Texas Instruments (Ireland)

Seeding spectral datasets by augmenting the data matrix with either the full spectrum or selected spectral features in order to bias multivariate analysis towards the solution of interest is explored. It is demonstrated that such seeding can have a profound effect on the endpoint of the analysis. Using Raman spectroscopic data of human lung adenocarcinoma cells (A549) in vitro, systematic perturbations to the spectra are introduced to simulate dose dependent exposure to a drug (cisplatin), and/or cellular response, representing reduced viability. Taking Principal Components Analysis (PCA) as the first example, seeding with the known spectral profiles of the drug exposure is demonstrated to greatly increase the ability of the algorithm to differentiate two distinct data subsets, representing control and exposed. The improved differentiation is quantified by further Linear Discriminant Analysis of the PCA data. Other examples of where seeding may be applied include, simulated datasets consisting of simultaneous changes in the spectral markers of exposure dose and cellular response, which are used for Multivariate Curve Resolution - Alternating Least Squares analysis (MCR-ALS). In the example presented, adding pure components to the dataset improves the ability of the algorithm to both model the systematic variation of concentration dependent data and extract the component spectra more accurately than the unseeded dataset. The seeded approach thus provides improved performance for differential analysis of datasets, as well as spectral unmixing analyses, to monitor, for example, the kinetic evolution of a reaction mixture, or metabolic pathway.

This paper presents the design and development of a low-cost, interactive Artificial Intelligence (AI) toolkit in the form of an animatronic ‘talking’ mask. The toolkit aims to democratise AI and make it more accessible, particularly for older adults, by facilitating hands-on engagement with AI technologies and engaging them in meaningful discussions about its opportunities and potential drawbacks. The toolkit is built using customisable features such as replaceable face, jaws and eyes, and modular software features like face tracking, speech recognition, speech-to-text and text-to-speech engines, and large language model (LLM). The customisable design provides a means to explore perceptions, raise awareness and encourage healthy discussions about different facets of AI, especially amongst technologically-challenged populations. It promotes co-design opportunities by leveraging its customisability to allow for adaption of its features to suit the technical abilities and preferences of the audience. Designed with an open-source approach, the project also encourages collaboration among makers and developers, fostering community-driven enhancements and adaptations. This accessibility empowers others to build upon the design, contributing to its evolution as a versatile tool for education and outreach. Evaluation of the platform with 3 distinct user groups highlights its potential to demystify AI technologies and engage diverse audiences in a conversation about the challenges and opportunities they present. This paper concludes with the identification of future opportunities for this research, including the use of co-creation methodologies with end users to extend the work to suit their needs and provide a public engagement framework for understanding emerging technologies.

While WiFi-based human activity recognition (HAR) learning models can achieve good accuracy in one environment (domain), they typically experience a substantial drop in accuracy in new environments where spatial, human, and other physical factors have changed. This work proposes an unsupervised domain adaptation approach for WiFi-based HAR that integrates feature alignment with domain adversarial neural networks (termed FA-DANN). We evaluate FA-DANN on two publicly available multi-environment WiFi datasets: GJWiFi and OPERAnet. Our baseline HAR model, built with a CNN and Attention-based BiLSTM (CNN-ABiLSTM) experiences an anticipated drop when tested in new environments with an average $\mathbf{F 1}$-score of $\mathbf{1 4. 3 4 \%}$ across both datasets and rises to an average $\mathbf{F 1}$-score of $\mathbf{8 5. 5 7 \%}$ across both datasets after applying FA-DANN, with traditional validation and an average F1-score of 76.48% across both datasets using Leave-One-Subject-Out Cross Validation (LOSOCV). We demonstrate that FA-DANN outperforms the state-of-the-art domain adversarial neural network (DANN) approach across both datasets by an average increase in F1-score of 19.98% using traditional validation and 17.67% using LOSOCV. Our work is a significant step in addressing the challenge of creating WiFi-based HAR models that can transfer to new environments / new subjects without the need to collect expensive labeled data for each new scenario.