Abstract
As machine learning technologies are increasingly adopted, the demand for responsible AI practices to ensure transparency and accountability grows. To better understand the decision-making processes of machine learning models, GASTeN was developed to generate realistic yet ambiguous synthetic data near a classifier’s decision boundary. However, the results were inconsistent, with few images in the low-confidence region and noise. Therefore, we propose a new GASTeN version with a modified architecture and a novel loss function. This new loss function incorporates a multi-objective measure with a Gaussian loss centered on the classifier probability, targeting the decision boundary. Our study found that while the original GASTeN architecture yields the highest Fréchet Inception Distance (FID) scores, the updated version achieves lower Average Confusion Distance (ACD) values and consistent performance across low-confidence regions. Both architectures produce realistic and ambiguous images, but the updated one is more reliable, with no instances of GAN mode collapse. Additionally, the introduction of the Gaussian loss enhanced this architecture by allowing for adjustable tolerance in image generation around the decision boundary. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

Abstract
We introduce VEMOCLAP: Video EMOtion Classifier using Pretrained features, the first readily available and open-source web application that analyzes the emotional content of any user-provided video. We improve our previous work, which exploits open-source pretrained models that work on video frames and audio, and then efficiently fuse the resulting pretrained features using multi-head cross-attention. Our approach increases the state-of-the-art classification accuracy on the Ekman-6 video emotion dataset by 4.3% and offers an online application for users to run our model on their own videos or YouTube videos. We invite the readers to try our application at serkansulun.com/app.

Abstract
[No abstract available]

Abstract
The advantages of Virtual Reality (VR) over traditional training, together with the development of VR technology, have contributed to an increase in the body of literature on training professionals with VR. However, there is a gap in the literature concerning the comparison of training in a Virtual Environment (VE) with the same training in a Real Environment (RE), which would contribute to a better understanding of the capabilities of VR in training. This paper presents a study with firefighters (N = 12) where the effect of a firefighter training exercise in a VE was evaluated and compared to that of the same exercise in a RE. The effect of environments was evaluated using psychophysiological measures by evaluating the perception of stress and fatigue, transfer of knowledge, sense of presence, cybersickness, and the actual stress measured through participants' Heart Rate Variability (HRV). The results showed a similar perception of stress and fatigue between the two environments; a positive, although not significant, effect of the VE on the transfer of knowledge; the display of moderately high presence values in the VE; the ability of the VE not to cause symptoms of cybersickness; and finally, obtaining signs of stress in participants' HRV in the RE and, to a lesser extent, signs of stress in the VE. Although the effect of the VE was shown to be non-comparable to that of the RE, the authors consider the results encouraging and discuss some key factors that should be addressed in the future to improve the results of the training VE.

Abstract
This study introduces the development of a client-based software layer within the FGPE project, aimed at enhancing the usability of the FGPE programming learning environment through client-side processing. The primary goal is to enable the evaluation of programming exercises and the application of gamification rules directly on the client-side, thereby facilitating offline functionality. This approach is particularly beneficial in regions with unreliable internet connectivity, as it allows continuous student interaction and feedback without the need for a constant server connection. The implementation promises to reduce server load significantly by shifting the evaluation workload to the client-side. This not only improves response times but also alleviates the burden on server resources, enhancing overall system efficiency. Two main strategies are explored: 1) caching the gamification service interface on the client-side, and 2) implementing a complete client-side gamification service that synchronizes with the server when online. Each approach is evaluated in terms of its impact on user experience, system performance, and potential security concerns. The findings suggest that while client-side processing offers considerable benefits in terms of scalability and user engagement, it also introduces challenges such as increased system complexity and potential data synchronization issues. The study concludes with recommendations for balancing these factors to optimize the design and implementation of client-based systems for educational environments. © Ricardo Queirós, Robertas Damaševicius, Rytis Maskeliunas, and Jakub Swacha;

Abstract
DeepAR is a popular probabilistic time series forecasting algorithm. According to the authors, DeepAR is particularly suitable to build global models using hundreds of related time series. For this reason, it is a common expectation that DeepAR obtains poor results in univariate forecasting [10]. However, there are no empirical studies that clearly support this. Here, we compare the performance of DeepAR with standard forecasting models to assess its performance regarding 1 step-ahead forecasts. We use 100 time series from the M4 competition to compare univariate DeepAR with univariate LSTM and SARIMAX models, both for point and quantile forecasts. Results show that DeepAR obtains good results, which contradicts common perception. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.