This week’s list of data news highlights covers July 22 – 28, 2017, and includes articles about an app store for genomic data and an effort to use the blockchain to make securities data more transparent.
DeepMind is developing a method to develop AI systems that can imagine the outcomes of their actions before they make them, which could help them perform complex tasks more effectively. The approach relies on neural networks called imagination-augmented agents (I2As), which can learn to identify information that could be useful in future decision-making and develop an array of strategies for future problem solving based on this data. In a test playing simple video games, I2A systems were able to considerably outperform standard methods while relying on less training.
2. Teaching Software to See Body Language
Researchers at Carnegie Mellon University have developed a computer vision system called OpenPose capable of tracking body, hand, and face movements in video in real time. Unlike most other body-tracking systems, OpenPose can distinguish between individual fingers, and thanks to the researchers’ training OpenPose to identify a particular pose from multiple angles, it is capable of interpreting 3D poses with just a single camera.
3. There’s Now an App Store for Your Genome
Genomics company Helix has launched a platform that allows users to buy genetic analysis services from different vendors. Users first purchase Helix’s sequencing service and then can buy additional analysis services in categories including health, nutrition, and ancestry to learn about their genetic predispositions to a variety of health and lifestyle factors, as well as their ethnic backgrounds.
4. Protecting the Coast Guard from Hoaxes with AI
The U.S. Coast Guard is using AI voice analysis software developed by researchers at Carnegie Mellon University to identify fake distress calls. The software can analyze a caller’s voice and determine the environment a person is calling from, such as inside a car or house, physical characteristics about the callers, and other factors that can help the Coast Guard determine if a caller is being truthful. Though fake calls only make up about one percent of all distress calls the Coast Guard receives per year, the Coast Guard still devotes a large amount of time and money to every call, such as launching search and rescue missions, and fake calls can strain its resources.
5. Building a Blockchain for Securities Data
Borsa Italiana, a subsidiary of the London Stock Exchange, has partnered with IBM to develop a blockchain system to digitize securities certificate data, which currently uses a paper-based system. By replacing paper securities information with a digital distributed ledger, securities transactions can be much more transparent as well as more efficient, as firms can automate the processing of contracts.
6. Fighting Insurance Fraud with Predictive Analytics
Insurance and roadside assistance company AA Ireland has developed a predictive analytics system to help identify fraudulent insurance claims. The system uses a technique called fuzzy matching that analyzes records of price quotes to identify customers that may have requested a quote from AA Ireland multiple times using slight variations in their personal information to try to obtain a better price. AA Ireland claims that the system will pay for itself within 18 months by reducing the losses to insurance fraud, which it says adds €50 to every €700 policy.
7. Growing Better Crops with AI
Farming company NatureSweet is using AI software to help control pests and protect against diseases on its tomato farms, which has improved weekly yields by two to four percent in tests. The system analyses a constant stream of photos from cameras inside NatureSweet’s greenhouses to detect any changes in plant health and flag when a potential infestation or disease appears. The software allows NatureSweet to constantly monitor plant health and is much more efficient than its previous approach, which required human employees to walk through greenhouses to observe the plants on a regular basis.
8. Using AI to Predict Schizophrenia Symptoms
Researchers at IBM and the University of Alberta have developed an AI system that can analyze functional magnetic resonance imaging (fMRI) scans of patients’ brains to identify if a patient has schizophrenia with 74 percent accuracy. Additionally, the system can analyze fMRI scans of a schizophrenia patient and identify brain network structures that could help determine how severe the patient’s symptoms might be, such as concentration disorders and erratic behavior.
9. Teaming Up to Protect Song Ownership Rights
Performing rights organizations Broadcast Music, Inc. (BMI) and the American Society of Composers, Authors, and Publishers (ASCAP) have announced they will develop a shared database of all of their musical works in their repertoires which can serve as a public authoritative source for music ownership rights. The shared database will make it easier for people to identify ownership rights for songs and facilitate licensing.
10. AI Turns 2D Images into 3D Shapes
Researchers at Purdue University have developed a machine learning system called SurfNet (short for “surface net”) capable of reconstructing a 2D image as a 3D model. The researchers trained the system on pairs of 2D and 3D images so it could learn how 3D objects are represented in two dimensions and generate a 3D representation using this knowledge. This approach could eventually help create 3D content for virtual environments without the need to develop 3D models.
Image: Mike Baird.
