The Copernicus Sentinel-2 mission operated by the European Space Agency (ESA) provides comprehensive and continuous multi-spectral observations of all the Earth’s land surface since mid-2015. Clouds and cloud shadows significantly decrease the usability of optical satellite data, especially in agricultural applications; therefore, an accurate and reliable cloud mask is mandatory for effective EO optical data exploitation. During the last few years, image segmentation techniques have developed rapidly with the exploitation of neural network capabilities. With this perspective, the KappaMask processor using U-Net architecture was developed with the ability to generate a classification mask over northern latitudes into the following classes: clear, cloud shadow, semi-transparent cloud (thin clouds), cloud and invalid. For training, a Sentinel-2 dataset covering the Northern European terrestrial area was labelled. KappaMask provides a 10 m classification mask for Sentinel-2 Level-2A (L2A) and Level-1C (L1C) products. The total dice coefficient on the test dataset, which was not seen by the model at any stage, was 80% for KappaMask L2A and 76% for KappaMask L1C for clear, cloud shadow, semi-transparent and cloud classes. A comparison with rule-based cloud mask methods was then performed on the same test dataset, where Sen2Cor reached 59% dice coefficient for clear, cloud shadow, semi-transparent and cloud classes, Fmask reached 61% for clear, cloud shadow and cloud classes and Maja reached 51% for clear and cloud classes. The closest machine learning open-source cloud classification mask, S2cloudless, had a 63% dice coefficient providing only cloud and clear classes, while KappaMask L2A, with a more complex classification schema, outperformed S2cloudless by 17%.

Building electric energy is characterized by a significant increase in its uses (e.g., vehicle charging), a rapidly declining cost of all related data collection, and a proliferation of smart grid concepts, including diverse and flexible electricity pricing schemes. Not surprisingly, an increased number of approaches have been proposed for its modeling and forecasting. In this work, we place our emphasis on three forecasting-related issues. First, we look at the forecasting explainability, that is, the ability to understand and explain to the user what shapes the forecast. To this extent, we rely on concepts and approaches that are inherently explainable, such as the evolutionary approach of genetic programming (GP) and its associated symbolic expressions, as well as the so-called SHAP (SHapley Additive eXplanations) values, which is a well-established model agnostic approach for explainability, especially in terms of feature importance. Second, we investigate the impact of the training timeframe on the forecasting accuracy; this is driven by the realization that fast training would allow for faster deployment of forecasting in real-life solutions. And third, we explore the concept of counterfactual analysis on actionable features, that is, features that the user can really act upon and which therefore present an inherent advantage when it comes to decision support. We have found that SHAP values can provide important insights into the model explainability. In our analysis, GP models demonstrated superior performance compared to neural network-based models (with a 20–30% reduction in Root Mean Square Error (RMSE)) and time series models (with a 20–40% lower RMSE), but a rather questionable potential to produce crisp and insightful symbolic expressions, allowing a better insight into the model performance. We have also found and reported here on an important potential, especially for practical, decision support, of counterfactuals built on actionable features, and short training timeframes.

Energy demand forecasting is practiced in several time frames; different explanatory variables are used in each case to serve different decision support mandates. For example, in the short, daily, term building level, forecasting may serve as a performance baseline. On the other end, we have long-term, policy-oriented forecasting exercises. TIMES (an acronym for The Integrated Markal Efom System) allows us to model supply and anticipated technology shifts over a long-term horizon, often extending as far away in time as 2100. Between these two time frames, we also have a mid-term forecasting time frame, that of a few years ahead. Investigations here are aimed at policy support, although in a more mid-term horizon, we address issues such as investment planning and pricing. In this paper, we develop and evaluate statistical and neural network approaches for this mid-term forecasting of final energy and electricity for the residential sector in six EU countries (Germany, the Netherlands, Sweden, Spain, Portugal and Greece). Various possible approaches to model the explanatory variables used are presented, discussed, and assessed as to their suitability. Our end goal extends beyond model accuracy; we also include interpretability and counterfactual concepts and analysis, aiming at the development of a modelling approach that can provide decision support for strategies aimed at influencing energy demand.

As machine learning models evolve, maintaining transparency demands more human-centric explainable AI techniques. Counterfactual explanations, with roots in human reasoning, identify the minimal input changes needed to obtain a given output and, hence, are crucial for supporting decision-making. Despite their importance, the evaluation of these explanations often lacks grounding in user studies and remains fragmented, with existing metrics not fully capturing human perspectives. To address this challenge, we developed a diverse set of 30 counterfactual scenarios and collected ratings across 8 evaluation metrics from 206 respondents. Subsequently, we fine-tuned different Large Language Models (LLMs) to predict average or individual human judgment across these metrics. Our methodology allowed LLMs to achieve an accuracy of up to 63% in zero-shot evaluations and 85% (over a 3-classes prediction) with fine-tuning across all metrics. The fine-tuned models predicting human ratings offer better comparability and scalability in evaluating different counterfactual explanation frameworks.

Building electric energy is characterized by a significant increase of its uses (e.g. vehicle charging), a rapidly declining cost of all related data collection and a proliferation of smart grid concepts, including diverse and flexible electricity pricing schemes. Not surprisingly, an increased number of approaches have been proposed for its modeling and forecasting. In this work, we place our emphasis on three forecasting related issues. First, on the forecasting explainability, i.e. the ability to understand and explain to the user what shapes the forecast. To this extent we rely on concepts and approaches that are inherently explainable, such as the evolutionary approach of genetic programming (GP) and its associated symbolic expressions, as well as the so-called SHAP (SHapley Additive eXplanations) values, which is a well established model agnostic approach for explainability, especially in terms of feature importance. Second, we investigate the impact of the training timeframe on the forecasting accuracy; this is driven by the realization that a fast training would allow for faster deployment of forecasting in real life solutions. And third, we explore the concept of counterfactual analysis on actionable features, i.e. features that the user can really act upon and which therefore present an inherent advantage when it comes to decision support. We have found that SHAP values can provide important insights into the model explainability. As for GP models, we have found comparable and in some cases superior accuracy when compared to its neural-network and time-series counterparts but a rather questionable potential to produce crisp and insightful symbolic expressions, allowing a better insight into the model performance. We have also found, and report here on an important potential especially for practical, decision support, solutions of counterfactuals built on actionable features and short training timeframes.

A fundamental problem in cell and tissue biology is finding cells in microscopy images. Traditionally, this detection has been performed by segmenting the pixel intensities. However, these methods struggle to delineate cells in more densely packed micrographs, where local decisions about boundaries are not trivial. Here, we develop a new methodology to decompose microscopy images into individual cells by making object-level decisions. We formulate the segmentation problem as training a flexible factorized representation of the image. To this end, we introduce Kaizen, an approach inspired by predictive coding in the brain that maintains an internal representation of an image while generating object hypotheses over the external image, and keeping the ones that improve the consistency of internal and external representations. We achieve this by training a Vector Quantised-Variational AutoEncoder (VQ-VAE). During inference, the VQ-VAE is iteratively applied on locations where the internal representation differs from the external image, making new guesses, and keeping only the ones that improve the overall image prediction until the internal representation matches the input. We demonstrate Kaizen’s merits on two fluorescence microscopy datasets, improving the separation of nuclei and neuronal cells in cell culture images.