Abstract
Concerns with the interpretability of ML models are growing as the technology is used in increasingly sensitive domains (e.g., health and public administration). Synthetic data can be used to understand models better, for instance, if the examples are generated close to the frontier between classes. However, data augmentation techniques, such as Generative Adversarial Networks (GAN), have been mostly used to generate training data that leads to better models. We propose a variation of GANs that, given a model, generates realistic data that is classified with low confidence by a given classifier. The generated examples can be used in order to gain insights on the frontier between classes. We empirically evaluate our approach on two well-known image classification benchmark datasets, MNIST and Fashion MNIST. Results show that the approach is able to generate images that are closer to the frontier when compared to the original ones, but still realistic. Manual inspection confirms that some of those images are confusing even for humans.

Abstract
The use of Virtual Reality (VR) technology to train professionals has increased over the years due to its advantages over traditional training. This paper presents a study comparing the effectiveness of a Virtual Environment (VE) and a Real Environment (RE) designed to train firefighters. To measure the effectiveness of the environments, a new method based on participants' Heart Rate Variability (HRV) was used. This method was complemented with self-reports, in the form of questionnaires, of fatigue, stress, sense of presence, and cybersickness. An additional questionnaire was used to measure and compare knowledge transfer enabled by the environments. The results from HRV analysis indicated that participants were under physiological stress in both environments, albeit with less intensity on the VE. Regarding reported fatigue and stress, the results showed that none of the environments increased such variables. The results of knowledge transfer showed that the VE obtained a significant increase while the RE obtained a positive but non-significant increase (median values, VE: before - 4 after - 7, p = .003; RE: before - 4 after - 5, p = .375). Lastly, the results of presence and cybersickness suggested that participants experienced high overall presence and no cybersickness. Considering all results, the authors conclude that the VE provided effective training but that its effectiveness was lower than that of the RE.

Abstract
Catharanthus roseus leaves produce a range of monoterpenoid indole alkaloids (MIAs) that include low levels of the anticancer drugs vinblastine and vincristine. The MIA pathway displays a complex architecture spanning different subcellular and cell-type localizations and is under complex regulation. As a result, the development of strategies to increase the levels of the anticancer MIAs has remained elusive. The pathway involves mesophyll specialised idioblasts where the late unsolved biosynthetic steps are thought to occur. Here, protoplasts of C. roseus leaf idioblasts were isolated by fluorescence-activated cell sorting, and their differential alkaloid and transcriptomic profiles were characterised. This involved the assembly of an improved C. roseus transcriptome from short- and long-read data, IDIO+. It was observed that C. roseus mesophyll idioblasts possess a distinctive transcriptomic profile associated with protection against biotic and abiotic stresses, and indicative that this cell type is a carbon sink, in contrast with surrounding mesophyll cells. Moreover, it is shown that idioblasts are a hotspot of alkaloid accumulation, suggesting that their transcriptome may hold the keys to the in-depth understanding of the MIA pathway and the success of strategies leading to higher levels of the anticancer drugs.

Abstract
Effective utility system management is fundamental and critical for ensuring the normal activities, operations, and services in cities and urban areas. In that regard, the advanced information and communication technologies underpinning smart cities enable close linkages and coordination of different subutility systems, which is now attracting research attention. To increase operational efficiency, we propose a two-stage optimal co-management model for an integrated urban utility system comprised of water, power, gas, and heating systems, namely, integrated water-energy hubs (IWEHs). The proposed IWEH facilitates coordination between multienergy and water sectors via close energy conversion and can enhance the operational efficiency of an integrated urban utility system. In particular, we incorporate social-aware peer-to-peer (P2P) resource trading in the optimization model, in which operators of an IWEH can trade energy and water with other interconnected IWEHs. To cope with renewable generation and load uncertainties and mitigate their negative impacts, a two-stage distributionally robust optimization (DRO) is developed to capture the uncertainties, using a semidefinite programming reformulation. To demonstrate our model's effectiveness and practical values, we design representative case studies that simulate four interconnected IWEH communities. The results show that DRO is more effective than robust optimization (RO) and stochastic optimization (SO) for avoiding excessive conservativeness and rendering practical utilities, without requiring enormous data samples. This work reveals a desirable methodological approach to optimize the water-energy-social nexus for increased economic and system-usage efficiency for the entire (integrated) urban utility system. Furthermore, the proposed model incorporates social participations by citizens to engage in urban utility management for increased operation efficiency of cities and urban areas.

Abstract

Abstract
Directed Energy Deposition (DED) is crucial in additive manufacturing for various industries like aerospace, automotive, and biomedical. Precise temperature control is essential due to high-power lasers and dynamic environmental changes. Employing Reinforcement Learning (RL) can help with temperature control, but challenges arise from standardization and sample efficiency. In this study, a model-based Reinforcement Learning (MBRL) approach is used to train a DED model, improving control and efficiency. Computational models evaluate melt pool geometry and temporal characteristics during the process. The study employs the Allen-Cahn phase field (AC-PF) model using the Finite Element Method (FEM) with the Multi-physics Object-Oriented Simulation Environment (MOOSE). MBRL, specifically Dyna-Q+, outperforms traditional Q-learning, requiring fewer samples. Insights from this research aid in advancing RL techniques for laser metal additive manufacturing.