Daily Archives: June 8, 2017

Artificial intelligence researchers at Facebook have figured out how to train their AI models for image recognition at eye-popping speeds.

The company announced the results of the effort to speed up training time at the Data@Scale event in Seattle this morning. Using Facebooks custom GPU (graphics processing unit) hardware and some new algorithms, researchers were able to train their models on 40,000 images a second, making it possible to get through the ImageNet dataset in under an hour with no loss of accuracy, said Pieter Noordhuis, a software engineer at Facebook.

You dont need a proper supercomputer to replicate these results, Noordhuis said.

The system works to associate images with words, which is called supervised learning, he said. Thousands of images from a training set are assigned a description (say, a cat) and the system is shown all of the images with an associated classification. Then, researchers present the system with images of the same object (say, a cat) but without the description attached. If the system knows its looking at a cat, its learning how to associate imagery with descriptive words.

The breakthrough allows Facebook AI researchers to start working on even bigger datasets; like, say, the billions of things posted to its website every day. Its also a display of Facebooks hardware expertise; the company made sure to note that its hardware is open-source, this means that for others to reap these benefits, theres no need for incredibly advanced TPUs, it said in a statement throwing some shade at Googles recent TPU announcement at Google I/O.

Facebook plans to release more details about its AI training work in a research paper published to its Facebook Research page.

Continued here:

Facebook's AI training models can now process 40000 images a second - GeekWire

June 8, 2017 by Joe Kullman ASU Professor Subbarao Kambhampati with one of the robots used in his lab teams research aimed at enabling effective collaboration between humans and intelligent robots. The wooden blocks spell out the name of the lab, Yochan, meaning thought or plan in the Sanskrit language. Credit: Marco-Alexis Chaira/ASU

Interest in artificial intelligence has exploded, with some predicting that machines will take over and others optimistically hoping that people will be freed up to explore creative pursuits.

According to Arizona State University Professor Subbarao Kambhampati, the reality will be more in the middlebut the technology will certainly bring about a restructuring of our society.

AI will accomplish a lot of good things, Kambhampati said, but we must also be vigilant about possible ramifications of the technology. And yes, some jobs will be lostbut maybe not the ones people most often think of.

The professor of computer science and engineering in ASU's Ira A. Fulton Schools of Engineering is well qualified to enter the debate. He has been doing work in the areacommonly called "AI"for more than three decades, and he is at the midpoint of a two-year term as president of the international Association for the Advancement of Artificial Intelligence (AAAI), the largest organization of scientists, engineers and others in the field.

Kambhampati, whose current research focuses on developing "human-aware" AI systems to enable people and intelligent machines to work collaboratively, is also on the board of trustees of the Partnership on Artificial Intelligence to Benefit People and Society (PAI), which aims to help establish industry-wide best practices and ethics guidelines.

The following interview is edited from a recent conversation with him.

Question: You became president of the AI association at a time when public awareness of these technologies and the issues they raise has exploded. What's sparking the widespread interest?

Answer: AI as a scientific field has actually been around since the 1950s and has made amazing, if fitful, progress in getting machines to show hallmarks of intelligence. The Deep Blue computer's win over the world chess champion in 1997 was a watershed moment, but even after that, AI remained a staid academic field. Most people didn't come into direct contact with AI technology until relatively recently.

With the recent advances of AI in perceptual intelligence, we all now have smartphones that can hear and talk back to us and recognize images. AI is now a very ubiquitous part of our everyday lives, so there's a visceral understanding of its impact.

Q: Plus, it's a big driver of major industries, right?

A: In 2008, for instance, few if any tech companies were mentioning investments and involvement in AI in their annual reports or quarterly earnings reports. Today you'll find about 300 major companies emphasizing their AI projects or ventures in those reports.

The members of the Partnership for Artificial Intelligence, which I am involved with, include Amazon, Facebook, Google's Deep Mind, IBM and Microsoft. So, yes, AI is now a very big deal.

Q: The big question about AI is what it means for not only business and the economy, but what it portends for society when AI machines are doing more jobs that people used to do. What's your perspective on that?

A: Elon Musk (the prominent engineer, inventor and tech entrepreneur) started this trend of AI fears by remarking that what keeps him up at night is the idea of super-intelligent machines that will become more powerful than humans. Then Stephen Hawking (renowned physicist and cosmologist) chimed in. Statements like that, coming from influential people, of course make the public worry.

I don't take such a pessimistic view. I think AI is going to do a lot of good things. But it is also going to be a very powerful technology that will shape and change our world. So we should remain vigilant of all the ramifications of this powerful technology and work to mitigate unintended consequences. Fortunately, this is a goal shared by both AAAI and PAI.

Q: Garry Kasparov, the former chess champion who was defeated by the Deep Blue computer, writes that we should embrace AI, that it will free people from work so that they can develop their intellectual and creative capabilities. Others are saying the same. Do you agree?

A: I think Kasparov and others who say this are maybe too optimistic. We see from the past that new technology has taken away certain jobs but also created new kinds of jobs. But it's not certain that will always be the case with the proliferation of AI.

It seems clear that some professions are going to disappear, and not just blue-collar jobs like trucking, but also high-paying white-collar jobs. There are going to be many fewer radiologists, because machines are already doing a better job of reading X-rays. Machines can also be much faster and better at doing the kind of information gathering and research now done by paralegals, for instance.

This is why we have to start thinking about how society is going to be restructured if AI technologies and systems are doing much of the work that people once did.

Q: What would such a restructuring look like?

A: This is quite an open question, and organizations like AAAI and PAI are trying to get ahead of the curve in answering it.

I do want to emphasize that I don't think it is solely the job of AI experts, or of industry, to think about these issues of long-term restructuring. This is something that society at large has to contend with. We also have to realize that AI consequences play into already existing social ills such as societal biases, wealth concentration and social alienation. We have to work to make sure that AI moderates rather than amplifies these trends.

Q: What can those in the AI field do proactively to produce the most positive outcomes from the expansion of the technology?

A: We can take potential impacts into consideration when deciding in what directions we want to take our research and development. Much research now, like mine, is focusing on systems that are not intended to replace humans but to augment and enhance what humans are doing. We want to enable humans and machines to work together to do things better than what humans can do alone.

For AI systems to work with humans, they need to acquire emotional and social intelligence, something humans expect from their co-workers. That's where human-aware AI comes into play.

Q: What keeps you excited about your research?

A: I've always thought that the biggest questions facing our age are about three fundamental things: the origin of the universe, the origin of life and the nature of intelligence.

AI research takes you to the heart of one of them. In developing AI systems, I get a window into the basic nature of intelligence. That's why I tell my students that it takes a particularly bad teacher to make AI uninteresting.

That is what hooked me into this work. And now I'm getting the opportunity to go beyond the technical aspects of the field and have a voice on issues of ethics and practices and societal outcomes. That is energizing me even more.

Explore further: AI 'good for the world'... says ultra-lifelike robot

Sophia smiles mischievously, bats her eyelids and tells a joke. Without the mess of cables that make up the back of her head, you could almost mistake her for a human.

Major technology firms have joined forces in a partnership on artificial intelligence, aiming to cooperate on "best practices" on using the technology "to benefit people and society."

Advances in artificial intelligence will soon lead to robots that are capable of nearly everything humans do, threatening tens of millions of jobs in the coming 30 years, experts warned Saturday.

A technology industry alliance devoted to making sure smart machines don't turn against humanity said Friday that Apple has signed on and will have a seat on the board.

The phrase "artificial intelligence" saturates Hollywood dramas from computers taking over spaceships, to sentient robots overpowering humans. Though the real world is perhaps more boring than Hollywood, artificial intelligence ...

Microsoft chief executive Satya Nadella said Wednesday tech developers have a responsibility to prevent a dystopian "1984" future as the US technology titan unveiled a fresh initiative to bring artificial intelligence into ...

An AI machine has taken the maths section of China's annual university entrance exam, finishing it faster than students but with a below average grade.

Globally, from China and Germany to the United States, electric vehicle (EV) subsidies have been championed as an effective strategy to boost production of renewable technology and reduce greenhouse gas emissions (GHG).

As global automakers compete to bring the first flying car to market, Czech pilot Pavel Brezina is trying a different tack: instead of creating a car that flies, he has made a "GyroDrive"a mini helicopter you can drive.

Apple's new HomePod speaker may be music to the ears of its loyal fans, but how much it can crank up volume in the smart speaker market remains to be heard.

Autonomous vehicles with no human backup will be put to the test on publicly traveled roads as early as next year in what may be the first attempt at unassisted autonomous piloting.

Using Earth-abundant materials, EPFL scientists have built the first low-cost system for splitting CO2 into CO, a reaction necessary for turning renewable energy into fuel.

Please sign in to add a comment. Registration is free, and takes less than a minute. Read more

See original here:

Artificial intelligence's potential impacts raise promising possibilities, societal challenges - Phys.Org

Quantitative analytical procedures are some of the most successful in the financial world, with an increasing number of money managers turning the grunt work of data processing over to computer algorithms and artificial intelligence (AI). One argument in favor of quantitative methods like these is that they remove the human element from the analytical process, thereby ensuring faster processing times, more thorough analysis, and the effective removal of emotions and potential bias from the process. Now, at least one company is looking to capitalize on the advantages that quantitative methods have over old-fashioned human ones, but in a new area: sports betting.

The new company, Stratagem, is based in London and was set up by an ex-hedge funder, Andreas Koukorinis. In an interview with Business Insider, Koukorinis described his initial efforts at harnessing the powers of quantitative analysis for the purposes of sports betting as "building these robots to let them run around on the floor." He and his team have been developing predictive analytics programs for sports betting procedures, using machine learning and AI to process vast data fields. With these computer systems in place, Koukorinis believes that he will gain an edge in the competitive and often-arbitrary world of sports betting.

Koukorinis has been developing Stratagem for several years, and the company now appears to be taking off. The fund has seen some success with its machine learning models, and Stratagem now has an internal syndicate which allows it to bet its own money and bring in a return. One of the next steps for the fund is to raise around 25 million in the next few months to allow for further growth. Investors will essentially be buying into a sports betting-focused hedge fund.

Charles McGarraugh, CEO of the fledgling company, believes that the model is a straightforward sell to potential investors. "Sports lend themselves well to this kind of predictive analytics because it's a large number of repeated events. And it's uncorrelated to the rest of the market. And the duration of the asset class is short."

Stratagem focuses on both data collection and processing. For the former, the company uses both public sources as well as its own data generation system. Once the data has been gathered, Stratagem uses its analytical tools to crunch the numbers in search of mispriced odds. The results so far have been promising.

Could this be the future of quant methods? Koukorinis and others with Stratagem believe so, seeing a strong connection between the world of sports betting and the hard data analysis that quant is specially designed for. Whether the company will beat the odds remains to be seen.

Continue reading here:

Sports Betting: The Next Big Thing for Artificial Intelligence - Investopedia

BEIJING: An artificial intelligence (AI) machine has taken the maths section of China's annual university entrance exam, finishing it faster than students but with a below average grade.

The artificial intelligence machine -- a tall black box containing 11 servers placed in the centre of a test room -- took two versions of the exam on Wednesday in Chengdu, Sichuan province.

The machine, called AI-MATHS, scored 105 out of 150 in 22 minutes. Students have two hours to complete the test, the official Xinhua news agency reported.

It then spent 10 minutes on another version and scored 100.

Beijing liberal art students who took the maths exam last year scored an average of 109.

Exam questions and the AI machine's answers were both shown on a big screen while three people kept score.

The AI was developed in 2014 by a Chengdu-based company, Zhunxingyunxue Technology, using big data, artificial intelligence and natural language recognition technologies from Tsinghua University.

"I hope next year the machine can improve its performance on logical reasoning and computer algorithms and score over 130," Lin Hui, the company's CEO, was quoted as saying by Xinhua.

"This is not a make-or-break test for a robot. The aim is to train artificial intelligence to learn the way humans reason and deal with numbers," Lin said.

The machine took only one of the four subjects in the crucially important entrance examination, the other three being Chinese, a foreign language and one comprehensive test in either liberal arts or science.

While AI is faster with numbers than humans, it struggles with language.

"For example, the robot had a hard time understanding the words 'students' and 'teachers' on the test and failed to understand the question, so it scored zero for that question," Lin said.

The test was the latest attempt to show how AI technology can perform in comparison to the human brain.

Last year, the Google-owned computer algorithm AlphaGo became the first computer programme to beat an elite player in a full match of the ancient Chinese game of Go.

AlphaGo won again last month, crushing the world's top player, Ke Jie of China, in a three-game sweep.

AlphaGo's feats have fuelled visions of AI that can not only perform pre-programmed tasks, but help humanity look at complex scientific, technical and medical mysteries in new ways.

See the original post:

Artificial Intelligence gets below average grade in Chinese university entrance exam - Economic Times

This AI will tell people when theyre likely to die -- and thats a good thing. Thats because scientists from the University of Adelaide in Australia have used deep learning technology to analyze the computerized tomography (CT) scans of patient organs, in what could one day serve as an early warning system to catch heart disease, cancer, and other diseases early so that intervention can take place.

Using a dataset of historical CT scans, and excluding other predictive factors like age, the system developed by the team was able to predict whether patients would die within five years around 70 percent of the time. The work was described in an article published in the journal Scientific Reports.

The goal of the research isn't really to predict death, but to produce a more accurate measurement of health, Dr. Luke Oakden-Rayner, a researcher on the project, told Digital Trends. A patient's risk of death is directly related to the health of their organs and tissues, but the changes of chronic diseases build up for decades before we get symptoms. By the time we recognize a disease is present it is often quite advanced. So we can take a known outcome, like death, and look back in time at the patient's medical scans to find patterns that relate to undetected disease. Our goal is to identify these changes earlier and more accurately so we can tailor our treatment to individuals.

The AI analyzes CT scans to make its decisions.

At present, this is still a proof-of-concept experiment, however, and Oakden-Rayner points out that theres a lot more work to be done before this becomes the transformative clinical tool it could be. For one thing, the AIs 70-percent predictive accuracy when looking at scans is in line with the manual predictions made by experts. That makes it a potential time-saving tool, or a good means of double-checking, but the hope is that it can be much more than that.

Our next major step is to expand our dataset, Oakden-Rayner continued. We used a very small cohort of 48 patients in this study to show that our approach can work, but in general deep learning works better if you can give it much more data. We are collecting and analyzing a dataset of tens of thousands of cases in the next stage of our project.

The team also aims to expand what the AI is looking for, to help spot things like strokes before they strike.

Follow this link:

Artificial intelligence can now predict if someone will die in the next 5 years - Fox News