Meeting Minds: Jenkin Lecture - Advances in Applied AI

This year the Jenkin Lecture and supporting lectures take place as part of Meeting Minds, on Saturday 21 September at the Maths Institute from 12.30-5pm.

Jenkin Lecture

14:30-15:30

Learning to Perceive and to Act - What did Generative AI ever do for us?

In recent years, large language models such as ChatGPT and Gemini have firmly established artificial intelligence as a mainstay technology of the future in the public’s perception. Driven by an abundance of unlabelled data and the advent of deep generative models, machines are now able to create content such as text, synthetic images, videos and sounds.

This talk will review how generative AI can be leveraged in robotics to create agents that learn to perceive and to act. In particular, it will highlight one of the most tantalising features of these models beyond direct content creation - the ability to learn structured latent spaces.

Professor Ingmar Posner leads the Applied Artificial Intelligence Lab (A2I) at Oxford University and is a Founding Director of the Oxford Robotics Institute (ORI). Ingmar's research aims to enable machines to robustly act and interact in the real world - for, with, and alongside humans. Ingmar’s expertise lies in the design of robot learning approaches for perception, action and interaction. 

It includes seminal work on large-scale learning from demonstration, unsupervised learning of scene dynamics and 3D object detection. Ingmar is the recipient of a number of best paper awards at recognised international venues in robotics and AI such as ICAPS, IROS and ISER. He has served as Associate Editor for IJRR as well as Area Chair for most of the major robotics conferences. An ELLIS Fellow, Ingmar convened workshops at venues such as ICML, NeurIPS and CVPR as well as two international Robotics Summits.

His work has been covered extensively in national and international press including the Wall Street Journal, the Financial Times, the New Scientist, the Guardian, various outlets of the BBC, and MIT TechReview. In 2016 he received the UK-SMMT Award for Automotive Innovation as well as an Oxford University Impact Award for his work in mobile autonomy. In 2023, the ORI received the Queen's Anniversary Award, the highest national honour available to universities and further education colleges across the UK. In 2014 Ingmar co-founded Oxa, a multi-award winning provider of mobile autonomy software solutions. He is recipient of an Amazon Research Award and currently serves as an Amazon Schola

Supporting lectures

12:30-13:30

Professor Manolis Chatzis graduated from the National Technical University of Athens in 2007 with a Diploma in Civil Engineering and obtained an MSc from NTUA in Structural Engineering in 2008. He then joined Columbia University in the City of New York, where he completed his PhD in the Department of Civil Engineering and Engineering Mechanics in 2012, under the supervision of Professor Andrew Smyth, successfully defending his thesis “The Dynamics of Rigid Bodies on Moving Deformable Media”. Manolis continued in Columbia as a Post Doctoral Research Scientist working on “System Identification and data fusion”. He was appointed an Associate Professor in the Department of Engineering Science and a Tutorial Fellow at Hertford in 2013.

Structural Health Monitoring: Current Practices and the Opportunities with AI

Infrastructure elements such as bridges, buildings and wind turbines are critical to the proper use of modern societies. Due to their long exposure to environmental loads, the accumulated fatigue and damage and the inevitable uncertainty during their construction, the status of their health is hard to assess. Damage is often not visible or can be found in locations that would be hard for an inspector to reach. The use of sensors for assessing the status of infrastructure elements, broadly termed as Structural Health Monitoring, is a mature field with several decades of field applications. In this presentation we will discuss the use of existing methods in the field and how such methods have progressed gradually from ‘white’ methods, which rely more heavily on modelling assumptions, towards ‘grayer’ methods that rely more on the data. More focus will be given on stochastic subspace identification methods and Bayesian methods such as non-linear Kalman Filters. Comparisons will be drawn with machine learning methods which have lately followed the opposite path, starting from fully agnostic ‘black box’ methods. The recent developments in AI related methods, motivate investigating the joint use of ‘white’ and ‘black’ methods for better assessing the health of infrastructure elements.

Konstantina Vogiatzaki is an Associate Professor of Engineering Science, and Fellow and Tutor in Engineering Science at Somerville. Konstantina leads research into multiscale computational modelling that help answer fundamental question in the broad field of fluid dynamics, phase change, turbulence, and heat transfer. She combines analytical methods, data-sets generated by supercomputers and more recently machine learning techniques to produce high fidelity models of the underlying thermo-fluid processes governing complex systems ranging from gas turbines and injection devices to carbon capture units and human organs. Having a strong interest in sustainability she is also working in close collaboration with industries helping them with the modelling tools she develops in the faster design and optimisation of environmentally responsible energy and propulsion solutions. 

AI-Driven Innovations in Energy and Transportation: From empowering traditional Computational Fluid Dynamics to improving Market Predictions

As modern societies confront the dual challenges of climate change and increasing energy demands, the integration of Artificial Intelligence (AI) into the design and optimisation of energy and transportation systems emerges as a pivotal opportunity for transformation.

This talk delves into the intersection of AI and energy, illustrating how high performance computing, machine learning techniques, and advanced Computational Fluid Dynamics (CFD) algorithms can significantly enhance system efficiency, optimize resource management, mitigate risks, and drive innovations in the next generation of energy and transportation systems’ deployment.

We will explore three key aspects of real-world applications: First, we will discuss how advanced machine learning algorithms, such as Deep Neural Networks, can improve the computational efficiency and accuracy of traditional CFD techniques. Second, we will demonstrate a framework for real time risk management utilizing synthetic training data generated from CFD, focusing on hydrogen-powered systems—an ideal showcase scenario due to the scarcity of real-life data. Finally, we will explore how machine learning can enhance the predictive capabilities of time series models, with specific applications in the energy market. Finally, the talk will briefly address the implications and engineering challenges of employing AI in the energy sector, advocating for strategies that ensure equitable access to energy solutions.