Latest News

Francois Lemarchand Francois Lemarchand

Posted on 2 November 2017

Francois shortlisted for Qualcomm AI Contest

Francois Lemarchand has recently been shortlisted for the final phase of the Qualcomm Artificial Intelligence Contest with his project of porting his aesthetic classification algorithms to mobile phones. The contest aims at developing 20 artificial intelligent mobile apps that would take full advantage of the next generation of mobile phones, ever more powerful and designed to welcome machine-learning algorithms. A brand new Snapdragon 835 Mobile Hardware Development Kit will now be sent by Qualcomm to each participant. The distributed hardware is a significant upgrade compared to today's phones and offer the opportunity to run power-greedy algorithms that would usually be processed on the cloud, necessitating an internet connection.

As for Francois' project about aesthetic classification of visual information, it is the perfect opportunity to develop an application that regroups all the different steps of building an aesthetic classification agent on the same device: taking photos, training the artificial intelligent agent, and getting user feedback. While the project has always aimed at developing an efficient aesthetic filter for photographs, being able to run such programs on a phone was out of the question at the beginning of the project a couple of years ago. In previous iterations, photos and user feedback had to be retrieved from experiments or online datasets, making the studies about visual preferences of communities, instead of one single person. At Off The Lip 2016, Francois' artificial intelligent agent challenged humans at the task of classifying photos on aesthetic criteria by matching previous judgements made by an online community. Overall, the AI performed as well as the whole group of human participants, but there were strong divergences between the AI's decisions and individual participants' decisions. Therefore, the new app will help to avoid the wisdom of the crowd phenomenon and turn the AI from learning a community's visual preferences into a personalised companion that will pick the user's best photos.