In recent times, pollution has emerged as a significant global concern, with European regulations stipulating limits on PM 2.5 particle levels. Addressing this challenge necessitates innovative approaches. Smart low-cost sensors suffer from imprecision, and can not replace legal stations in terms of accuracy, however, their potential to amplify the capillarity of air quality evaluation on the territory is not under discussion. In this paper, we propose an AI system to correct PM 2.5 levels in low-cost sensor data. Our research focuses on data from Turin, Italy, emphasizing the impact of humidity on low-cost sensor accuracy. In this study, different Neural Network architectures that vary the number of neurons per layer, consecutive records and batch sizes were used and compared to gain a deeper understanding of the network’s performance under various conditions. The AirMLP7-1500 model, with an impressive R-squared score of 0.932, stands out for its ability to correct PM 2.5 measurements. While our approach is tailored to the city of Turin, it offers a systematic methodology for the definition of those models and holds the promise to significantly improve the accuracy of air quality data collected from low-cost sensors, increasing the awareness of citizens and municipalities about this critical environmental information.

Generated Scalable Vector Graphics (SVG) images demand evaluation criteria tuned to their symbolic and vectorial nature: criteria that existing metrics such as FID, LPIPS, or CLIPScore fail to satisfy. In this paper, we introduce SVGauge, the first human-aligned, reference based metric for text-to-SVG generation. SVGauge jointly measures (i) visual fidelity, obtained by extracting SigLIP image embeddings and refining them with PCA and whitening for domain alignment, and (ii) semantic consistency, captured by comparing BLIP-2-generated captions of the SVGs against the original prompts in the combined space of SBERT and TF-IDF. Evaluation on the proposed SHE benchmark shows that SVGauge attains the highest correlation with human judgments and reproduces system-level rankings of eight zero-shot LLM-based generators more faithfully than existing metrics. Our results highlight the necessity of vector-specific evaluation and provide a practical tool for benchmarking future text-to-SVG generation models.

The rapid spread of misinformation poses a significant challenge in the digital age, with false claims appearing across multiple modalities, particularly combinations of textual and visual content such as images and videos. While automatic fact-checking plays a crucial role in countering misinformation, traditional approaches predominantly rely on textual data, often neglecting the multimodal nature of modern misinformation. In this survey, we provide a comprehensive evaluation of multimodal datasets designed for automatic fact-checking that combine textual and visual information, systematically analyzing their sources, annotation methodologies, and key statistical properties, such as class distribution, topic diversity, and label availability. Additionally, we assess the usability of these datasets in real-world scenarios, discussing their limitations, biases, and potential risks, such as information leakage. Motivated by the practical difficulties we encountered when attempting to integrate existing datasets into our own multimodal fact-checking pipeline, our work also offers concrete guidance to help researchers choose the most suitable resources. By identifying gaps and challenges in existing datasets, our survey aims to support the development of more reliable and scalable multimodal fact-checking systems.