Abstract
Gastric intestinal metaplasia (GIM) characterization is challenging for humans and AI models. Deep learning solutions for this task are sensitive to training data, which is particularly concerning given the wide range of acquisition conditions, sampling biases, and overall scarcity of high-quality datasets.In this paper, we set forth the GIM self-similarity hypothesis where we assume that an underlying stationary self-similar process governs the structural changes observed in the mucosa. To validate this hypothesis we show that a deep learning model can map an adequately placed patch to the endoscopic grading of GIM (EGGIM) of the entire still frame.To evaluate our approach, we collected and annotated both retrospective and prospective datasets with EGGIM scores. Our results are promising: using leave-one-patient-out cross-validation, the predictions from a ResNet-50 model can be used to correctly stratify the risk for 57 out of 65 patients with perfect sensitivity on an extremely biased dataset.

Abstract
Indoor Positioning Systems (IPSs) are essential for applications requiring accurate location awareness in indoor environments. However, achieving high precision remains challenging due to signal interference and environmental variability. This study proposes a multimodal IPS that integrates Bluetooth Received Signal Strength Indicator (RSSI) measurements and video imagery using machine learning (ML) and ensemble learning techniques. The system was implemented and deployed in the Hall of Biodiversity at the Natural History and Science Museum of the University of Porto. The venue presented significant deployment issues, namely restrictions on beacon placement and lighting conditions. We trained independent ML models on RSSI and video datasets, and combined them through ensemble learning methods. The experimental results from test scenarios, which included simulated visitor trajectories, showed that ensemble models consistently outperformed the RSSI-based and video-based models. These findings demonstrate that the use of multimodal data can significantly improve IPS accuracy despite constraints such as multipath interference, low lighting, and limited beacon infrastructure. Overall, they highlight the potential of multimodal data for deployments in complex indoor environments.

Abstract
Recent verification tools aim to make formal verification more accessible for software engineers by automating most of the verification process. However, the manual work and expertise required to write verification helper code, such as loop invariants and auxiliary lemmas and assertions, remains a barrier. This paper explores the use of Large Language Models (LLMs) to automate the generation of loop invariants for programs in Dafny. We tested the approach on a curated dataset of 100 programs in Dafny involving arrays, strings, and numeric types. Using a multimodel approach that combines GPT-4o and Claude 3.5 Sonnet, correct loop invariants (passing the Dafny verifier) were generated at the first attempt for 92% of the programs, and in at most five attempts for 95% of the programs. Additionally, we developed an extension to the Dafny plugin for Visual Studio Code to incorporate automatic loop invariant generation into the IDE. Our work stands out from related approaches by handling a broader class of problems and offering IDE integration.

Abstract
[No abstract available]

Abstract

Abstract