Matterhorn Studio

We're #hiring a new Machine Learning Researcher in Greater London, England. Apply today or share this post with your network.

We're #hiring a new Industry-PostDoc in Machine Learning for Material Science in London, England. Apply today or share this post with your network.

NeurIPS 2023 After-Party: 14th December, New Orleans We are looking forward to host you for an informal evening of Bayesian Optimisation and Lab Automation enthusiasts, parallel to NeurIPS 2023, sponsored by Matterhorn Studio! We have reserved tables at a bar with views across the Mississipi, close to the south end of Woldenberg Park. Details: https://lnkd.in/dTEZDNmw The exact location will be sent to attendants on the day, if they are accepted on the guest-list, to avoid overcrowding and to focus the event on people with interest in Bayesian Optimisation and Lab Automation. This After-Party is in the context of conferences earlier this year: Come around to discuss Bayesian Optimisation and Lab Automation with a cross-over to materials and chemistry for biosynthetics, alternative proteins, hydrogen, fusion, foundation industries, materials 4.0, plastics, in the context of ChemAI 2023, the recent Henry Royce National Institute Conference in Manchester and the Accelerate23 conference in Toronto (http://accelerate23.ca).

Join us tomorrow 2pm for talk 4 out 4 for our November Research Series: Gbetondji Dovonon will discuss our recent NeurIPS workshop paper on “Long-run Behaviour of Multi-fidelity Bayesian Optimisation”. Link: https://lnkd.in/ekQP5_PW Abstract: Multi-fidelity Bayesian Optimisation (MFBO) has been shown to generally converge faster than single-fidelity Bayesian Optimisation (SFBO). Inspired by recent benchmark papers, we are investigating the long-run behaviour of MFBO, based on observations in the literature that it might under-perform in certain scenarios. An under-performance of MBFO in the long-run could significantly undermine its application to many research tasks, especially when we are not able to identify when the under-performance begins. We create a simple benchmark study, showcase empirical results and discuss scenarios, concluding with inconclusive results.

Upcoming Neurips23 workshop paper together with Lars Puiman at Delft University of Technology, here is Mahdi Eskandari presentation from the seminar last week.

Join us tomorrow 2pm for talk 3 out 4 for our November Research Series: Zoe Wang at Imperial College London will present on “Closed-loop Optimisation of Deformable Mirrors for Laser Beam Aberration Correction”. Link: https://lnkd.in/etK_HYWf Abstract: Laser beam optimisation is important for the creation of high intensity laser focal spots used in applications such as laser acceleration of particles from a thin foil target, with potential applications in areas such as hadron therapy. By presenting a closed-loop optimisation of low actuator count deformable mirrors (DM) for correction of spatial phase aberrations in a high power laser beam, we explore optimisation algorithms across systems with 5 and 9 actuators with a 12-bit control system (with $10^{18}$ and 3x$10^{32}$ search spaces) in terms of required lab time and speed and robustness of convergence. Conventional approaches such as genetic algorithms are comparatively brute force and can be inefficient when applied to a large search space. Hence, we also evaluate methods using uncertainty estimates, such as Bayesian Optimisation, that aim to maximise information gain at every experimentation step. We also discuss practical issues, such as methods for efficiently determining a simple single valued metric of laser beam quality with image data from a camera. Averaging noisy 2D visual measurements increases precision, but comes at the cost of slower experimental runs, and so it is important to find a balance between speed, accuracy and robustness against becoming trapped in a local minimum. Overall, the system showcases the challenges of a high throughput closed-loop optimisation system linked to image based measurement, and further work will focus on the integration of physical expert knowledge such as the representation of a laser beam aberration or a correction using a group of actuators described by a Zernike polynomial.

Join us also tonight at Leidseplein as a warm-up for ChemAI! https://lnkd.in/eu6aZ42p

Join us today at 2pm for our Talk #2 of the #November #Research #Series Mahdi Eskandari will discuss our recent NeurIPS workshop paper on “Multi-fidelity Bayesian Optimisation for Syngas Fermentation Simulators”. Time: Nov 14, 2023 02:00 PM London Link: https://lnkd.in/eSmC7NgB Abstract: A Bayesian optimization approach for maximizing the gas conversion rate in syngas fermentation is presented. We have access to an expensive-to-evaluate, computational fluid dynamic (CFD) reactor model and a cheap ideal-mixing based reactor model. The goal is to maximize the gas conversion rate with respect to the input variables. Due to the high cost of the industrial simulator, a multi-fidelity Bayesian optimization is adopted to solve the optimization problem using both high and low fidelities. We first describe the problem of syngas fermentation followed by our approach to solving simulator optimisation using multiple fidelities. We discuss concerns regarding significant differences in fidelity cost and their impact on fidelity-sampling and conclude with a discussion on the integration of real-world fermentation data.

Join our November Research Series Talk #2 next Tuesday at 2pm London! Mahdi Eskandari will discuss our recent NeurIPS workshop paper on "Multi-fidelity Bayesian Optimisation for Syngas Fermentation Simulators". Time: Nov 14, 2023 02:00 PM London Link: https://lnkd.in/eSmC7NgB Details: https://lnkd.in/euBtewUj Abstract: A Bayesian optimization approach for maximizing the gas conversion rate in syngas fermentation is presented. We have access to an expensive-to-evaluate, computational fluid dynamic (CFD) reactor model and a cheap ideal-mixing based reactor model. The goal is to maximize the gas conversion rate with respect to the input variables. Due to the high cost of the industrial simulator, a multi-fidelity Bayesian optimization is adopted to solve the optimization problem using both high and low fidelities. We first describe the problem of syngas fermentation followed by our approach to solving simulator optimisation using multiple fidelities. We discuss concerns regarding significant differences in fidelity cost and their impact on fidelity-sampling and conclude with a discussion on the integration of real-world fermentation data.