Machines have long helped us understand the world around us, but only recently are we relying on their assistance in recognising patterns. Still, such machine learning has rapidly become an integral part of society, in speech recognition or information retrieval; and in science, for detecting patterns in nature and the unvarying cosmos.
Now, the need is growing rapidly to run machine-learning neural networks in real time. Self-driving cars and responsive manufacturing require fast feedback for complex systems. On a more fundamental level, self-learning machines help us unveil a dynamical Universe we did not know existed. Bright, millisecond duration transients appear all over the radio sky, pointing to extreme energies. Gravitational waves whipped up by merging black holes and neutron stars peak for 0.1 second, maybe once per year.
Our overall aim is to understand how such real-time studies can be achieved, and apply these to the explosive Universe.
Our main goal is to discover the physics that powers radio transients and compact-object mergers, by real-time detection of gravitational-wave and electromagnetic emission. Our previous work is world-leading in this field. We will now accelerate essential algorithms and pipelines, to understand the immediate energetics, composition and ejecta of these merging stars, for the first time.
Our second goal is to understand how deep neural networks can be made more efficient, and how data-driven auto-tuned pipelines, on highly parallel accelerator hardware, can produce optimal results for all the sciences.
To achieve these goals, we initiate a highly innovative program: CORTEX, the Center for Optimal, Real-Time Machine Studies of the Explosive Universe. CORTEX bridges fundamental research to society, and connects academic, applied, public and industry partners. Together we will provide the complex, real-time systems that are essential to, for the first time, reveal the exotic, dynamical Universe.
