Post has attachment
The geographic location of a quarter billion images from news articles have been determined from the images alone (no geotags) using machine learning. It's part of the Global Database of Events, Language, and Tone (GDELT) project. "One might naturally ask why visual geocoding is required when modern images typically contain a wealth of embedded information via hidden metadata fields like EXIF. The answer is that when looking globally, just 0.26% of contemporary news images contain GPS coordinates and when processing historical images, such as from library archives, none of the imagery contains such enrichments. Image captions are also far less useful than one might expect when it comes to identifying location. Photographs appearing in news coverage may not contain a caption at all (and those appearing in library archives typically do not), while for those images that do have captions, often the captions describe the contents of the image but do not offer insights as to its location."

"Over the course of 2016 the GDELT Project processed more than 234 million images from online news outlets throughout the world through the Google Cloud Vision API, generating more than 1.8TB of JSON data describing them in detail. Of those quarter billion images, the Vision API flagged 3,468,424 of them (1.5%) as depicting a precise geographic location it was able to confidently identify, featuring 101,090 distinct locations on earth."

Post has attachment
Google Cloud Platform has just made GPUs, specifically Nvidia Tesla K80 GPUs, available for machine learning. "You can now spin up NVIDIA GPU-based VMs in three GCP regions: us-east1, asia-east1 and europe-west1, using the gcloud command-line tool. Support for creating GPU VMs using the Cloud Console appears next week. If you need extra computational power for deep learning, you can attach up to eight GPUs (4 K80 boards) to any custom Google Compute Engine virtual machine."

Post has attachment
"As Ford Motor Co. has been developing self-driving cars, the US automaker has started noticing a problem during test drives: Engineers monitoring the robot rides are dozing off."

"These are trained engineers who are there to observe what's happening. But it's human nature that you start trusting the vehicle more and more and that you feel you don't need to be paying attention."

"The struggle to prevent snoozing-while-cruising has yielded a radical decision: Ford will venture to take the human out of the loop by removing the steering wheel, brake and gas pedals from its driverless cars debuting in 2021."

Post has attachment
Amazon has started a blog for AI on Amazon Web Services (AWS).

Post has attachment
AI will make life meaningless, says Elon Musk. Or at least that's the headline. But, there's a video, which I was able to track down. At about 52:00 in this video:

https://www.youtube.com/watch?v=jBuLgBX2bKQ

Elon Musk says, "How are people going to have meaning? A lot of people derive their meaning from their employment. So if you don't have... if you're not needed, if there isn't a need for your labor, how do you... where's the meaning? Do you have meaning? Do you feel useless?"

He seems to not so much asserting that "AI will make life meaningless" but pondering out loud.

The part of the video with Elon Musk starts at 27:00 and there's other interesting bits in it. Unfortunately, there is a lot of background noise in the sound. Anyway, he predicts in 10 years, all cars sold will have full autonomy. Today, Tesla cars have all the sensors needed for full autonomy, and probably enough compute power to be safer than a human. So it's mostly about the software. Asked about traffic congestion, he says the solution is to build tunnels, and to do them in a "3D" way -- i.e. have 50 levels of tunnels. Traffic congestion happens because streets are 2D and people go between buildings all at once. It's a good example of Musk's "reasoning from first principles." The question, which he brings up, is whether such a thing can be done cheaply enough, fast enough, and with a high margin of safety.

He predicts with the adoption of electric cars, demand for electricity will triple.

After talking about AI and employment, he goes on to merging biological and digital intelligence. He proposes that a "high-bandwidth connection to the brain will solve the control problem and the usefulness problem." (He doesn't explain the 'control problem' here, but has elsewhere -- it is the problem of AI escaping human control.)

He advises young people to learn physics, because the way of thinking in physics is the best way to learn to think about things that are counterintuitive. And you have to think that you are to some degree wrong, and the goal is to be less wrong over time. You learn to do less wishful thinking, where you ignore the real truth, in favor of what you want to be true.

Post has attachment
"Artificial intelligence grows a nose." "Participants were given the volunteer ratings for two-thirds of the odors, along with the chemical structure of the molecules that produced them. They were also given more than 4800 descriptors for each molecule, such as the atoms included, their arrangement, and geometry, which constituted a separate set of more than 2 million data points. These data were then used to train their computer models in predicting smells from chemical structural information."

"We learned that we can very specifically assign structural features to descriptions of the odor,' Meyer says. For example, molecules with sulfur groups tend to produce a 'garlicky' smell, and molecules with a similar chemical structure to vanillin, from vanilla beans, predicts whether subjects will perceive a 'bakery' smell."

Post has attachment
MD Anderson, the cancer center that is part of the University of Texas that started using the IBM Watson cognitive computing system for its cancer research in 2013 has put the project on hold. "MD Anderson is actively requesting bids from other contractors who might replace IBM in future efforts. And a scathing report from auditors at the University of Texas says the project cost MD Anderson more than $62 million and yet did not meet its goals."

"As a public institution, we decided to go out to the marketplace for competitive bids to see where the industry has progressed."

Post has attachment
Air force robotic snake for aircraft inspections. "Remote Access Nondestructive Evaluation (RANDE) is a flexible, robotic snake-arm tool that can reach into confined areas to perform required inspections, or simply look into tight spaces."

"Typically, when military depot or field personnel perform routine inspections on hard-to-reach components such as the interior of aircraft wings, they first have to remove the wing, then remove additional structure within the wing so that inspectors can reach in with specialized equipment. With RANDE, the need to remove the wing for inspection can be eliminated. Instead, maintainers only need to remove the necessary external access panels and maneuver the snake arm through an access hole as small as three inches in diameter."

Post has attachment
List of about 30 books on machine learning available on the internet for free (center of page).

Post has attachment
"Inside Wikipedia's attempt to use artificial intelligence to combat harassment." "They started with 100,000 comments from Wikipedia talk pages, where editors hash out their disagreements. Next, 4,000 crowdworkers evaluated the comments for personal attacks. Each comment was inspected by 10 different people. After being trained on the dataset, the algorithm could determine the probability of a given comment being a personal attack as reliably as a team of three human moderators."

"The Detox team then ran 63 million English Wikipedia comments posted between 2001 to 2015 through the algorithm and analyzed the results for patterns in abusive comments."

"Although comments from unregistered users were six times more likely to contain an attack, more than half of all abusive comments came from registered, identifiable users. What's more, the abuse wasn't coming from an isolated group of trolls. Almost 80 per cent of all abusive comments were made by over 9000 'low-toxicity users' -- people who made less than 5 abusive comments in a year. On the flip side, nearly 10 percent of all attacks on the platform were made by just 34 highly toxic users."
Wait while more posts are being loaded