Abstract
This paper address the use of deep learning approaches for visual based navigation in confined underwater environments. State-of-the-art algorithms have shown the tremendous potential deep learning architectures can have for visual navigation implementations, though they are still mostly outperformed by classical feature-based techniques. In this work, we apply current state-of-the-art deep learning methods for visual-based robot navigation to the more challenging underwater environment, providing both an underwater visual dataset acquired in real operational mission scenarios and an assessment of state-of-the-art algorithms on the underwater context. We extend current work by proposing a novel pose optimization architecture for the purpose of correcting visual odometry estimate drift using a Visual-Inertial fusion network, consisted of a neural network architecture anchored on an Inertial supervision learning scheme. Our Visual-Inertial Fusion Network was shown to improve results an average of 50% for trajectory estimates, also producing more visually consistent trajectory estimates for both our underwater application scenarios.

Abstract
Recent Low Power Wide Area Networks (LPWAN) protocols are receiving increased attention from industry and academia to offer accessibility for Internet of Things (IoT) connected remote sensors and actuators. In this work, we present a formal study of LoRaWAN security, an increasingly popular technology, which defines the structure and operation of LPWAN networks based on the LoRa physical layer. There are previously known security vulnerabilities in LoRaWAN that lead to the proposal of several improvements, some already incorporated into the latest protocol specification. Our analysis of LoRaWAN security uses Scyther, a formal security analysis tool and focuses on the key exchange portion of versions 1.0 (released in 2015) and 1.1 (the latest, released in 2017). For version 1.0, which is still the most widely deployed version of LoRaWAN, we show that our formal model allowed to uncover weaknesses that can be related to previously reported vulnerabilities. Our model did not find weaknesses in the latest version of the protocol (v1.1), and we discuss what this means in practice for the security of LoRaWAN as well as important aspects of our model and tools employed that should be considered. The Scyther model developed provides realistic models for LoRaWAN v1.0 and v1.1 that can be used and extended to formally analyze, inspect, and explore the security features of the protocols. This, in turn, can clarify the methodology for achieving secrecy, integrity, and authentication for designers and developers interested in these LPWAN standards. We believe that our model and discussion of the protocols security properties are beneficial for both researchers and practitioners. To the best of our knowledge, this is the first work that presents a formal security analysis of LoRaWAN.

Abstract
This study describes a novel dataset with retinal image quality annotation, defined by three different retinal experts, and presents an inter-observer analysis for quality assessment that can be used as gold-standard for future studies. A state-of-the-art algorithm for retinal image quality assessment is also analysed and compared against the specialists performance. Results show that, for 71% of the images present in the dataset, the three experts agree on the given image quality label. The results obtained for accuracy, specificity and sensitivity when comparing one expert against another were in the ranges [83.0 - 85.2]%, [72.7 - 92.9]% and [80.0 - 94.7]%, respectively. The evaluated automatic quality assessment method, despite not being trained on the novel dataset, presents a performance which is within inter-observer variability.

Abstract
The ever-increasing demand for higher quality live streams is driving the need for better networking infrastructures, specially when disseminating content over highly congested areas, such as stadiums, concerts and museums. Traditional approaches to handle this type of scenario relies on a combination of cellular data, through 4G distributed antenna arrays (DAS), with a high count of WiFi (802.11) access points. This obvious requires a substantial upfront cost for equipment, planning and deployment. Recently, new efforts have been introduced to securely leverage the capabilities of wireless multipath, including WiFi multicast, 4G, and device-to-device communications. In order to solve these issues, we propose an approach that lessens the requirements imposed on the wireless infrastructures while potentially expanding wireless coverage through the crowd-sourcing of mobile devices. In order to achieve this, we propose a novel pervasive approach that combines secure distributed systems, WiFi multicast, erasure coding, source coding and opportunistic offloading that makes use of hyperlocal mobile edge clouds. We empirically show that our solution is able to offer a 11 fold reduction on the infrastructural WiFi bandwidth usage without having to modify any existing software or firmware stacks while ensuring stream integrity, authorization and authentication.

Abstract
In this paper, we identify a subset of types in the rank 2 intersection type system, where types do not contain positive occurrences of intersections. We extend the notion of pre-grammar of a type and address the type-inhabitation problem for types in this subset, as well as their intersections.

Abstract
Zooplankton plays a key -role on Earth's ecosystem, emerging in the oceans and rivers in great quantities and diversity, making it an important and rather common topic on scientific studies. Given the numbers of different species it is not only necessary to study their numbers but also their classification. In this paper a possible solution for the zooplankton in situ detection and classification problem in real-time is proposed using a portable deep learning approach based on Convolutional Neural Networks deployed on INESC TEC's MarinEye system. For detection a Single Shot Detection model with MobileNet was used, and ZooplanktoNet for classification. System portability is guaranteed with the use of MovidiusTMNeural Compute Stick as the deep learning motor.