ICAI: The Labs – Deep Learning in NL
This month, ICAI: The Labs is focused on Deep Learning in the Netherlands. The QUVA Lab and the e/MTIC Lab each present their work and discuss challenges and developments made in this field.
The QUVA Lab is a collaboration between Qualcomm and University of Amsterdam. The mission of the QUVA-lab is to perform world-class research on deep vision.
The e/MTIC lab is a collaboration between Eindhoven University of Technology, Catharina Hospital, Maxima Medical Center, Kempenhaeghe Epilepsy and Sleep Center and Philips. The main purpose of e/MTIC is to provide a fast track to high-tech health innovations.
12.00 (noon): Introduction QUVA Lab by Cees Snoek (UvA)
12:05: Philip Lippe on “Efficient Neural Causal Discovery without Acyclicity Constraints” (UvA)
12.20: Introduction e/MTIC Lab bij Paul Merkus (TU/Eindhoven)
12:25: Ben Luijten on “Deep learning for ultrasound signal processing” (TU/Eindhoven)
12.40: Discussion what’s next for Deep Learning in NL
Title: ‘Efficient Neural Causal Discovery without Acyclicity Constraints’
Learning the structure of a causal graphical model using both observational and interventional data is a fundamental problem in many scientific fields. A promising direction is continuous optimization for score-based methods, which efficiently learn the causal graph in a data-driven manner. However, to date, those methods require constrained optimization to enforce acyclicity or lack convergence guarantees. In this work, we present ENCO, an efficient structure learning method leveraging observational and interventional data. ENCO formulates the graph search as an optimization of independent edge likelihoods with the edge orientation being modeled as a separate parameter. Consequently, we can provide convergence guarantees of ENCO under mild conditions without constraining the score function with respect to acyclicity. In experiments, we show that ENCO can efficiently recover graphs with hundreds of nodes, an order of magnitude larger than what was previously possible, while handling deterministic variables and latent confounders.
Title: ‘Deep learning for ultrasound signal processing’
The quality of ultrasound images is strongly dependant on the underlying reconstruction algorithms, transforming acoustic reflections into an interpretable image. Conventionally these algorithms have been relatively simple, due to computational constraints, and striving for real-time framerates. In this presentation we will discuss artificial intelligence solutions that lie at the interface of deep learning, and conventional signal processing. In particular we focus on the front-end signal processing (e.g. sampling, quantization and beamforming), and how end-to-end learning of these steps can transform the next generation of ultrasound devices.