#LboroAppliedAI Seminars Semester 2, 2021
We are continuing our series of #LboroAppliedAI online sessions this Semester! The series is organized by Dr Lise Jaillant and Dr Valerie Pinfield and sponsored by the Centre for Research in Communication and Culture.
The objective is to bring together colleagues and PGRs who are interested in Artificial Intelligence and its applications in a wide range of fields.
You can find the recordings of the Semester 1 sessions here (Lboro log in necessary).
To register, please contact email@example.com
Thurs 14th Jan 1pm:
Multi-Agent Reinforcement Learning: Challenges and Real world Applications
A fundamental goal of AI is to develop intelligent agents. Multi-agent learning involves developing decision making algorithms for autonomous agents in environments where multiple intelligent entities interact with each other. This talk will present key developments in the area of multi-agent policy learning: namely Multi-agent reinforcement learning and multi-agent imitation learning, which are emerging as key techniques to address the problem of multi agent policy learning. The talk will relate to emerging applications of multi-agent policy learning such as driverless vehicle control, sports analytics, urban planning and autonomous generation of video game content. A significant part of the talk will discuss recent attempts at addressing key challenges relate to multi-agent policy learning, such as non-stationarity, communication, selective attention and curriculum learning. The talk will conclude with a discussion on challenges to move theoretical results in the real world applications where agents are required to learn from limited experience, and engineering efforts that are required to do so.
Click here for the recording of the talk (Lboro log in necessary).
Thurs 25th Feb 1pm
Three meeting points between AI and Conversation Analysis
I’ll outline three projects that study different configurations of the relationship between AI and CA: using AI as a tool for doing CA, using CA as a means of improving AI, and exploring the reflexive relationships between AI-based voice interfaces and everyday interactions in a naturalistic setting.
Bio: Saul is a lecturer in social sciences (social psychology) in the communication and media group at Loughborough University. His research explores human interaction in all its forms, including empirical work on how politicians shake hands, how couples dance, how people draw, evaluate art, and how we deal with miscommunication in interaction. He is currently leading a British Academy-funded project studying how disabled people and their personal assistants work with AI-based virtual assistants during everyday domestic routines.
Thurs 4th March 4pm:
Connected and autonomous vehicles
Through Artificial Intelligence, Autonomous Vehicles will soon start to make decisions (1) without the need for drivers; or (2) on behalf of drivers. There are obvious potential benefits for smart and safe mobility by taking the human out of the loop and relying on the vehicle technology to negotiate a safe, efficient, and reliable path through traffic.
However, to achieve these potential benefits, substantial human factors challenges need to be addressed, and confidence in the capability of AI needs to be developed amongst a potentially sceptical public. Some of these challenges include:
- The issue of trust and reliance in the vehicle when it is fully autonomous
- Reliability on the vehicles to perform as necessary in safety-critical situations
- Misuse or over-reliance on the systems
- Negative impacts of drivers disengaging from the task e.g., consequences for situational awareness, fatigue, driver comfort.
Therefore, there is a need to understand driver/operator requirements when the role shifts from active vehicle control to passive monitoring of the system automated through AI.
This presentation will examine some of these human factors challenges and will provide some results from trials of a prototype Autonomous ‘Pod’ that was conducted at QEOP during 2020.
Thurs 25th March 4pm:
AI techniques for design of low carbon energy systems
One of the practical steps to reducing global carbon emission is the wide application of renewable energy devices owing to their high efficiency and green energy conversion. These include fuel cells, for example, where a fuel source (e.g. hydrogen) undergoes electrochemical reaction to produce electricity, and produces a waste product (e.g. water). Many such energy devices incorporate porous materials which contribute to their effective operation in a number of ways, including providing pathways for gas fuel, liquid waste product, electronic conduction and catalyst sites for electrochemical reaction. Mathematical methods such as computational modelling and recent popular artificial intelligence techniques can help to understand the physical dynamics in these porous structures and to achieve the best designs for their operation in energy systems.
In this talk, we will review the evolution of these mathematical approaches and their typical applications in modelling various porous energy materials and devices. We will particularly highlight the successful application of several deep learning methods more familiar in image analysis applications (e.g. generative adversarial networks GANs– and convolutional neural networks CNNs). GANs may be familiar as the means of producing deep-fake images. CNNs are popular in image classification and feature extraction. These can be used to generate candidate material structures, to identify the structure that leads to optimum performance in the energy system. The target for AI in these systems is to achieve autonomous optimisation of structures, operating conditions of the component materials and even the manufacturing process used to produce them.
Thurs 8th July, 4pm:
Exploring deductive and inductive approaches to generating chemical process knowledge through machine learning
Recent advances in computer science and machine learning have provided an unprecedented opportunity to transform traditional chemical and biochemical engineering research into the era of digital manufacturing. Despite the huge success of these data-driven modelling techniques, chemical and biochemical processes are predominantly governed by their physical mechanisms. As a result, primary research in this domain seeks to quantify the knowledge behind the data rather than the data itself. For natural science and engineering research, knowledge is discovered via two routes: induction (observation – pattern – hypothesis – theory) and deduction (theory – hypothesis – observation – confirmation). In this presentation, we will discuss how to adopt different machine learning algorithms to generate knowledge via deduction and distil knowledge via induction through black-box modelling, hybrid modelling, and interpretable modelling approaches. A number of examples will be illustrated, covering central themes within process systems engineering, including predictive modelling, optimisation, control, soft-sensing, and dimensionality reduction.