Qualcomm announce winners of the European QInF programme

Burga Tekin from EPFL, Alessandro Davide Ialongo from Cambridge University, and Bert De Brabandere from KUL have been selected as the winning students for their outstanding proposals. Each winning student will receive $40,000 as part of the fellowship along with mentoring by a Qualcomm researcher.

“This year’s proposals focused on cutting-edge areas of research including deep learning, computer vision, and robotics,” said Peter Rauber, Senior Director of engineering at Qualcomm International. “Qualcomm Research prides itself on pushing the innovation envelope and working with our ecosystem partners to spur the development of breakthrough inventions that can serve as real game changers for mobile devices. QInF continues to be an invaluable resource for finding, collaborating, and mentoring some of the world’s brightest PhD students hailing from Europe’s most prestigious technical universities.”

Europe QInF programme winners:

• Allessandro Lalongo, supervised by Professor Carl Rasmussen from the University of Cambridge, has been selected for his proposal; ‘Learning and Decision-making for Autonomous Behaviour.’ The proposal addresses the common problem where data is streamed by many inaccurate sensors. Here, learning is hindered by the inaccuracy of the data sources, and control actions based on these are therefore unlikely to be reliable. Allessandro’s research aims to solve this by modelling the latent system’s dynamics with Gaussian processes to capture complex, non-linear behaviour while quantifying and propagating uncertainty. The research will allow robots to take decisions autonomously and more reliably.
• Bert De Brabandere, supervised by Prof. Luc van Gool has been selected for his proposal; ‘Learning to Learn Visual Representations Without Supervision.’ Visual representation learning is concerned with finding good image representations that can be reused for different tasks like classification, detection and segmentation. This project strives to learn such a representation in an unsupervised way by designing local Hebbian-like learning rules that do not rely on an explicit back-propagated supervision signal to update network parameters, but instead the neurons will self-organise and adapt in a mechanistic way on the basis of locally available information like activations from neighbouring neurons.
• Bugra Tekin, supervised by Professor Pascal Fua, has been selected for his proposal; ‘Real-time 3D Human Pose Estimation in the Wild’ which focuses on real-time human pose estimation in unconstrained outdoor situations. Bugra will investigate a novel multi-stream neural network architecture to learn features related to 3D pose without requiring 3D pose annotations for training. The project will exploit readily available 3D synthetic data and 2D pose annotations to estimate 3D human pose. The methods developed could also be used for various machine learning problems with limited ground-truth data.

The six invited universities for QInF Europe all have outstanding programmes in the areas of interest. The schools are located in Belgium (KU Leuven), the Netherlands (Delft University), Switzerland (ETHZ and EPFL), and the United Kingdom (Imperial College and Cambridge University).

Charles Bergan, Vice President of Engineering at Qualcomm Technologies, said: “This year, we received a record number of top-notch proposals that warranted serious consideration. Each delivered fresh, innovative, and leading edge ideas consistent with Qualcomm Research’s high standard. Narrowing the selection process down to eight finalists, and later three winners did require significant deliberation among the reviewers and judges. Ultimately, we chose Burga Tekin, Alessandro Davide Ialongo, and Bert De Brabandere for their groundbreaking work. We are pleased to congratulate them on this accomplishment and look forward to working with them, turning their concepts into reality.”

Charles participated for a sixth time in this year’s QInF Europe judging panel, along with the head of Qualcomm’s European research offices, Peter Rauber. Additionally, five members of Qualcomm Research’s Amsterdam team also served on the panel. 

The QInF Europe finalist event was hosted by QUVA, the joint research lab, announced by Qualcomm Technologies and the University of Amsterdam, focused on advancing machine learning techniques for computer vision. One of the three QUVA directors, Professor Max Welling, provided an interesting keynote speech on ‘Bayesian extensions to deep learning with an application to compression.’