BMW Machine Learning Weekly — Week 11
May 24 — June 6, 2018
News about Machine Learning (ML), Artificial Intelligence (AI) and related research areas.
Meet Norman: The World’s First Psychopath AI
Scientists at the MIT trained an AI algorithm dubbed “Norman” to become a psychopath by only exposing it to macabre Reddit images of gruesome deaths and violence, according to a new study. Named after Anthony Perkins’ character in Alfred Hitchcock’s 1960 film Psycho, the AI was fed only a continuous stream of violent images before being tested with Rorschach inkblot tests. (Rorschach inkplot tests have the basic idea that when a person is shown an ambiguous, meaningless image, the mind will work hard at imposing meaning on the image. It is considered a personality test.) The imagery detected by Norman produced spooky interpretations of electrocutions and speeding car deaths where a standard AI would only see umbrellas and wedding cakes. The goal of the MIT researches was to prove that the method of input used to teach an ML algorithm can greatly influence its later behavior. The scientists argued that when algorithms are accused of being biased or unfair, such as the high-profile cases of Facebook news or Google Photos, “the culprit is often not the algorithm itself but the biased data that was fed into it.”
Privacy Filters for Photos
As concerns over privacy and data security on social networks grow, researchers from the University of Toronto have created an algorithm to dynamically disrupt facial recognition systems. Since privacy is a real issue at hand with facial recognition becoming increasingly better, this anti-facial-recognition system can benefit to combat that ability. Their solution leverages a deep learning technique called adversarial training, which pits two AI algorithms against each other. Two sets of neural networks were designed: the first working to identify faces, and the second working to disrupt the facial recognition task of the first. The two are constantly battling and learning from each other, setting up an ongoing AI arms race. In addition to disabling facial recognition, the new technology also disrupts image-based search, feature identification, emotion and ethnicity estimation, and all other face-based attributes that could be extracted automatically. Next, the team hopes to make the privacy filter publicly available, either via an app or a website.
AI Can Identify Wildlife as Accurately as Humans
Motion-sensor cameras are increasingly being used to track wildlife across the globe, from tigers in India to aardvarks in Africa. But combing through the millions of images captured by these systems is a time-consuming task. Now, scientists have discovered that artificial intelligence is as effective as human volunteers — and much faster — at identifying species in these largely untapped photo repositories. In a new study published this week in the Proceedings of the National Academy of Sciences, a team of researchers, led by computer scientist Mohammad Sadegh Norouzzadeh at the University of Wyoming, tested whether a type of artificial intelligence called deep neural networks could correctly identify and count species, determine animals’ ages, and classify their behaviors. They analyzed AI’s capabilities using 3.2 million images from the Snapshot Serengeti dataset, which contains photos from 225 camera traps in Tanzania’s Serengeti National Park since 2011.
Noteworthy
- Drop It Like It’s Hot
Microsoft submerged a bunch of its server machines into the sea by the coast of Scotland to keep them cool. The idea is to use the sea as a natural coolant, helping to reduce the energy needed to run the center. Read more… - Playing Piano: Machine or Human
Listen to Chopin played by a machine and by a human. - Stock Market Prediction with Natural Language Processing (NLP)
A Microsoft research team attempted to predict stock performance with NLP to interpret the earnings releases, with steps taken to purify the input by removing stop words, punctuation, and other ephemera. The model then attempted to find a relationship between the language content of the releases and the following impact on the stock price. Read more… - Applying ML to Find Your Perfect Wine
Bright Cellars, a wine store, is applying the same concept to wine as Spotify and Netflix are doing with music and movies: their wine-pairing algorithm matches members to their ideal wines by having them take a simple quiz. Read more…
