Rachel Cardell-Oliver

In a Socratic dialogue, a teacher engages a student in a series of questions to prompt critical thinking, guiding the student to clarify and elaborate on their own view and to consider alternative perspectives. In this experience report, we analyse the suitability of large language models (LLMs) as automated Socratic tutors. Postgraduate computer science students from two universities and multiple courses held one-to-one interactive dialogues with LLMs on professional ethics and technology ethics scenarios. We found that the LLMs' questions were diverse and broadly scaffolded critical thinking. Students were able to demonstrate most of the cognitive levels of Bloom's Taxonomy in their answers. However, we also observed some significant limitations of LLMs as Socratic tutors, including unfounded praise of poor responses and reformulating answers for students rather than guiding students to do this themselves. Further, although many students interacted effectively with the LLM, others did not engage authentically with the process. These limitations can potentially be mitigated via prompt engineering and activity framing. Our work contributes an LLM-supported learning activity and our lessons learned on using current state-of-the-art LLMs as Socratic tutors for teaching ethics in computing. In addition, our codebooks provide a set of criteria to drive the evaluation of LLMs used for similar activities.

Understanding locations significant to an individual’s routine travels – such as home and work – gives important context to the individual’s travel patterns. Termed 'anchoring points', these represent places where activities are repeated in space and time around which other activities are arranged. While data from public transport automated fare collection (AFC) systems provides a structured and detailed view of movement patterns with minimal data collection efforts, it cannot provide the same level of background detail of these locations that can be achieved with more labour-intensive data collection methods such as user travel surveys. Where previous studies have developed methods based on AFC data to identify these anchoring points for each individual as points, this work considers these locations as spatial regions 'anchoring regions', acknowledging that individuals may have multiple transport options available to them to access the same activities. Further, this work also develops an approach for allowing individuals to have a flexible number of these anchoring regions, recognising that individual routines and circumstances may differ greatly. Using temporal patterns (such as time of day and duration of time spent at the location) to identify likely activity labels and then the distribution of these activities within each region, a Gaussian mixture model is used to characterise these regions. This resulted in five region activity clusters, with the activity distributions compared to land use and the final clusters of type: 'residences', 'workplaces', 'education', 'residences/leisure', 'workplaces/leisure' . While a sample of the full dataset was used for clustering, the stability of the resulting region activity clusters was tested with a newly developed methodology to validate the results across sample sizes.

Complex urban systems can be difficult to monitor, diagnose and manage because the complete states of such systems are only partially observable with sensors. State estimation techniques can be used to determine the underlying dynamic behavior of such complex systems with their highly non-linear processes and external time-variant influences. States can be estimated by clustering observed sensor readings. However, clustering performance degrades as the number of sensors and readings (i.e. feature dimension) increases. To address this problem, we propose a framework that learns a feature-centric lower dimensional representation of data for clustering to support analysis of system dynamics. We propose Unsupervised Feature Attention with Compact Representation (UFACR) to rank features contributing to a cluster assignment. These weighted features are then used to learn a reduced-dimension temporal representation of the data with a deep-learning model. The resulting low-dimensional representation can be effectively clustered into states. UFACR is evaluated on real-world and synthetic wastewater treatment plant data sets, and feature ranking outcomes were validated by Wastewater treatment domain experts. Our quantitative and qualitative experimental analyses demonstrate the effectiveness of UFACR for uncovering system dynamics in an automated and unsupervised manner to offer guidance to wastewater engineers to enhance industrial productivity and treatment efficiency.

Data from smart water meters can provide valuable insights but it also exposes sensitive information about the behaviour of individuals. This short paper considers local privacy protection for allowing households to selectively hide secrets about their behaviours, as well as sharing their data for analysis. A single water meter signal can be disaggregated into activity channels, with each channel reporting the times and volume of specific activities such as garden irrigation or absences. Adversaries are assumed to have common knowledge of the distinctive distributions of such activities including their likely times of day, temporal- and auto-correlations, and seasonal patterns. Pufferfish privacy is a general purpose framework for secrets to be protected, with mechanisms for protecting those secrets given an adversary’s prior knowledge of the generating mechanisms for the data. In this short paper we propose a Pufferfish activity-aware privacy protection framework for water meter data. Using a real-world dataset of hourly water meter readings for 3557 households, we demonstrate how to hide secrets about the times of leaks, absences and garden irrigation whilst preserving utility for accurate analysis of the data of individuals and populations.

Deep learning is being widely utilized in industrial process monitoring, control, and optimization. However, in the wastewater industry, its applications are still underexplored. This is because deep learning requires a large amount of labeled training data to induce effective predictive models. Owing to the high cost of sensors and frequency and delay in sampling and laboratory analytics, wastewater treatment process data can be sparse with varying frequencies. One option to address training data limitations is to use transfer learning. However, owing to the large covariate shift between the commonly adopted source domains for transfer learning and the target domain of wastewater processes, this approach leads to unacceptable performance. We address this issue by proposing a novel synthetic data generation method for deep predictive modeling of wastewater plants. Employing a Markov process that utilizes random walk, our technique enables the generation of abundant annotated data for our target domain. The method preserves the temporal dynamics and distribution of the original data, thereby closely mimicking the potential original samples of the domain. We extensively evaluate our method over two different high-rate-algae-based treatment datasets, demonstrating considerable performance gains over existing transfer learning. Our proposed algorithm can assist plant operators to deploy responsive supportive models with limited data.