Office sensor systems often fail to present environmental data related to office well-being in ways that are accessible and comprehensible for non-expert users. To address this, we introduce Situated Data Sensemaking, a concept that enhances individuals’ ability to understand and use data by embedding it within their physical environment. Implemented through OfficeSense, a system that physicalizes environmental parameters—light levels, sound, air quality (CO2), and temperature—within office spaces, this concept was evaluated in a 4-week study (N=11) including a baseline week. OfficeSense influenced perceived data literacy and understanding of office well-being. Findings indicate that Situated Data Sensemaking supported participants in aligning their perceptions of environmental conditions with real-time data, influenced by spatial and temporal contexts, personal preferences, and collaborative interactions. This study demonstrates that embedding data in shared physical spaces can enhance users’ confidence in interpreting environmental data and foster collective awareness of workplace conditions.
Despite their strong sequential modeling capabilities, recurrent neural networks (RNNs) remain vulnerable to noise, outlier-contaminated, and missing values in real-world time series, which undermines robustness and generalization. Conventional forecasting models struggle when inputs are corrupted, missing, or contaminated by outliers. Moreover, existing approaches that inject innovation signals either tie them to the output head, complicating training, or fail to provide gate-level corrective signals for broadly applicable forecasting. To address these limitations, we propose Error-Projected RNNs (EP-RNNs), a lightweight, architecture-agnostic modification that augments gated cells with a hidden-space innovation signal. Each gate receives a learnable corrective signal derived from the difference between the current input aligned proposal and the carried hidden state, decoupled from the output head and fully compatible with standard backpropagation through time. This mechanism enhances robustness, adaptability, and predictive accuracy under noise, missingness, outliers, and distribution shifts. We construct synthetic stress tests as controlled perturbation suites that add noise, induce missing values, and insert outliers, enabling us to evaluate how projected error aligns with realized future error and how it influences gate weights. We then validate EP-RNNs on these stress tests and on three real-world domains, including traffic flow, influenza-like illness, and stock market forecasting. Experimental results, evaluated using MSE, RMSE, and MAPE, demonstrate substantial and consistent improvements over standard LSTM, GRU, and BiLSTM models. Specifically, EP-LSTM achieves up to 83% improvement, EP-GRU up to 41%, and EP-BiLSTM up to 76% improvement on real-world datasets, highlighting the practical effectiveness of gate-level error modulation for reliable time series forecasting.
Transdisciplinary research (TDR) is increasingly recognized as an important approach to make research on complex environmental and sustainability issues such as climate risks and biodiversity loss more societally relevant and actionable. Deliberately designing early stages of TDR- phase-0 - is considered essential to its overall success. However, limited knowledge is available about how to shape and undertake phase-0 of TDR. We provide a detailed account of TDR processes in the WildlifeNL project, which were designed with explicit and extensive attention to phase-0. We saw that designing and shaping the project was a constant process of decisionmaking about delineation; defining consortium composition and participation of others outside the consortium ('who'), selecting cases and sites ('where'), and determining model systems, questions, and approaches ('what'). We demonstrate the interwovenness and interdependency between these dimensions based on several examples from WildlifeNL. The WildlifeNL team primarily focused on diverse stakeholder representation and building on existing partnerships. They thereby prioritized the 'who' dimension, while leaving 'what' and 'where' more flexible to adapt. The interconnectedness and interdependency confront those designing and implementing TDR with the question how to balance and prioritize the different dimensions, which defines how they make decisions on who, where and what are included and excluded in their TDR. We advocate for reflection and transparency about these messy and power-laden processes and suggest our framework as a structure for doing this. We argue that different TDR objectives and impact orientations ('why') may also prompt assigning different relative weight to the 'who', 'where' and 'what' dimensions.
How competent parents feel about their parenting and how children feel that they are heard and seen are key elements of a good cooperation between parents, children, and professionals. Based on a questionnaire study (N = 479), the views of parents, children, youth care professionals and special education teachers in diverse contexts were described. The findings suggest that professionals find it difficult to support parents' self-efficacy concerning parenting competences. Also, children feel less heard and seen then professionals estimate this. Professionals should focus more on supporting parenting skills, especially for at risk families, and give enduring attention to children's voice.