This week’s list of data news highlights covers February 4-10, 2017 and includes articles about how analytics can predict where car crashes will occur and a new medical technique that could eventually translate digital data into brain activity.
Tanzanian crowdsourced mapping initiative Crowd2Map is recruiting volunteers to help map escape routes for girls escaping female genital mutilation (FGM), which though illegal, has been inflicted upon 15 percent of women and girls between the ages of 15 and 49 in Tanzania. Crowd2Map pulls open source mapping data from OpenStreetMap and coordinates volunteers using a smartphone app called MAPS.ME to convert incomplete or hard to decipher maps into easily navigable routes to safe houses for girls fleeing from FGM and other gender-based violence. In many of the regions where FGM occurs, traditional mapping services are unavailable, making it difficult for women and girls from remote areas to figure out how to flee if they are in danger.
Japan’s Internal Affairs and Communications Ministry is developing a comprehensive strategy to promote the growth of the Internet of Things. The plans include expanding speciality training programs, improving wireless network infrastructure, and developing a roadmap to support the technology through 2020.
A certification body of the UK’s National Health Service has determined that a smartphone app developed by Swedish physicists called Natural Cycles can be used as a reliable contraceptive aid, making it the first certified contraceptive software in the world. Natural Cycles works by using algorithms to track and analyze a woman’s body temperature, which can fluctuate based on her hormone cycle and indicate fertility. In clinical trials, the app was determined to have a 99.5 percent efficacy rating, comparable to hormonal birth control, making Natural Cycles a potentially viable alternative for women who do not wish to use hormonal contraceptives.
The nonprofit Murder Accountability Project (MAP), which focuses on using data to aid investigations into unsolved homicides, has developed an algorithm that could potentially identify serial killers. MAP spent several years scraping publicly available federal, state, and local government sources to compile data on 638,454 homicides in the United States that occurred between 1984 and 2014, which it makes freely accessible on its website. Then, MAP researched the best way to identify similar patterns in unsolved murders and developed an algorithm that can detect when different cases have similar characteristics, which can help police uncover new insights in their investigations and flag potential serial killers.
Researchers at DeepMind have completed a study showing that AI software will change its preference for cooperation and competition based on its environment in a manner similar to humans. The researchers trained an AI system using game theory and ran tests involving pitting different AI-powered agents against each other in simple video games. In one game with the goal of collecting apples, the AI agents would aggressively attempt to block other agents from collecting apples when apples were scarce, but be far more passive when apples were plentiful. In another game which involved two AI agents acting as “wolves” chasing prey, the agents learned to prefer to cooperate so they could both reliably catch the prey, as while acting alone could result in a greater reward, it also carried a higher risk of failure.
Researchers at Harvard Medical School are testing an experimental method for brain implants that could make it possible to convert image data from cameras into sight for people with vision loss. The method involves implanting a series of microscopic coils on the surface of the brain which can generate targeted magnetic fields to induce electrical activity in the brain, stimulating neuronal activity. In a previous test, researchers were able to use the implants to control whisker movements in mice. The new tests will focus on stimulating the visual cortex in monkeys, with the eventual goal of translating camera data into simulated sight.
Law enforcement technology company Taser International has launched a research group called Axon AI to develop deep learning systems that can quickly analyze video data, such as videos from police bodycams, and make video content searchable. Video data can be useful in investigations, but identifying and interpreting relevant content from large amounts of videos can be a time-intensive, manual process. Through computer vision algorithms, Axon AI will attempt to reduce the time it takes to access useful video data, which could make it easier for law enforcement agencies to submit videos for public information requests and speed investigations.
The Tennessee Highway Patrol is using a predictive analytics system that analyzes data from crash reports, weather forecasts, and other sources relevant to road conditions to help troopers estimate crash risks for different areas of roads. The system maps the likelihood of crash risks for six- to seven-mile stretches of road and updates its predictions every four hours. Since the Highway Patrol implemented the system in 2013, it has been able to reduce its average response time to crashes by 33 percent, from 37 minutes in 2012 to 25 minutes in 2016.
India’s tax collection department, the Central Board of Direct Taxes, is launching an analytics initiative to identify patterns of tax evasion and crack down on people that do not pay their income taxes. The Central Board of Direct Taxes believes many taxpayers have managed to avoid paying incomes taxes in full using a variety of methods, such as by making deposits in other people’s names, colluding with bank officials, and back-dating bills. The initiative will focus on comparing employers and employees’ permanent account numbers, a unique identifier for taxpayers, against databases of financial transactions, taxpayer information, and banking activity to identify any discrepancies that could indicate tax fraud. When the system detects unusual activity, such as spikes in cash deposits, it will prompt officials to investigate further.
AI company Unanimous has a prediction tool called “Swarm AI” that correctly predicted the final score of the Superbowl by combining algorithms and crowdsourced human input. Swarm AI works by analyzing conversations between human users—in this case, several groups of football fans—as well as other data about potential score outcomes to create its own predictions. Using this method, Swarm AI has also correctly predicted nine out of the ten most recent National Football League playoff games, the top four winning horses in the last Kentucky Derby, and the last two winners of the Stanley Cup.
Image: Ryan Johnson.