Eric Howard

Eric Howard is a Postdoctoral Research Fellow in Theoretical Physics and cosmology at CSIRO, Astronomy and Space Science and Macquarie University Centre for Quantum Engineering, Quantum Materials and Applications Group, working on quantum sensors. He is also Adjunct Research Fellow in computational condensed matter at Griffith University. He earned his PhD in black hole physics at Macquarie University and his current research interests are in quantum many-body physics, quantum sensing and optomechanics, non-equilibrium and phase transitions, quantum information and thermalization, condensed matter systems, machine learning and their applications to cosmology, string theory and quantum gravity.


Affiliation – Macquarie University Position – Postdoctoral Research Fellow Homepage – https://www.researchgate.net/profile/Eric_M_Howard

Talks

Efficient machine learning methods for cosmology

Machine learning-based analysis techniques are gaining interest in cosmology due to their computational ability to generate complex models in order to analyse and interpret large scale structure data sets, such as the matter density fields comprised of nonlinear complex features, like halos, filaments, sheets and voids. Data-driven cosmological discovery has seen a remarkable rise in the last decade, leading to unprecedented improvements in the ways we can gain knowledge, interpret and extract cosmological features from large data volumes, constraining cosmological parameters and modelling the large structure formation of the Universe. The cross-fertilization between cosmology and machine learning require the integration of traditional statistical techniques with modern machine learning models, providing promising opportunities with significant advantages for state-of-the art cosmological simulations. Multiple applications, from cosmic web simulations and predicting the cosmic structure in the non-linear regime to multi-wavelength structure Identification, 21cm reionization models or predicting dark matter annihilation and halo formation may benefit from robust and efficient data analysis methods, such as convolutional neural networks or generative adversarial networks. We present a number of powerful machine learning algorithms (classification, regression, reinforcement learning) and data-analysis tools that can be used to predict the non-perturbative cosmological structure and non-Gaussian features hierarchically formed over all scales in the Universe, justifying the advantage of employing such methods for use in cosmology.