Ceramic materials contain several defects, one of which is porosity. At the time of writing, porosity measurement is a manual and time-consuming process performed by a human operator. With advances in deep learning for computer vision, this thesis explores to what degree convolutional neural networks and semantic segmentation can reliably measure porosity from microscope images. Combining classical image processing techniques with deep learning, images were automatically labeled and then used for training semantic segmentation neural networks leveraging transfer learning. Deep learning-based methods were more robust and could more reliably identify porosity in a larger variety of images than solely relying on classical image processing techniques.

This thesis is concerned with the topic of person re-identification. Many tracking algorithms today cannot keep track of players reentering the scene from different angles and times. Therefore, in this thesis, current literature is explored to gather information about the topic, and a current state-of-the-art model is tested. The person re-identification techniques will be applied to Padel games due to the collaboration with PadelPlay AB. The purpose of the thesis is to keep track of players during full matches of Padel with correct identities. To this, a current state-of-the-art model is applied to an existing tracking algorithm to enhance its capabilities.  Furthermore, the purpose is broken down into two research questions. Firstly, how well does an existing person re-id model perform on Padel matches when it comes to keeping a consistent and accurate id on all players. Secondly, how can this model be improved upon to perform better in the new domain, being the sport of Padel? To be able to answer the research questions, a Padel dataset is created for benchmarking purposes. The state-of-the-art model is tested on the new dataset to see how it handles a new domain. Additionally, the same state-of-the-art model is retrained on the Padel dataset to answer the second research question.  The results show that the state-of-the-art model that is previously trained on the Market-1501 dataset is highly generalizable on the Padel dataset and performs closely to the new model that is purely trained on the Padel dataset. Although they perform alike, the new model trained on the Padel dataset is slightly better as seen through both the quantitative and qualitative evaluations. Furthermore, the application of re-identification technology to keep track of players yielded significantly higher results than conventional solutions such as YOLOv5 with Deepsort.

With the recent advantages of machine learning in cheminformatics, the drug discovery process has been accelerated; providing a high impact in the field of medicine and public health. Molecular property and activity prediction are key elements in the early stages of drug discovery by helping prioritize the experiments and reduce the experimental work. In this thesis, a novel approach for multi-task regression using a text-based Transformer model is introduced and thoroughly explored for training on a number of properties or activities simultaneously. This multi-task regression with Transformer based model is inspired by the field of Natural Language Processing (NLP) which uses prefix tokens to distinguish between each task. In order to investigate our architecture two data categories are used; 133 biological activities from ExCAPE database and three physical chemistry properties from MoleculeNet benchmark datasets. The Transformer model consists of the embedding layer with positional encoding, a number of encoder layers, and a Feedforward Neural Network (FNN) to turn it into a regression problem. The molecules are represented as a string of characters using the Simplified Molecular-Input Line-Entry System (SMILES) which is a ’chemistry language’ with its own syntax. In addition, the effect of Transfer Learning is explored by experimenting with two pretrained Transformer models, pretrained on 1.5 million and on 100 million molecules. The text-base Transformer models are compared with a feature-based Support Vector Regression (SVR) with the Tanimoto kernel where the input molecules are encoded as Extended Connectivity Fingerprint (ECFP), which are calculated features. The results have shown that Transfer Learning is crucial for improving the performance on both property and activity predictions. On bioactivity tasks, the larger pretrained Transformer on 100 million molecules achieved comparable performance to the feature-based SVR model; however, overall SVR performed better on the majority of the bioactivity tasks. On the other hand, on physicochemistry property tasks, the larger pretrained Transformer outperformed SVR on all three tasks. Concluding, the multi-task regression architecture with the prefix token had comparable performance with the traditional feature-based approach on predicting different molecular properties or activities. Lastly, using the larger pretrained models trained on a wide chemical space can play a key role in improving the performance of Transformer models on these tasks.

Transformer-based architectures have in recent years advanced state-of-the-art performance in Natural Language Processing. Researchers have successfully adapted such models to downstream tasks within NLP in a domain-specific setting. This thesis examines the application of these models to the legal domain by doing Named Entity Recognition (NER) in a setting of scarce training data. Three different pre-trained BERT models are fine-tuned on a set of 101 court case documents, whereof one model is pre-trained on legal corpora and the other two on general corpora. Experiments are run to evaluate the models’ predictive performance given smaller or larger quantities of data to fine-tune on. Results show that BERT models work reasonably well for NER with legal data. Unlike many other domain-specific BERT models, the BERT model trained on legal corpora does not outperform the base models. Modest amounts of annotated data seem sufficient for reasonably good performance.

I detta arbete genomförs utvecklingen av ett realtids objektdetekteringssystem för iOS. För detta ändamål används YOLO, en ett-stegs objektdetekterare och ett s.k. ihoplänkat neuralt nätverk vilket åstadkommer betydligt bättre prestanda än övriga realtidsdetek- terare i termer av hastighet och precision. En dörrdetekterare baserad på YOLO tränas och implementeras i en systemutvecklingsprocess. Maskininlärningsprocessen sammanfat- tas och praxis för att undvika överträning eller “overfitting” samt för att öka precision och hastighet diskuteras och appliceras. Vidare genomförs en rad experiment vilka pekar på att dataaugmentation och inkludering av negativ data i ett dataset medför ökad precision. Hyperparameteroptimisering och kunskapsöverföring pekas även ut som medel för att öka en objektdetekringsmodells prestanda. Författaren lyckas öka modellens mAP, ett sätt att mäta precision för objektdetekterare, från 63.76% till 86.73% utifrån de erfarenheter som dras av experimenten. En modells tendens för överträning utforskas även med resultat som pekar på att träning med över 300 epoker rimligen orsakar en övertränad modell. / This paper features the creation of a real time object detection system for mobile iOS using YOLO, a state-of-the-art one stage object detector and convoluted neural network far surpassing other real time object detectors in speed and accuracy. In this process an object detecting model is trained to detect doors. The machine learning process is outlined and practices to combat overfitting and increasing accuracy and speed are discussed. A series of experiments are conducted, the results of which suggests that data augmentation, including negative data in a dataset, hyperparameter optimisation and transfer learning are viable techniques in improving the performance of an object detection model. The author is able to increase mAP, a measurement of accuracy for object detectors, from 63.76% to 86.73% based on the results of experiments. The tendency for overfitting is also explored and results suggest that training beyond 300 epochs is likely to produce an overfitted model.

No description available.

No description available.

With more aircraft in the air than ever before, there is a need for automating the surveillance of the airspace. It is widely known that aircraft with different intentions fly in different flight patterns. Support systems for finding different flight patterns are therefore needed. In this thesis, we investigate the possibility of detecting circular flight patterns using deep learning models. The basis for detection is ADS-B data which is continuously transmitted by aircraft containing information related to the aircraft status. Two deep learning models are constructed to solve the binary classification problem of detecting circular flight patterns. The first model is a Long Short-Term Memory (LSTM) model and utilizes techniques such as sliding window and bidirectional LSTM layers to solve the given task. The second model is a Convolutional Neural Network (CNN) and utilizes transfer learning. For the CNN model, the trajectory data is converted into image representations which are fed into a pre-trained model with a custom final dense layer. While ADS-B is openly available, finding specific flight patterns and producing a labeled data set of that pattern is hard and time-consuming. The data set is therefore expanded using other sources of data. Two additional sources of trajectory data are added to the data set; radar and simulated data. Training a model on data of a different distribution than the model is being evaluated on can be problematic and introduces a new source of error known as training-validation mismatch. One of the main goals of this thesis is to be able to quantify the size of this error to decide if using data from other sources is a viable option. The results show that the CNN model outperforms the LSTM model and achieves an accuracy of 98.2%. The results also show that there is a cost, in terms of accuracy, associated with not only training on ADS-B data. For the CNN model that cost was a 1-4% loss in accuracy depending on the training data used. The corresponding cost for the LSTM model was 2-10%.

Training a deep neural network from scratch can be very expensive in terms of resources.In addition, training a neural network on a new task is usually done by training themodel form scratch. Recently there are new approaches in machine learning which usesthe knowledge from a pre-trained deep neural network on a new task. The technique ofreusing the knowledge from previously trained deep neural networks is called Transferlearning. In this paper we are going to evaluate transfer learning capabilities of deep neuralnetwork architectures for image classification. This research attempts to implementtransfer learning with different datasets and models in order to investigate transfer learningin different situations.

Melanoma is a life threatening type of skin cancer with numerous fatal incidences all over the world. The 5-year survival rate is very high for cases that are diagnosed in early stage. So, early detection of melanoma is of vital importance. Except for several techniques that clinicians apply so as to improve the reliability of detecting melanoma, many automated algorithms and mobile applications have been developed for the same purpose.In this paper, deep learning model designed from scratch as well as the pretrained models Inception v3 and VGG-16 are used with the aim of developing a reliable tool that can be used for melanoma detection by clinicians and individual users. Dermatologists who use dermoscopes can take advantage of the algorithms trained on dermoscopical images and acquire a confirmation about their diagnosis. On the other hand, the models trained on clinical images can be used on mobile applications, since a cell phone camera takes images similar to them.The results using Inception v3 model for dermoscopical images achieved accuracy 91.4%, sensitivity 87.8% and specificity 92.3%. For clinical images, the VGG-16 model achieved accuracy 86.3%, sensitivity 84.5% and specificity 88.8%. The results are compared to those of clinicians, which shows that the algorithms can be used reliably for the detection of melanoma.