On the use of CNNs with patterned stride for medical image analysis

The use of deep learning techniques for early and accurate medical image diagnosis has grown significantly in recent years, with some encouraging results across many medical specialties, pathologies, and image types. One of the most popular deep neural network architectures is the convolutional neural network (CNN), widely used for medical image classification and segmentation, among other tasks. One of the configuration parameters of a CNN is called stride and it regulates how sparsely the image is sampled during the convolutional process. This paper explores the idea of applying a patterned stride strategy: pixels closer to the center are processed with a smaller stride concentrating the amount of information sampled, and pixels away from the center are processed with larger strides consequently making those areas to be sampled more sparsely. We apply this method to different medical image classification tasks and demonstrate experimentally how the proposed patterned stride mechanism outperforms a baseline solution with the same computational cost (processing and memory). We also discuss the relevance and potential future extensions of the proposed method.

LightGuide

Precise and reliable directional feedback is crucial for electronic traveling aids that guide visually impaired people along safe paths. A large proportion of visually impaired people can determine light position using their light perception. This work presents LightGuide, a directional feedback solution that indicates a safe direction of travel via the position of a light within the user's visual field. We prototyped LightGuide using an LED strip attached to the brim of a cap, and conducted three user studies to explore the effectiveness of LightGuide compared to HapticBag, a state-of-the-art baseline solution that indicates directions through on-shoulder vibrations. Results showed that, with LightGuide, participants turned to target directions in place more quickly and smoothly, and navigated along basic and complex paths more efficiently, smoothly, and accurately than HapticBag. Users' subjective feedback implied that LightGuide was easy to learn and intuitive to use. The potential limitations of using LightGuide in real environments are subsequently discussed.

Privacy-aware character pattern matching over outsourced encrypted data

Providing a method to efficiently search into outsourced encrypted data, without forsaking strong privacy guarantees, is a pressing concern rising from the separation of data ownership and data management typical of cloud-based applications. While several existing solutions allow a client to look-up the occurrences of a substring in an outsourced document collection, the practical application requirements in terms of privacy and efficiency call for the improvement of such solutions. In this work, we present a privacy-preserving substring search protocol with a polylogarithmic communication cost and a limited computational effort on the server side. The proposed protocol provides search pattern and access pattern privacy, for both exact string search, and character-pattern search with wildcards. Its extension to a multi-user setting shows significant savings in terms of outsourced storage w.r.t. a baseline solution where the whole dataset is replicated. The performance figures of an optimized implementation of our protocol, searching into a remotely stored genomic dataset, validate the practicality of the approach exhibiting a data transfer of less than 50 kiB to execute a query over a document of 40 MiB, with execution times on client and server in the range of a few seconds and a few minutes, respectively.

An Improved Relative GNSS Tracking Method Utilizing Single Frequency Receivers

The Global Navigation Satellite Systems (GNSS) becomes the primary choice for device localization in outdoor situations. At the same time, many applications do not require precise absolute Earth coordinates, but instead, inferring the geometric configuration information of the constituent nodes in the system by relative positioning. The Real-Time Kinematic (RTK) technique shows its efficiency and accuracy in calculating the relative position. However, when the cycle slips occur, the RTK method may take a long time to obtain a fixed ambiguity value, and the positioning result will be a “float” solution with a low meter accuracy. The novel method presented in this paper is based on the Relative GNSS Tracking Algorithm (Regtrack). It calculates the changes in the relative baseline between two receivers without an ambiguity estimation. The dead reckoning method is used to give out the relative baseline solution while a parallel running Extended Kalman Filter (EKF) method reinitiates the relative baseline when too many validation failures happen. We conducted both static and kinematic tests to assess the performance of the new methodology. The experimental results show that the proposed strategy can give accurate millimeter-scale solutions of relative motion vectors in adjacent two epochs. The relative baseline solution can be sub-decimeter level with or without the base station is holding static. In the meantime, when the initial tracking point and base station coordinates are precisely obtained, the tracking result error can be only 40 cm away from the ground truth after a 25 min drive test in an urban environment. The efficiency test shows that the proposed method can be a real-time method, the time that calculates one epoch of measurement data is no more than 80 ms and is less than 10 ms for best results. The novel method can be used as a more robust and accurate ambiguity free tracking approach for outdoor applications.

Water Resource Systems Analysis for Water Scarcity Management: The Thames Water Case Study

Optimisation tools are a practical solution to problems involving the complex and interdependent constituents of water resource systems and offer the opportunity to engage with practitioners as an integral part of the optimisation process. A multiobjective genetic algorithm is employed in conjunction with a detailed water resource model to optimise the “Lower Thames Control Diagram”, a set of control curves subject to a large number of constraints. The Diagram is used to regulate abstraction of water for the public drinking water supply for London, UK, and to maintain downstream environmental and navigational flows. The optimisation is undertaken with the aim of increasing the amount of water that can be supplied (deployable output) through solely operational changes. A significant improvement of 33 Ml/day (1% or £59.4 million of equivalent investment in alternative resources) of deployable output was achieved through the optimisation, improving the performance of the system whilst maintaining the level of service constraints without negatively impacting on the amount of water released downstream. A further 0.2% (£11.9 million equivalent) was found to be realisable through an additional low-cost intervention. A more realistic comparison of solutions indicated even larger savings for the utility, as the baseline solution did not satisfy the basic problem constraints. The optimised configuration of the Lower Thames Control Diagram was adopted by the water utility and the environmental regulators and is currently in use.

On the Validity of Scaling Transient Conjugate Heat Transfer Characteristics

Accurate prediction of unsteady thermal loads is of paramount importance in several engineering disciplines and applications. Performing time-accurate unsteady conjugate heat transfer (CHT) simulations presents considerable challenges due to the markedly different time scales between the solid and fluid domains. Two methods have been recently proposed, aimed at addressing this issue: multiscale modeling (MSM) and equalized time-scales (ET). The former is based on the separation of the disparate short and long temporal scales of the solution and subsequent averaging of the flow/energy equations. In the latter, the equalization of the time scales is achieved through manipulation of the solid's thermal properties. Both methods are very appealing due to the possibility of being easily implemented on an existing solver. It becomes, thus, relevant to assess their performance and/or limitations. This paper work presents a comparative study of the two methods for the prediction of transient thermal load, first using a simplified case of a solid body with uniform temperature, then through the investigation of the prewarming phase of a steam turbine. Both methods are then compared against a reference baseline fully coupled (FC) CHT solution. The results show how the MSM allows greater accuracy and robustness with considerable saving in computational cost with respect to the baseline solution.