 (and plans to raise again in the coming months) from SoftBank, Navar Corporation, Verizon Ventures, pro golf legend Greg Norman, and other backers. Moreover, it counts among its customers the NBA’s Boston Celtics, Golden State Warriors, and Toronto Raptors as well as over 80 other NCAA and NJCAA (National Junior College Athletic Association) programs and the United States Tennis Association’s National Campus.

At the 2020 OurCrowd Global Investor summit in Jerusalem this week, VentureBeat caught up with founder and CEO Chen Shachar to get the skinny on PlaySight’s technology — specifically its use of AI and machine learning.

Live broadcasting

PlaySight works with customers to build what it calls SmartCourts, connected systems consisting of cameras installed around fields, courts, gyms, and rinks. At a high level, the cameras and software layer atop them provide automated live streaming, tagging software, and predictive analytics, as well as audio annotation and drawing tools for positioning.

The technology has roots in the Israeli military — Shachar and cofounders Evgeni Khazanov and Yoram Bentur spent years developing weaponry and war simulators. The first SmartCourt was installed in 2014 and the number of installations now stands at “hundreds” of athletic centers, school gymnasiums, clubs, and federations, which use it to set goals and build regimens or choose from drills guided by coaches like Darren Cahill and Paul Annacone.

PlaySight’s eponymous PlaySight Edge service offers multi-angle video recording and ball- and player-tracking with a broadcasting interface that lets users zoom in, play in slow motion, and rewind to the beginning for instant replays. With a mobile app, they’re able to review and share sports video and data, and to create highlights with tagged and bookmarked plays uploaded to PlaySight’s sports network.

AI and machine learning

PlaySight currently supports over 30 sports, including baseball, volleyball, swimming, lacrosse, gymnastics, dancing, wrestling, and cricket. While analytics isn’t available for all of them, sports for which it is tap AI models trained on thousands of hours of video recordings.

PlaySight refers to its family of models as SmartTracker, and Shachar says they improve each game by learning the most efficient ways to track the action. Off-the-shelf cameras supplied by PlaySight capture the footage they ingest and analyze, but SmartTracker also works with existing cameras (from a single camera up to 10) that meet certain baseline requirements. There’s also a portable solution courtesy of a partnership with LiveU that can stream and record video without the need for cabling.

Subscribers to PlaySight’s SmartCourt Pro tennis plan see serves and strokes tracked and tagged by type, speed, spin, and more. Additionally, they get an overview of performance with a 3D shot map, which breaks down points with automatically generated analytics, percentages, and performance data.

Adding a new sport can take months to roughly a year, depending on the complexity of the said sport. According to Shachar, basics like player tracking are easy enough — PlaySight’s pretrained models generalize well to new playing fields and rules of play. But building something more bespoke, like a soccer model that keeps tabs on goal scoring and incurred penalties, would require more development time and effort.

And some tasks — like tracking pucks during a hockey match, when struck pucks can reach upwards of 109.2 miles per hour — are beyond the PlaySight platform’s capabilities. It’s not so much the models that are the limited factor, but the camera’s (or cameras’) frame rate. “It’s feasible to do it with higher-speed cameras, but if you’re trying to achieve a mass-market solution, it [doesn’t make sense],” said Shachar. “We’re trying to reduce the number of cameras [required] to reduce the cost of the system.

Competition

PlaySight is far from the only startup vying for share of the lucrative AI sports analytics segment, it’s worth noting.

Keemotion — whose customers include Columbia’s volleyball team, professional soccer leagues, NBA teams, and several NCAA Division I athletic programs — captures footage from multiple cameras to autonomously track balls and adjust the center of the viewport during the most exciting moments. Like PlaySight, it annotates recorded footage and streams it to handheld tablets, which coaches and support staff can use to view replays and hone in on specific points of the floor.

As for Tel Aviv, Israel-based computer vision startup Minute.ly, it offers a real-time software analysis tool that automatically divvies up livestreams into attention-grabbing, heart-thumping five-to-seven-second clips. Meanwhile, sports tech company Hudl supports video analysis workflows for youth, high school, club, and professional teams. Startup Stats captures up to 2,700 data points per game analyzed, tapping machine learning algorithms to tally up the distance players run, the trajectory and speed of balls, and teammates’ touches.

Elsewhere, for the past several years, IBM has tapped its Watson AI service to derive highlights for U.S. Open tennis matches, taking into account crowd noises, emotional player reactions, and other factors. A platform developed by Fujitsu uses sensors to track the movements of gymnasts, and to analyze those movements using AI that measures skeletal positions, speeds, and various angles. On the academic side, researchers at the Queensland University of Technology proposed last year an AI system that’s not only capable of anticipating a tennis opponent’s actions but doing so with “player-level” behavioral patterns.

But in contrast to some of its competitors, Shachar says that PlaySight’s focus has been — and remains — keeping product and service pricing low. To this end, over 20 high schools, colleges, and universities have installed PlaySight’s technology across several sports and venues, including ice hockey and soccer.

“We’re [working] hard to develop technology that can democratize sports analytics,” he said. “In the next three years, most sports facilities or the majority of sports facilities are going to be smart — they’re going to be connected. And the AI will just get better and better.”

Let’s block ads! (Why?)

Big Data – VentureBeat

Category Archives: Big Data

Self-driving car companies complain California test data may mislead

Google launches TensorFlow library for optimizing fairness constraints

Researchers propose data set to measure few-shot learning performance

Untrue News indexes and flags false and misleading news articles

Industry chief tells Facebook to adapt to EU

Force for Good Or Bad?

California’s data privacy rules get clearer

1. The scope of data & businesses subject to CCPA processes is clearer

2. We now have more details on how opt-out requests and do not track will work

3. The rules on how to provide consumer notices have new detail

What’s next for privacy compliance?

Roboflow: Popular autonomous vehicle data set contains critical flaws

AI Weekly: Machine learning could lead cybersecurity into uncharted territory

PlaySight trained AI on thousands of hours of videos to understand sports

Live broadcasting

AI and machine learning

Competition

Recent Posts

Categories

Archives