By completing a jig-saw-puzzle during the official opening on September 15, Senior Vice President Nagraj Kashyap and President Dymph van den Boom marked the start of the new public-private partnership between Qualcomm and the University of Amsterdam: the QUVA-lab.

The mission of the QUVA-lab is to perform world-class research on deep vision. Such vision strives to automatically interpret with the aid of deep learning what happens where, when and why in images and video. Deep learning is a form of machine learning with neural networks, loosely inspired by how neurons process information in the brain. Research projects in the lab will focus on learning to recognize objects in images from a single example, personalized event detection and summarization in video, and privacy preserving deep learning. The research will be published in the best academic venues and secured in patents.

One billion processors

Qualcomm is the world-leading provider of processer and radio technology for mobile devices and especially smartphones. Qualcomm ships over one billion processors annually, which includes wireless radio processors (2G/3G/4G, WIFI, Bluetooth) as well as the CPUs, GPUs and DSPs that integrate into the Qualcomm Snapdragon System-on-Chip. Bringing computer vision together with machine learning—with an emphasis on mobile and embedded use cases—will foster new approaches to more intelligence in smartphone cameras, robotics, automotive and Internet of Everything applications.

The establishment of the QUVA-lab was motivated by Qualcomm’s acquisition of Euvision Technologies during the summer of 2014. The spin-off of the UvA-lab of Arnold Smeulders, at the University of Amsterdam, became Qualcomm Research Netherlands. Qualcomm Research Netherlands conducts R&D in computer vision by learning. QUVA-lab, which will form part of the UvA’s Informatics Institute at Amsterdam Science Park, is a further extension of the academic research in computer vision and deep learning.

15 to 20 researchers

The agreement has been made for a period of five years and will involve between 15 to 20 researchers. The research lab’s management team will be comprised of Max Welling, professor of Machine Learning at the UvA, Cees Snoek, principal engineer at Qualcomm and associate professor at the UvA, and Arnold Smeulders. The QUVA-lab is located in the vicinity of the Amsterdam office of Qualcomm Research, which will facilitate further collaboration opportunities.

Deep learning

A deep neural network receives data for instance in the form of an image. It scans this image with small templates (e.g. parts of objects) which generate “feature maps” after one layer of processing. Features represent the activities of neurons that get activated if a template finds a match somewhere in the image. These fields of activities are then scanned again for suspicious correlations by new templates in the second layer of the network resulting in new feature maps etc. Where neurons in lower layers usually search for edges, neurons higher up in the hierarchy become sensitive to more abstract concepts covering a larger area of the image, such as an entire face. The final layers are trained to detect and classify the objects present in the image.

In recent years new deep learning algorithms have seen the light of day but their backbone is still an algorithm invented in the eighties: error backpropagation. In contrast to the feedforward processing of information necessary to detect and classify objects in images, the learning updates that tune the parameters of the network run backward through the network, computing how each synapse needs to change in order to make the network perform better. This is iterated millions of times until the updates converge. Yoshua Bengio, one of the founding fathers of the field delivered a keynote on deep learning during the opening of the QUVA-lab.