27/12/2016 2:55 PM IST | Updated 27/12/2016 2:57 PM IST

Apple Talks About Image Recognition In Its First Public AI Paper

The paper lines out how to teach a system to identify real images.


NEW YORK -- Breaking with its tradition of keeping research topics a secret, Apple has come out in the open, publishing its first artificial intelligence (AI) research paper that focuses on advanced image recognition.

Apple's first public research paper on AI was penned by vision expert Ashish Shrivastava and a team of engineers including Tomas Pfister, Oncel Tuzel, Wenda Wang, Russ Webb and Apple Director of Artificial Intelligence Research Josh Susskind, reported on Tuesday.

Shrivastava holds a PhD in computer vision from the University of Maryland.

Titled 'Learning from Simulated and Unsupervised Images through Adversarial Training', the paper describes techniques of training computer vision algorithms to recognise objects using synthetic, or computer generated, images.

However, learning from synthetic images may not achieve the desired performance owing to a gap between synthetic and real image distributions.

To reduce this gap, Apple has proposed Simulated plus Unsupervised (S+U) learning, where the task is to learn a model to improve the realism of a simulator's output using unlabeled real data while preserving the annotation information from the simulator.

The company has developed a method for S+U learning that uses an adversarial network similar to Generative Adversarial Networks (GANs), but with synthetic images as inputs instead of random vectors.

The improved realism enables the training of better machine-learning models on large datasets without any data collection or human annotation effort.

Moreover, since machine learning models can be sensitive to artifacts in the synthetic data, S+U learning should generate images without artifacts.