AsianScientist (Dec. 8, 2018) – A research team from the RIKEN Center for Advanced Intelligence Project (AIP) has developed a method of machine learning that allows artificial intelligence (AI) to make classifications without negative data. They presented their work at the 2018 Neural Information Processing Systems Conference.
Humans classify things all the time. For example, we are able to distinguish between objects and faces. For AI to do the same, ‘classification technology’ is necessary, which means that the computer must be trained to form a boundary that separates positive and negative data. Positive data could consist of photos of happy faces, while negative data could comprise photos of sad faces.
However, in some contexts, negative data is hard to find, or less abundant. For instance, there may be fewer photos of sad faces since most people smile in front of a camera. The lack of negative data could therefore hinder the computer from forming a clear classification boundary.
In the present study, researchers led by Dr. Takashi Ishida of RIKEN AIP reduced the reliance on negative data by adding a confidence score to positive data. The score mathematically corresponds to the probability of whether a data point belongs to a positive class or not.
“Previous classification methods could not cope with the situation where negative data were not available, but we have made it possible for computers to learn with only positive data, as long as we have a confidence score for our positive data,” Ishida explained.
To test how well the system functioned, the researchers applied it to a set of photos that contained various apparel. For example, they chose T-shirts as the positive class and sandals as the negative class. They then attached a confidence score to photos of T-shirts. They found that the AI could form a classification boundary even without accessing the negative data, or the photos of sandals.
“This discovery could expand the range of applications where classification technology can be used. In the near future, we hope to put our technology to use in various research fields, such as natural language processing, computer vision, robotics and bioinformatics,” said Ishida.
The article can be found at: Ishida et al. (2018) Binary Classification from Positive-Confidence Data.
Disclaimer: This article does not necessarily reflect the views of AsianScientist or its staff.