Last year, Applied AI Letters announced the formation of its Associate Editorial Board. The purpose of this body was to provide a platform for rising stars in the field of applied AI, to enable them to have a voice in the world of publishing, and so to ensure that Applied AI Letters was a venue which maintained its close connection to the community from which it was born. As part of their role on the Associate Editorial Board, these talented practitioners were asked to go out into the community and find research works, which they felt were particularly exciting and bring them to Applied AI Letters for wider publication and dissemination. Today, I am proud to write this Editorial announcing the first fruit of this effort. In this issue, we can highlight two of these carefully selected pieces of research, chosen by two of our Associate Editorial Board members, Dr Helge Spieker and Dr Philippe Schwaller. Helge is a postdoctoral researcher at Simula Research Laboratory in Norway, whose research interests are in the applications of artificial intelligence and machine learning, especially in the domains of validation and verification of software-based autonomous systems, industrial robotics, and sports. He also maintains an interest in the integration of machine learning and symbolic AI methods, such as constraint programming, to enable robust hybrid intelligent systems that are reliable and robust. Helge has commissioned an article entitled ‘Deep learning to predict power output from respiratory inductive plethysmography data’.1 As we move into an age where our own personal health data are becoming increasingly important to personalise our lifestyles, including exercise regimes, there is a new scrutiny being placed on the equipment, which is required to generate these data. A reliance on expensive equipment could drive inequalities between those who are financially able to partake in this monitoring and those who cannot. This work takes the case of measuring power output, a common metric used to define exercise intensity. Traditional technologies require expensive equipment, which must also be embedded into the technology, which limit usage. In this paper, an AI model based upon respiratory inductive plethysmography (a non-invasive method for monitoring breathing) and heart rate was built and tested against standard sensor-based power output measurements. The model was able to function with low error and provides evidence of the potential for AI methods to democratise data gathering in this field. Philippe Schwaller is an assistant professor at EPFL (Swiss Federal Institute of Technology Lausanne), having previously worked at IBM Research – Europe, in Zurich. He completed a PhD in Chemistry and Molecular Sciences at the University of Bern, an MPhil in Physics at the University of Cambridge, and an MSc in Materials Science & Engineering at EPFL. His main research focus is on accelerating the discovery and synthesis of novel materials and molecules using AI techniques. Philippe has commissioned an article entitled ‘Generative model-enhanced human motion prediction’.2 Generative models are sometimes referred to as computational creativity agents, as they are adept at identifying and plausibly filling in the white space in data. Whilst we are commonly used to seeing these models in technologies such as deep fakes, as part of a generative adversarial network, this paper uses this family of techniques to augment a human motion data set. These data sets are often prone to building models, which have weaknesses to rare events, and for them to be useful in the ‘real world’, it is necessary to build some robustness to these out of distribution events. In this paper, this robustness is achieved by using generative models to enhance the data set with data which more closely represents the true generating distribution. This work shows that this approach can dramatically improve model robustness without a significant performance penalty. I hope that you enjoy these articles, we will continue this series in the next issue with some more articles championed by our associate editors.
Search from 250 M+ research papersSearch from 250 M+ research papers
Dec 1, 2023
Esmaeal Ghavanloo 
Open Access