Calendar

Sequential optimization is one of the fastest growing areas of machine learning. In this presentation we deep dive into sequential optimization based on Gaussian process models (aka Bayesian optimization). We will take a look at the analysis of popular algorithms such as UCB and Thompson sampling and wrap up with an overview of recent results and open problems.

Recommended reading:

Srinivas et al. 2010 (https://arxiv.org/abs/0912.3995), recipient of ICML 2020 Test of Time award
Chowdhury and Gopalan 2017 (https://arxiv.org/abs/1704.00445)
Vakili et al. 2020 (https://arxiv.org/abs/2009.06966)
Wilson et al. 2020 (https://arxiv.org/abs/2002.09309)

The lecture will outline a theory of heterogeneous catalysis that allows a detailed understanding of elementary chemical processes at transition metal surfaces and singles out the most important parameters determining catalytic activity and selectivity. It will be shown how scaling relations allow the identification of descriptors of catalytic activity and how they can be used to construct activity and selectivity maps. The maps can be used to define catalyst design rules and examples of their use will be given.

Link to live online lecture: https://meet.google.com/efz-imrr-rgh

Join Zoom Meeting
https://us02web.zoom.us/j/89506620956?pwd=Q1VqNlQwWXh5eVBBZCtYd3IxeXd5Zz09

Meeting ID: 895 0662 0956
Passcode: 211168

CANCELLED

Recently, there has been a renewed interest in the problem of obtaining control laws directly from measured data. This talk focuses on the design of state feedback controllers using finite (and possibly noise-corrupted) trajectories of linear systems. The main ingredients of our approach are generalizations of the well-known Finsler’s lemma and Yakubovich’s S-lemma. These results allow us to come up with non-conservative control design methods that rely on tractable linear matrix inequalities.

Zoom link: https://zoom.us/j/92943573797

Abstract not available

Computer simulations are becoming useful in providing insight in the physical and chemical processes taking places in nature.
Simulations yield molecular level understanding, which is often complementary information to the understanding provided by experimental investigations.
Yet, they are only useful when when they can accurately model the physical system.
High accuracy is often only obtained by resorting to first principles, and by modelling the quantum mechanics features of the system of interest at the atomic level.
Thriving nanotechnologies and exciting experiments pose big challenges to computational approaches. On the one hand, the systems to be simulated are large and computationally expensive, and their physical and thermal properties require sampling of a large phase space (using molecular dynamics or other techniques).
On the other hand, the high accuracy required to evaluate inter-atomic interactions often means using very accurate and expensive approaches to solve the Schrodinger equation.
We discuss here some of the most accurate approaches available to assess the ground state electronic states and their properties in molecular systems, solids and surfaces, namely quantum Monte Carlo (QMC) methods.
QMC simulations are computationally expensive and often demands the employment of high performance computers.
However, recent developments have drastically reduced the overall cost of QMC, especially in the evaluation of interaction energies.
QMC methods can be used to benchmark cheaper but less accurate approaches (such as density functional theory, or empirical force fields) promoting their further developments. The combination of this hierarchy of methods, coupled with machine learning techniques, then provides high accuracy for systems whose size would preclude a full quantum mechanics approach.

Abstract not available

*The seminar will take place via vidyo “here”:https://indico.cern.ch/event/996329/.
The explicit url is: https://indico.cern.ch/event/996329/.*

The precision of the data collected at the Large Hadron Collider (LHC) demands an increasing theoretical accuracy at the multi-differential level to fully exploit the LHC potential. In this talk I will introduce a formalism for the resummation of transverse observables in direct space (RadISH formalism). As a phenomenological application, I will show accurate QCD predictions for single and double-differential observables for various colour-singlet processes at the LHC.

Abstract not available

Abstract not available

Humans are able to perform a wide range of joint actions, from carrying heavy objects together to having conversations. What are the mechanisms enabling joint action? This talk will provide an overview of experimental research that has begun to unravel the behavioural, cognitive, and neural processes supporting joint action planning and coordination. While much of this research has focused on egalitarian dyadic joint actions, new findings also shed light on role and task distributions in more complex group contexts. It will be discussed what we can learn from joint action research for increasing affiliation and cooperation, for improving the design of collaborative robots, and for enhancing our understanding of aesthetic experiences during observation of joint performances.

Claire is a Consultant Haematologist and Honorary Senior Lecturer in Haematology at UCL with a particular interest in CAR T-cells for cancer. She completed a PhD in Cellular Immunotherapy at UCL in the laboratory of Professor Karl Peggs and subsequently undertook a Clinician Scientist post in Dr Martin Pule’s Laboratory to work on the UCL CAR T-cell program. Claire’s current role involves pre-clinical development of novel CAR projects, GMP CAR T-cell manufacture and Clinical Trial design for academic CAR T-cell studies at UCL. She is also responsible for the development of a clinical service at UCLH to support patients recruited to CAR T-cell studies and those receiving CAR T on the NHS.
https://www.youtube.com/c/DarwinCollegeLectureSeries

During learning, populations of neurons alter their connectivity and activity patterns, enabling the brain to construct a model of the external world. Conventional wisdom holds that the durability of a such a model is reflected in the stability of neural responses and the stability of synaptic connections that form memory engrams. However, recent experimental findings have challenged this idea, revealing that neural population activity in circuits involved in sensory perception, motor planning and spatial memory continually change over time during familiar behavioural tasks. This continual change suggests significant redundancy in neural representations, with many circuit configurations providing equivalent function. I will describe recent work that explores the consequences of such redundancy for learning and for task representation. Despite large changes in neural activity, we find cortical responses in sensorimotor tasks admit a relatively stable readout at the population level. Furthermore, we find that redundancy in circuit connectivity can make a task easier to learn and compensate for deficiencies in biological learning rules. Finally, if neuronal connections are subject to an unavoidable level of turnover, the level of plasticity required to optimally maintain a memory is generally lower than the total change due to turnover itself, predicting continual reconfiguration of an engram.

Abstract not available

Abstract not available

Over the past twenty years or so, velocity-map imaging has revolutionised experimental work in the field of chemical reaction dynamics by allowing the complete scattering distribution for a chosen reaction product to be imaged directly in the gas phase in a single measurement. The technique has provided fascinating and often quantum-state resolved insights into the mechanisms of a wide variety of photochemical and bimolecular reactions.

The development of universal ionization methods and ultrafast time-of-flight imaging sensors has ushered in the era of multi-mass velocity-map imaging, in which multiple reaction products can be imaged in a single experiment. As well as offering a vast reduction in data acquisition time, multi-
mass imaging data sets can be analysed using a statistical data processing technique known as covariance analysis or covariance mapping to reveal correlations amongst the reaction products.

This talk will provide an introduction velocity-map imaging and covariance mapping, and will show how these techniques can be used to unravel complex multi-step reaction mechanisms and to provide information on gas-phase molecular structures.

Abstract not available

*Register your interest here:* https://forms.gle/4udDM3G13grY7Dko9

*Abstract:*

In this talk, I will review progress in the field of gravitational wave detection from the first days of the aluminium bar detectors to the present time where the laser interferometer detectors Advanced LIGO and Advanced Virgo have allowed gravitational waves to be detected and are opening up a new field of gravitational multi-messenger astrophysics. Many experimental challenges had to be overcome and new challenges are presenting themselves as we look to further enhance the performance of ground based detectors and look to lower frequencies with the space based detector LISA. Further, the most recent discoveries by the collaboration will be discussed.

*Speaker profile:*

Professor Sir James Hough (Jim) is a graduate of the University of Glasgow where he became Professor of Experimental Physics in 1986 and is the emeritus holder of the Kelvin Chair of Natural Philosophy. He was Director of the University’s Institute for Gravitational Research from 2000 to 2009 and is now Associate Director. His research interests are centred on gravitational wave detection on the ground and in space. Prof Hough was elected to the Royal Society of Edinburgh in 1991, the Fellowship of the Institute of Physics in 1993, and the Royal Society of London in 2003. For his wide-ranging research and advisory work, Sir James was awarded the Officer of the Most Excellent Order of the British Empire (OBE) in the 2013 Queen’s Birthday Honours, and a Knighthood in the 2018 Queen’s Birthday Honours.

Two modes of Subantarctic Mode Water (SAMW) form in the Pacific, in regions of deep winter mixed layers, on the northern side of the Subantarctic Front. Each water mass has experienced significant interannual variability in recent years. In this study, mixed layer temperature and salinity budgets were computed in the formation regions, to determine the drivers of variability in the mixed layer properties. The dominant drivers are shown to be surface buoyancy fluxes, horizontal advection and entrainment. While surface buoyancy fluxes set the depth of the winter mixed layer, strong advection in the lead up to the deepening of the mixed layer also drives strong variability in mixed layer salinity. Salt advection in each water mass formation region is strongly correlated with sea ice area in the northern Ross Sea, at lags of up to two years. Correlation is also found between salt advection in the southeast Pacific SAMW formation region and sea ice area in the northern Amundsen/Bellingshausen Sea, at lags of up to six months.

A central goal of computational physics and chemistry is to predict material properties using
first principles methods based on the fundamental laws of quantum mechanics. However,
the high computational costs of these methods typically prevent rigorous predictions of
macroscopic quantities at finite temperatures, such as chemical potential, heat capacity and thermal conductivity.

In this talk, I will first discuss how to enable such predictions by combining advanced statistical mechanics
with data-driven machine learning interatomic potentials. As an example [1], for the
omnipresent and technologically essential system of water, a first-principles thermodynamic
description not only leads to excellent agreement with experiments, but also reveals the
crucial role of nuclear quantum fluctuations in modulating the thermodynamic stabilities of
different phases of water. As another example [2], we simulated the high pressure hydrogen system
with converged system size and simulation length, and found, contrary to established beliefs,
supercritical behaviour of liquid hydrogen above the melting line.
Besides the computation of thermodynamic properties, I will talk about transport properties:
Ref [3] proposed a method to compute the heat conductivities of liquid just from equilibrium molecular dynamics trajectories.

During the second part of the talk, I will rationalize why machine learning potentials work at all, and in particular, the locality argument.
I’ll show that a machine-learning potential trained on liquid water alone can predict the properties of diverse ice phases,
because all the local environments characterising the ice phases are found in liquid water [4].

References:
[1] Bingqing Cheng, Edgar A Engel, Jörg Behler, Christoph Dellago, Michele Ceriotti. (2019) ab initio thermodynamics of liquid and solid water. Proceedings of the National Academy of Sciences, 116 (4), 1110-1115.
[2] Bingqing Cheng, Guglielmo Mazzola, Chris J. Pickard, Michele Ceriotti. (2020) Evidence for supercritical behaviour of high-pressure liquid hydrogen. Nature, 585, 217–220
[3] Bingqing Cheng, Daan Frenkel. (2020) Computing the Heat Conductivity of Fluids from Density Fluctuations. Physical Review Letters, 125, 130602
[4] Bartomeu Monserrat, Jan Gerit Brandenburg, Edgar A. Engel, Bingqing Cheng. (2020) Liquid water contains the building blocks of diverse ice phases. Nature Communications 11.1: 1-8.

Abstract not available

Abstract not available

Abstract not available

Abstract not available

Over the past 15 years, machine-learning interatomic potentials have evolved from a promising idea to a wide field of materials modeling. The idea is that a true interatomic interaction energy (or an accurate quantum-mechanical model of it) can be approximated as a function of positions of neighbors of each atom with a flexible (systematically improvable) functional form. Within this field, researchers are working on different directions: studying ways to improve accuracy, efficiency, and the range of applicability of potentials, constructing potentials for different atomistic systems, or combining these potentials with other algorithms to enable materials properties calculation that was out-of-reach for more traditional methods. I will present my work in this field under my favorite angle: machine-learning potentials as a computational technology to seamlessly accelerate quantum-mechanical calculations.

Namely, I will present my version of machine-learning potentials, moment tensor potentials, and an active learning algorithm automating the procedure of assembling the dataset. I will show how the two algorithms combined allow for an automatic acceleration by orders of magnitude in such applications as constructing convex hulls of stable alloy structures or computing vibrational and configurational free energy of alloys. Moreover, machine-learning potentials can be used as a screening tool before the final quantum-mechanical calculation, offering a speedup of several orders of magnitude without committing any numerical error.

Abstract not available

Abstract not available

Abstract not available