Ilan Price

Learn more
A photograph of Ilan smiling at the camera

Former position

Doctoral Student

Cohort year

2018

Partner Institution

University of Oxford
Menu

Research areas

Information theory (Statistical methods & theory) Neural networks Machine learning Supervised learning Deep learning

Bio

Ilan Price is a Research Scientist at Google DeepMind, where his current work is focused on using machine learning to advance probabilistic weather forecasting. 

Ilan also recently submitted his DPhil in Mathematics at the University of Oxford, in which he focused his research on sparsity and efficiency in deep learning.

Research interests

Deep learning research has seen a massive surge in research and applications over the last six years. It achieves state-of-the-art (and sometime even super-human) performance on classical supervised learning tasks, and is the basis many of the applications of machine learning that we all interact with every day. However, despite these successes there remain a large number of crucial open questions. Why do these methods generalise (and therefore perform) as well as they do? Are there more principles ways of designing deep neural networks? And can we guarantee when they will work? Ilan’s research will contribute to this growing body of theory for deep learning, with a particular focus on understanding adversarial examples - datapoint engineered to trick well trained models - with a view to building safer, more robust algorithms.

Ilan's doctoral research results touching on four areas: Mathematical theory concerning sparse random DNNs; an architecture for training extremely sparse DNNs with minimal trainable parameters; a technique for shrinking the memory footprint of trained CNNs by compressing their feature maps and foldding this compression into the architecture; and techniques for generating and training DNNs with sparse activations.

Selected publications and papers

'Price, Ilan, and Jared Tanner. "Dense for the price of sparse: Improved performance of sparsely initialized networks via a subspace offset." International Conference on Machine Learning. PMLR, 2021. (Jul 18 -24 2021)

Trajectory growth lower bounds for random sparse deep ReLU networks. Price, Ilan, and Jared Tanner. "Trajectory growth lower bounds for random sparse deep ReLU networks." 2021 20th IEEE International Conference on Machine Learning and Applications (ICMLA). IEEE, 2021. (Dec 13 - 15, 2021)

Price, Ilan, and Jared Tanner. "Improved Projection Learning for Lower Dimensional Feature Maps." ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2023. (Upcoming: Jun 4 - 9 2023)

 

Turing Events

Turing Publications