Användningen av artificiell intelligens (AI) har tredubblats på ett år och och anses av vissa vara det viktigaste paradigmskiftet i teknikhistorien. Den rådande AI-kapplöpningen riskerar att underminera frågor om etik och hållbarhet, vilket kan ge förödande konsekvenser. Artificiell intelligens har i flera fall visat sig avbilda, och till och med förstärka, befintliga snedvridningar i samhället i form av fördomar och värderingar. Detta fenomen kallas algoritmisk snedvridning (algorithmic bias). Denna studie syftar till att formulera ett ramverk för att minimera risken att algoritmisk snedvridning uppstår i AI-projekt och att anpassa det efter ett medelstort konsultbolag. Studiens första del är en litteraturstudie på snedvridningar - både ur ett kognitivt och ur ett algoritmiskt perspektiv. Den andra delen är en undersökning av existerande rekommendationer från EU, AI Sustainability Center, Google och Facebook. Den tredje och sista delen består av ett empiriskt bidrag i form av en kvalitativ intervjustudie, som har använts för att justera ett initialt ramverk i en iterativ process. / In the use of the third generation Artificial Intelligence (AI) for the development of products and services, there are many hidden risks that may be difficult to detect at an early stage. One of the risks with the use of machine learning algorithms is algorithmic bias which, in simplified terms, means that implicit prejudices and values are comprised in the implementation of AI. A well-known case is Google’s image recognition algorithm, which identified black people as gorillas. The purpose of this master thesis is to create a framework with the aim to minimise the risk of algorithmic bias in AI development projects. To succeed with this task, the project has been divided into three parts. The first part is a literature study of the phenomenon bias, both from a human perspective as well as from an algorithmic bias perspective. The second part is an investigation of existing frameworks and recommendations published by Facebook, Google, AI Sustainability Center and the EU. The third part consists in an empirical contribution in the form of a qualitative interview study which has been used to create and adapt an initial general framework. The framework was created using an iterative methodology where two whole iterations were performed. The first version of the framework was created using insights from the literature studies as well as from existing recommendations. To validate the first version, the framework was presented for one of Cybercom’s customers in the private sector, who also got the possibility to ask questions and give feedback regarding the framework. The second version of the framework was created using results from the qualitative interview studies with machine learning experts at Cybercom. As a validation of the applicability of the framework on real projects and customers, a second qualitative interview study was performed together with Sida - one of Cybercom’s customers in the public sector. Since the framework was formed in a circular process, the second version of the framework should not be treated as constant or complete. The interview study at Sida is considered the beginning of a third iteration, which in future studies could be further developed.

The purpose of this study was to gain a greater understanding of potential challenges from a moral stress perspective that Swedish military leaders can face when implementing autonomous weapons systems. Two questions were asked to investigate this: what demands may arise and how can leaders be impacted? The study was conducted as a literature study and data from nine peer reviewed articles and a research report from the Swedish Defense Research Institute were analyzed via thematic analysis. The result seems to show that the lack of control, the lack of trust and difficulty in demanding responsibility from an autonomous weapon system creates moral leadership challenges. Without control over the autonomous weapon system, the consequences of its actions risk going against the leader's moral and this creates problems with how leadership can and should be conducted. The study also seems to show that autonomous weapon systems can lead to a moral impact on leaders, since autonomous weapon systems risk leading to increased distancing and risking contributing to increased violence. Given the moral leadership challenges and the moral influence made visible in this study, there seems to be a great risk of moral stress and even moral injury if autonomous weapon systems are used for actions that go against the leader's morality.

The focus of this study was deep learning. A small, autonomous robot car was used for obstacle avoidance experiments. The robot car used a camera for taking images of its surroundings. A convolutional neural network used the images for obstacle detection. The available dataset of 31 022 images was trained with the Xception model. We compared two different implementations for making the robot car avoid obstacles. Mapping image classes to steering commands was used as a reference implementation. The main implementation of this study was to separate obstacle detection and steering logic in different modules. The former reached an obstacle avoidance ratio of 80 %, the latter reached 88 %. Different hyperparameters were looked at during training. We found that frozen layers and number of epochs were important to optimize. Weights were loaded from ImageNet before training. Frozen layers decided how many layers that were trainable after that. Training all layers (no frozen layers) was proven to work best. Number of epochs decided how many epochs a model trained. We found that it was important to train between 10-25 epochs. The best model used no frozen layers and trained for 21 epochs. It reached a test accuracy of 85.2 %.

Over the last decade, data has become a highly valuable resource. Electrical power grids deal with large quantities of data, and continuously collect this for analytical purposes. Anomalies that occur within this data is important to identify since they could cause nonoptimal performance within the substations, or in worse cases damage to the substations themselves. However, large datasets in the order of millions are hard or even impossible to gain a reasonable overview of the data manually. When collecting data from electrical power grids, predefined triggering criteria are often used to indicate that an event has occurred within the specific system. This makes it difficult to search for events that are unknown to the operator of the deployed acquisition system. Clustering, an unsupervised machine learning method, can be utilised for fault prediction within systems generating large amounts of multivariate time-series data without labels and can group data more efficiently and without the bias of a human operator. A large number of clustering techniques exist, as well as methods for extracting information from the data itself, and identification of these was of utmost importance. This thesis work presents a study of the methods involved in the creation of such a clustering system which is suitable for the specific type of data. The objective of the study was to identify methods that enables finding the underlying structures of the data and cluster the data based on these. The signals were split into multiple frequency sub-bands and from these features could be extracted and evaluated. Using suitable combinations of features the data was clustered with two different clustering algorithms, CLARA and CLARANS, and evaluated with established quality analysis methods. The results indicate that CLARA performed overall best on all the tested feature sets. The formed clusters hold valuable information such as indications of unknown events within the system, and if similar events are clustered together this can assist a human operator further to investigate the importance of the clusters themselves. A further conclusion from the results is that research into the use of more optimised clustering algorithms is necessary so that expansion into larger datasets can be considered.

With self-driving cars on the horizon, vehicle autonomy and its problems is a hot topic. In this study we are using convolutional neural networks to make a robot car avoid obstacles. The robot car has a monocular camera, and our approach is to use the images taken by the camera as input, and then output a steering command. Using this method the car is to avoid any object in front of it. In order to lower the amount of training data we use models that are pretrained on ImageNet, a large image database containing millions of images. The model are then trained on our own dataset, which contains of images taken directly by the robot car while driving around. The images are then labeled with the steering command used while taking the image. While training we experiment with using different amounts of frozen layers. A frozen layer is a layer that has been pretrained on ImageNet, but are not trained on our dataset. The Xception, MobileNet and VGG16 architectures are tested and compared to each other. We find that a lower amount of frozen layer produces better results, and our best model, which used the Xception architecture, achieved 81.19% accuracy on our test set. During a qualitative test the car avoid collisions 78.57% of the time.

Volume measurements of timber loads is done in conjunction with timber trade. When dealing with goods of major economic values such as these, it is important to achieve an impartial and fair assessment when determining price-based volumes. With the help of Saab’s missile targeting technology, CIND AB develops products for digital volume measurement of timber loads. Currently there is a system in operation that automatically reconstructs timber trucks in motion to create measurable images of them. Future iterations of the system is expected to fully automate the scaling by generating a volumetric representation of the timber and calculate its external gross volume. The first challenge towards this development is to separate the timber load from the truck. This thesis aims to evaluate and implement appropriate method for semantic pixel-wise segmentation of timber loads in real time. Image segmentation is a classic but difficult problem in computer vision. To achieve greater robustness, it is therefore important to carefully study and make use of the conditions given by the existing system. Variations in timber type, truck type and packing together create unique combinations that the system must be able to handle. The system must work around the clock in different weather conditions while maintaining high precision and performance.

Large collections of historical document images have been collected by companies and government institutions for decades. More recently, these collections have been made available to a larger public via the Internet. However, to make accessing them truly useful, the contained images need to be made readable and searchable. One step in that direction is document image binarization, the separation of text foreground from page background. This separation makes the text shown in the document images easier to process by humans and other image processing algorithms alike. While reasonably well working binarization algorithms exist, it is not sufficient to just being able to perform the separation of foreground and background well. This separation also has to be achieved in an efficient manner, in terms of execution time, but also in terms of training data used by machine learning based methods. This is necessary to make binarization not only theoretically possible, but also practically viable. In this thesis, we explore different ways to achieve efficient binarization in terms of execution time by improving the implementation and the algorithm of a state-of-the-art binarization method. We find that parameter prediction, as well as mapping the algorithm onto the graphics processing unit (GPU) help to improve its execution performance. Furthermore, we propose a binarization algorithm based on recurrent neural networks and evaluate the choice of its design parameters with respect to their impact on execution time and binarization quality. Here, we identify a trade-off between binarization quality and execution performance based on the algorithm’s footprint size and show that dynamically weighted training loss tends to improve the binarization quality. Lastly, we address the problem of training data efficiency by evaluating the use of interactive machine learning for reducing the required amount of training data for our recurrent neural network based method. We show that user feedback can help to achieve better binarization quality with less training data and that visualized uncertainty helps to guide users to give more relevant feedback. / Scalable resource-efficient systems for big data analytics

In mineral processing the grinding comminution process is an integral part since it is often the bottleneck of the concentrating process, thus small improvements may lead to large savings. By implementing a Reinforcement Learning controller this thesis aims to investigate if it is possible to control the grinding process more efficiently compared to traditional control strategies. Based on a calibrated plant simulation we compare existing control strategies with Proximal Policy Optimization and show possible increase in profitability under certain conditions.

In recent years, development of Convolutional Neural Networks has enabled high performing semantic segmentation models. Generally, these deep learning based segmentation methods require a large amount of annotated data. Acquiring such annotated data for semantic segmentation is a tedious and expensive task. Within machine learning, active learning involves in the selection of new data in order to limit the usage of annotated data. In active learning, the model is trained for several iterations and additional samples are selected that the model is uncertain of. The model is then retrained on additional samples and the process is repeated again. In this thesis, an active learning framework has been applied to road segmentation which is semantic segmentation of objects related to road scenes. The uncertainty in the samples is estimated with Monte Carlo dropout. In Monte Carlo dropout, several dropout masks are applied to the model and the variance is captured, working as an estimate of the model’s uncertainty. Other metrics to rank the uncertainty evaluated in this work are: a baseline method that selects samples randomly, the entropy in the default predictions and three additional variations/extensions of Monte Carlo dropout. Both the active learning framework and uncertainty estimation are implemented in the thesis. Monte Carlo dropout performs slightly better than the baseline in 3 out of 4 metrics. Entropy outperforms all other implemented methods in all metrics. The three additional methods do not perform better than Monte Carlo dropout. An analysis of what kind of uncertainty Monte Carlo dropout capture is performed together with a comparison of the samples selected by baseline and Monte Carlo dropout. Future development and possible improvements are also discussed.

Deep learning has dominated the computer vision field since 2012, but a common criticism of deep learning methods is their dependence on large amounts of data. To combat this criticism research into data-efficient deep learning is growing. The foremost success in data-efficient deep learning is transfer learning with networks pre-trained on the ImageNet dataset. Pre-trained networks have achieved state-of-the-art performance on many tasks. We consider the pre-trained network method for a new task where we have to collect the data. We hypothesize that the data efficiency of pre-trained networks can be improved through informed data collection. After exhaustive experiments on CaffeNet and VGG16, we conclude that the data efficiency indeed can be improved. Furthermore, we investigate an alternative approach to data-efficient learning, namely adding domain knowledge in the form of a spatial transformer to the pre-trained networks. We find that spatial transformers are difficult to train and seem to not improve data efficiency.