Traversable functors see widespread use in purely functional programming as an approach to the iterator pattern. Unlike other commonly used functor families, free constructions of traversable functors have not yet been described. Free constructions have previously found powerful applications in purely functional programming, as they embody the concept of providing the minimal amount of structure needed to create members of a complex family out of members of a simpler, underlying family. This thesis introduces Cofree Traversable Functors, together with a provably valid implementation, thereby developing a family of free constructions for traversable functors. As free constructions, cofree traversable functors may be used in order to create novel traversable functors from regular functors. Cofree traversable functors may also be leveraged in order to manipulate traversable functors generically.

Code reuse has the benefits of saving time and resources but poses a risk whenattempting to tailor copied code for a new purpose or in cases when such copies arebuggy or otherwise faulty. In the field of data science, the web application JupyterNotebook is a popular tool for creating computational notebooks, documentscontaining both plain text and code snippets, many of which are publicly available oncode hosting sites such as GitHub. This thesis describes the acquisition ofapproximately 2.6 million computational notebooks and analysis of this data set.By hashing the contents of every code snippet, using the MD5 hashing algorithm,cloned snippets were found through snippets producing identical hashes. Bysubsequently mapping the snippets to their corresponding notebooks, the relativeoriginality of a notebook could be determined. This analysis shows that nearly 95% ofnotebooks are written in some version of Python. Furthermore, nearly 54% ofnotebooks in the data set are comprised of code blocks also found in othernotebooks and, on average, approximately 70% of the code in any given notebookis copied from elsewhere.

With the advent of deep learning algorithms, the possibilities to analyse great amounts of data has been unlocked. As EEG-readers become both more affordable and of higher quality it enables interesting applications. In this thesis a convolutional neural network (CNN) is developed and implemented to evaluate the possibility to detect patterns in brain signals gathered from perceiving simple visual stimuli. Historically, brain signals are filtered into brain waves. These are cruder representations of brain signals than the raw data collected from the electrodes in the EEG-device. The software developed in this thesis enables the collection and structuring of raw EEG data such as it can be analysed by CNNs that classify images. The results show some emerging patterns, in contrast, to a completely noisy set of data and points towards the possibility for higher performance through further experimentation with data collection and modifications of the CNN. The deep learning pattern detection was implemented using the keras python library with tensorflow backend and open source software from openBCI. Data was collected from 14 participants using the 16-channel device (Cyton board + Daisy) in order to provide a proof of concept.

Wireless sensors offer the possibility to monitor critical parameters in our environment, which enables applications to optimize processes or anticipate and detect critical events. Environmental monitoring and predictive maintenance of non-electric systems are two important application domains that have different computational requirements but similar power constraints. In this thesis is presented an iterative wireless sensor node design that supports both applications equally well. In particular, the platform is useful for building demonstrators and evaluating proof of concept designs because the system can be used for the rapid prototyping of models out of standard machine learning frameworks with reasonable performance. At the same time, the platform can run for several years during the environmental monitoring with a battery. Additionally can the system be powered by solar harvesting to enable its use in a "deploy and forget" manner. For this purpose, the system hardware has been optimized and a radio module was selected which enables the transmission of measurements over several kilometers. To recommend the radio configuration for a minimal energy consumption, different settings have been compared in terms of their required energy and transmission range. The power budget of the platform has been generated and optimized, to increase the system run-time and enable the maximal amount of measurements within its energy constraints. Finally, the illumination in greenhouses has been analyzed which showed to provide enough energy to power the platform with a 45x15mm photovoltaic module. In combination with a single coin cell battery could be achieved a continuous system run-time of more than ten years for environmental monitoring applications with this platform.

Recent work in computer architecture and machine learning has seen various groups begin exploring the viability of using neural networks to augment conventional processor designs. Of particular interest is using the predictive capabilities of techniques in natural language processing to assist traditional CPU memory prefetching methods. This work demonstrates one of these proposed techniques, and examines some of the challenges associated with producing satisfactory and consistently reproducible results. Special attention is given to data acquisition and preprocessing as different methods. This is important since the handling training data can enormously influence on the final prediction accuracy of the network. Finally, a number of changes to improve these methods are suggested. These include ways to raise accuracy, reduce network overhead, and to improve the consistency of results. This work shows that augmenting an LSTM prefetcher with a simple stream prefetcher leads to moderate improvements in prediction accuracy. This could be a way to start reducing the size of neural networks so they are usable in real hardware.

he monad is a mathematical concept, used by Haskell to describe — among other things — Input/Output. Many are intimidated by it since it stems from abstract mathematics — namely Category Theory. However, the mathematics required to use and understand the monad is straight forward and intuitive, and can be explained through incremental definitions and proofs. This paper intends to construct and explain the monad from the ground up and show some example uses for it.

We develop a hybrid method to couple finite difference methods and finite element methods in a nonconforming multiblock fashion. The aim is to optimize computational efficiency when complex geometries present. The proposed coupling technique requires minimal changes in the existing schemes while maintaining strict stability, accuracy, and conservation. Analysis and computational results are shown for a linear problem (to the advection-diffusion equation) and a nonlinear problem (to the viscous Burger's equation) in two spatial dimensions

Chronic wounds are both a burden for patients and a major cost factor for a developed country's health budget. This research project investigates into designing IT systems for the treatment of chronic wounds by focusing on sharing, retrieving and entering information. To understand the context and the users, a pre-study was conducted followed by semi-structured interviews. The data gathered through the semi-structured interviews was analysed by applying a thematic analysis which resulted in five major themes. Based on these themes, problems were identified and solutions provided in the form of functional and non-functional requirements for an IT system. The major requirements are that the system should 1) provide fast and easy access to relevant information, 2) be easy to use, 3) adapt to the work environment, 4) reflect on established work processes and 5) focus on the user’s expectations and behaviour. A lowfidelity prototype was created based on the identified requirements and evaluated by co ducting a focus group. Overall, the feedback from the focus group was positive.

Cloud computing platforms lie at the very heart of today’s mobile and web-based applications. Cloud service providers must satisfy computational performance agreed through service level agreements (SLA) and simultaneously keep their operational costs, clusters health, and other cloud parameters within acceptable ranges in order to achieve business success. Using traditionally available monitoring tools is not sufficient to understand in depth how these different factors affect each other. Therefore, intelligent systems able to predict operational parameters from the usage behavior of a cloud data center can potentially be beneficial. This project aims to develop an algorithmic approach that models the relationship between cloud usage parameters such as CPU and memory usage and the cloud cluster’s health parameters such as temperature. Neural network models are trained using data from different machines, and experimental results show that the models deliver promising results in terms of modeling machines’ health parameters using usage parameters.

In this study a Deep Reinforcement Learning algorithm, MCTS-CNN, is applied on the Vehicle Routing Problem (VRP) in warehouses. Results in a simulated environment show that a Convolutional Neural Network (CNN) can be pre-trained on VRP transition state features and then effectively used post-training within Monte Carlo Tree Search (MCTS). When pre-training works well enough better results on warehouse VRP’s were often obtained than by a state of the art VRP Two-Phase algorithm. Although there are a number of issues that render current deployment pre-mature in two real warehouse environments MCTS-CNN shows high potential because of its strong scalability characteristics.