Browsing by Subject "deep learning"

Sort by: Order: Results:

Now showing items 1-17 of 17
  • Airaksinen, Manu; Juvela, Lauri; Alku, Paavo; Rasanen, Okko (IEEE, 2019)
    International Conference on Acoustics Speech and Signal Processing ICASSP
    This study explores various speech data augmentation methods for the task of noise-robust fundamental frequency (F0) estimation with neural networks. The explored augmentation strategies are split into additive noise and channel-based augmentation and into vocoder-based augmentation methods. In vocoder-based augmentation, a glottal vocoder is used to enhance the accuracy of ground truth F0 used for training of the neural network, as well as to expand the training data diversity in terms of F0 patterns and vocal tract lengths of the talkers. Evaluations on the PTDB-TUG corpus indicate that noise and channel augmentation can be used to greatly increase the noise robustness of trained models, and that vocoder-based ground truth enhancement further increases model performance. For smaller datasets, vocoder-based diversity augmentation can also be used to increase performance. The best-performing proposed method greatly outperformed the compared F0 estimation methods in terms of noise robustness.
  • Alcantara, Jose Carlos (Helsingin yliopisto, 2020)
    A recent machine learning technique called federated learning (Konecny, McMahan, et. al., 2016) offers a new paradigm for distributed learning. It consists of performing machine learning on multiple edge devices and simultaneously optimizing a global model for all of them, without transmitting user data. The goal for this thesis was to prove the benefits of applying federated learning to forecasting telecom key performance indicator (KPI) values from radio network cells. After performing experiments with different data sources' aggregations and comparing against a centralized learning model, the results revealed that a federated model can shorten the training time for modelling new radio cells. Moreover, the amount of transferred data to a central server is minimized drastically while keeping equivalent performance to a traditional centralized model. These experiments were performed with multi-layer perceptron as model architecture after comparing its performance against LSTM. Both, input and output data were sequences of KPI values.
  • Nabavi, Seyed Azad; Hossein Motlagh, Naser; Zaidan, Martha Arbayani; Aslani, Alireza; Zakeri, Behnam (2021)
    Buildings are responsible for 33% of final energy consumption, and 40% of direct and indirect CO2 emissions globally. While energy consumption is steadily rising globally, managing building energy utilization by on-site renewable energy generation can help responding to this demand. This paper proposes a deep learning method based on a discrete wavelet transformation and long short-term memory method (DWT-LSTM) and a scheduling framework for the integrated modelling and management of energy demand and supply for buildings. This method analyzes several factors including electricity price, uncertainty in climatic factors, availability of renewable energy sources (wind and solar), energy consumption patterns in buildings, and the non-linear relationships between these parameters on hourly, daily, weekly and monthly intervals. The method enables monitoring and controlling renewable energy generation, the share of energy imports from the grid, employment of saving strategy based on the user priority list, and energy storage management to minimize the reliance on the grid and electricity cost, especially during the peak hours. The results demonstrate that the proposed method can forecast building energy demand and energy supply with a high level of accuracy, showing a 3.63-8.57% error range in hourly data prediction for one month ahead. The combination of the deep learning forecasting, energy storage, and scheduling algorithm enables reducing annual energy import from the grid by 84%, which offers electricity cost savings by 87%. Finally, two smart active buildings configurations are financially analyzed for the next thirty years. Based on the results, the proposed smart building with solar Photo-Voltaic (PV), wind turbine, inverter, and 40.5 kWh energy storage has a financial breakeven point after 9 years with wind turbine and 8 years without it. This implies that implementing wind turbines in the proposed building is not financially beneficial.
  • Maljanen, Katri (Helsingin yliopisto, 2021)
    Cancer is a leading cause of death worldwide. Unlike its name would suggest, cancer is not a single disease. It is a group of diseases that arises from the expansion of a somatic cell clone. This expansion is thought to be a result of mutations that confer a selective advantage to the cell clone. These mutations that are advantageous to cells that result in their proliferation and escape of normal cell constraints are called driver mutations. The genes that contain driver mutations are known as driver genes. Studying these mutations and genes is important for understanding how cancer forms and evolves. Various methods have been developed that can discover these mutations and genes. This thesis focuses on a method called Deep Mutation Modelling, a deep learning based approach to predicting the probability of mutations. Deep Mutation Modelling’s output probabilities offer the possibility of creating sample and cancer type specific probability scores for mutations that reflect the pathogenicity of the mutations. Most methods in the past have made scores that are the same for all cancer types. Deep Mutation Modelling offers the opportunity to make a more personalised score. The main objectives of this thesis were to examine the Deep Mutation Modelling output as it was unknown what kind of features it has, see how the output compares against other scoring methods and how the probabilities work in mutation hotspots. Lastly, could the probabilities be used in a common driver gene discovery method. Overall, the goal was to see if Deep Mutation Modelling works and if it is competitive with other known methods. The findings indicate that Deep Mutation Modelling works in predicting driver mutations, but that it does not have sufficient power to do this reliably and requires further improvements.
  • Yli-Jyrä, Anssi Mikael (MTA Research Institute for Linguistics, 2019)
    Giving birth to Finite State Phonology is classically attributed to Johnson (1972), and Kaplan and Kay (1994). However, there is an ear- lier discovery that was very close to this achievement. In 1965, Hennie presented a very general sufficient condition for regularity of Turing machines. Although this discovery happened chronologically before Generative Phonology (Chomsky and Halle, 1968), it is a mystery why its relevance has not been realized until recently (Yli-Jyrä, 2017). The antique work of Hennie provides enough generality to advance even today’s frontier of finite-state phonology. First, it lets us construct a finite-state transducer from any grammar implemented by a tightly bounded one- tape Turing machine. If the machine runs in o(n log n), the construction is possible, and this case is reasonably decidable. Second, it can be used to model the regularity in context-sensitive derivations. For example, the suffixation in hunspell dictionaries (Németh et al., 2004) corresponds to time-bounded two-way computations performed by a Hennie machine. Thirdly, it challenges us to look for new forgotten islands of regularity where Hennie’s condition does not necessarily hold.
  • Räsänen, Janne; Salmivuori, Mari; Pölönen, Ilkka; Grönroos, Mari; Neittaanmäki, Noora (2021)
    Pigmented basal cell carcinomas can be difficult to distinguish from melanocytic tumours. Hyperspectral imaging is a non-invasive imaging technique that measures the reflectance spectra of skin in vivo. The aim of this prospective pilot study was to use a convolutional neural network classifier in hyperspectral images for differential diagnosis between pigmented basal cell carcinomas and melanoma. A total of 26 pigmented lesions (10 pigmented basal cell carcinomas, 12 melanomas in situ, 4 invasive melanomas) were imaged with hyperspectral imaging and excised for histopatho-logical diagnosis. For 2-class classifier (melano-cytic tumours vs pigmented basal cell carcinomas) using the majority of the pixels to predict the class of the whole lesion, the results showed a sensitivity of 100% (95% confidence interval 81-100%), specificity of 90% (95% confidence interval 60-98%) and positive predictive value of 94% (95% confidence interval 73-99%). These results indicate that a convolutional neural network classifier can differentiate melanocytic tumours from pigmented basal cell carcinomas in hyperspectral images. Further studies are warranted in order to confirm these preliminary results, using larger samples and multiple tumour types, including all types of melanocytic lesions.
  • Kurki, Lauri (Helsingin yliopisto, 2021)
    Atomic force microscopy (AFM) is a widely utilized characterization method capable of capturing atomic level detail in individual organic molecules. However, an AFM image contains relatively little information about the deeper atoms in a molecule and thus interpretation of AFM images of non-planar molecules offers significant challenges for human experts. An end-to-end solution starting from an AFM imaging system ending in an automated image interpreter would be a valuable asset for all research utilizing AFM. Machine learning has become a ubiquitous tool in all areas of science. Artificial neural networks (ANNs), a specific machine learning tool, have also arisen as a popular method many fields including medical imaging, self-driving cars and facial recognition systems. In recent years, progress towards interpreting AFM images from more complicated samples has been made utilizing ANNs. In this thesis, we aim to predict sample structures from AFM images by modeling the molecule as a graph and using a generative model to build the molecular structure atom-by-atom and bond-by-bond. The generative model uses two types of ANNs, a convolutional attention mechanism to process the AFM images and a graph neural network to process the generated molecule. The model is trained and tested using simulated AFM images. The results of the thesis show that the model has the capability to learn even slight details from complicated AFM images, especially when the model only adds a single atom to the molecule. However, there are challenges to overcome in the generative model for it to become a part of a fully capable end-to-end AFM process.
  • Barin Pacela, Vitória (Helsingin yliopisto, 2021)
    Independent Component Analysis (ICA) aims to separate the observed signals into their underlying independent components responsible for generating the observations. Most research in ICA has focused on continuous signals, while the methodology for binary and discrete signals is less developed. Yet, binary observations are equally present in various fields and applications, such as causal discovery, signal processing, and bioinformatics. In the last decade, Boolean OR and XOR mixtures have been shown to be identifiable by ICA, but such models suffer from limited expressivity, calling for new methods to solve the problem. In this thesis, "Independent Component Analysis for Binary Data", we estimate the mixing matrix of ICA from binary observations and an additionally observed auxiliary variable by employing a linear model inspired by the Identifiable Variational Autoencoder (iVAE), which exploits the non-stationarity of the data. The model is optimized with a gradient-based algorithm that uses second-order optimization with limited memory, resulting in a training time in the order of seconds for the particular study cases. We investigate which conditions can lead to the reconstruction of the mixing matrix, concluding that the method is able to identify the mixing matrix when the number of observed variables is greater than the number of sources. In such cases, the linear binary iVAE can reconstruct the mixing matrix up to order and scale indeterminacies, which are considered in the evaluation with the Mean Cosine Similarity Score. Furthermore, the model can reconstruct the mixing matrix even under a limited sample size. Therefore, this work demonstrates the potential for applications in real-world data and also offers a possibility to study and formalize identifiability in future work. In summary, the most important contributions of this thesis are the empirical study of the conditions that enable the mixing matrix reconstruction using the binary iVAE, and the empirical results on the performance and efficiency of the model. The latter was achieved through a new combination of existing methods, including modifications and simplifications of a linear binary iVAE model and the optimization of such a model under limited computational resources.
  • Savarese, Marco; Välipakka, Salla; Johari, Mridul; Hackman, Peter; Udd, Bjarne (2020)
    Human genes have a variable length. Those having a coding sequence of extraordinary length and a high number of exons were almost impossible to sequence using the traditional Sanger-based gene-by-gene approach. High-throughput sequencing has partly overcome the size-related technical issues, enabling a straightforward, rapid and relatively inexpensive analysis of large genes. Several large genes (e.g. TTN, NEB, RYR1, DMD) are recognized as disease-causing in patients with skeletal muscle diseases. However, because of their sheer size, the clinical interpretation of variants in these genes is probably the most challenging aspect of the high-throughput genetic investigation in the field of skeletal muscle diseases. The main aim of this review is to discuss the technical and interpretative issues related to the diagnostic investigation of large genes and to reflect upon the current state of the art and the future advancements in the field. © 2020 - IOS Press and the authors. All rights reserved.
  • Passos, Ives C.; Ballester, Pedro L.; Barros, Rodrigo C.; Librenza-Garcia, Diego; Mwangi, Benson; Birmaher, Boris; Brietzke, Elisa; Hajek, Tomas; Lopez Jaramillo, Carlos; Mansur, Rodrigo B.; Alda, Martin; Haarman, Bartholomeus C. M.; Isometsa, Erkki; Lam, Raymond W.; McIntyre, Roger S.; Minuzzi, Luciano; Kessing, Lars V.; Yatham, Lakshmi N.; Duffy, Anne; Kapczinski, Flavio (2019)
    Objectives The International Society for Bipolar Disorders Big Data Task Force assembled leading researchers in the field of bipolar disorder (BD), machine learning, and big data with extensive experience to evaluate the rationale of machine learning and big data analytics strategies for BD. Method A task force was convened to examine and integrate findings from the scientific literature related to machine learning and big data based studies to clarify terminology and to describe challenges and potential applications in the field of BD. We also systematically searched PubMed, Embase, and Web of Science for articles published up to January 2019 that used machine learning in BD. Results The results suggested that big data analytics has the potential to provide risk calculators to aid in treatment decisions and predict clinical prognosis, including suicidality, for individual patients. This approach can advance diagnosis by enabling discovery of more relevant data-driven phenotypes, as well as by predicting transition to the disorder in high-risk unaffected subjects. We also discuss the most frequent challenges that big data analytics applications can face, such as heterogeneity, lack of external validation and replication of some studies, cost and non-stationary distribution of the data, and lack of appropriate funding. Conclusion Machine learning-based studies, including atheoretical data-driven big data approaches, provide an opportunity to more accurately detect those who are at risk, parse-relevant phenotypes as well as inform treatment selection and prognosis. However, several methodological challenges need to be addressed in order to translate research findings to clinical settings.
  • Gierlach, Mateusz Tadeusz (Helsingin yliopisto, 2020)
    Visual fashion understanding (VFU) is a discipline which aims to solve tasks related to clothing recognition, such as garment categorization, garment’s attributes prediction or clothes retrieval, with the use of computer vision algorithms trained on fashion-related data. Having surveyed VFU- related scientific literature, I conclude that, because of the fact that at the heart of all VFU tasks is the same issue of visually understanding garments, those VFU tasks are in fact related. I present a hypothesis that building larger multi-task learning models dedicated to predicting multiple VFU tasks at once might lead to better generalization properties of VFU models. I assess the validity of my hypothesis by implementing two deep learning solutions dedicated primarily to category and attribute prediction. First solution uses multi-task learning concept of sharing features from ad- ditional branch dedicated to localization task of landmarks’ position prediction. Second solution does not share knowledge from localization branch. Comparison of those two implementations con- firmed my hypothesis, as sharing knowledge between tasks increased category prediction accuracy by 53% and attributes prediction recall by 149%. I conclude that multi-task learning improves generalization properties of deep learning-based visual fashion understanding models across tasks.
  • Kutvonen, Konsta (Helsingin yliopisto, 2020)
    With modern computer vision algorithms, it is possible to solve many different kinds of problems, such as object detection, image classification, and image segmentation. In some cases, like in the case of a camera-based self-driving car, the task can't yet be adequately solved as a direct mapping from image to action with a single model. In such situations, we need more complex systems that can solve multiple computer vision tasks to understand the environment and act based on it for acceptable results. Training each task on their own can be expensive in terms of storage required for all weights and especially for the inference time as the output of several large models is needed. Fortunately, many state-of-the-art solutions to these problems use Convolutional Neural Networks and often feature some ImageNet backbone in their architecture. With multi-task learning, we can combine some of the tasks into a single model, sharing the convolutional weights in the network. Sharing the weights allows for training smaller models that produce outputs faster and require less computational resources, which is essential, especially when the models are run on embedded devices with constrained computation capability and no ability to rely on the cloud. In this thesis, we will present some state-of-the-art models to solve image classification and object detection problems. We will define multi-task learning, how we can train multi-task models, and take a look at various multi-task models and how they exhibit the benefits of multi-task learning. Finally, to evaluate how training multi-task models changes the basic training paradigm and to find what issues arise, we will train multiple multi-task models. The models will mainly focus on image classification and object detection using various data sets. They will combine multiple tasks into a single model, and we will observe the impact of training the tasks in a multi-task setting.
  • Rasse, Tobias M.; Hollandi, Reka; Horvath, Peter (2020)
    Various pre-trained deep learning models for the segmentation of bioimages have been made available as developer-to-end-user solutions. They are optimized for ease of use and usually require neither knowledge of machine learning nor coding skills. However, individually testing these tools is tedious and success is uncertain. Here, we present the Open Segmentation Framework (OpSeF), a Python framework for deep learning-based instance segmentation. OpSeF aims at facilitating the collaboration of biomedical users with experienced image analysts. It builds on the analysts' knowledge in Python, machine learning, and workflow design to solve complex analysis tasks at any scale in a reproducible, well-documented way. OpSeF defines standard inputs and outputs, thereby facilitating modular workflow design and interoperability with other software. Users play an important role in problem definition, quality control, and manual refinement of results. OpSeF semi-automates preprocessing, convolutional neural network (CNN)-based segmentation in 2D or 3D, and postprocessing. It facilitates benchmarking of multiple models in parallel. OpSeF streamlines the optimization of parameters for pre- and postprocessing such, that an available model may frequently be used without retraining. Even if sufficiently good results are not achievable with this approach, intermediate results can inform the analysts in the selection of the most promising CNN-architecture in which the biomedical user might invest the effort of manually labeling training data. We provide Jupyter notebooks that document sample workflows based on various image collections. Analysts may find these notebooks useful to illustrate common segmentation challenges, as they prepare the advanced user for gradually taking over some of their tasks and completing their projects independently. The notebooks may also be used to explore the analysis options available within OpSeF in an interactive way and to document and share final workflows. Currently, three mechanistically distinct CNN-based segmentation methods, the U-Net implementation used in Cellprofiler 3.0, StarDist, and Cellpose have been integrated within OpSeF. The addition of new networks requires little; the addition of new models requires no coding skills. Thus, OpSeF might soon become both an interactive model repository, in which pre-trained models might be shared, evaluated, and reused with ease.
  • Hiippala, Tuomo (IEEE, 2017)
    This paper presents a system that uses machine learning to recognize military vehicles in social media images. To do so, the system draws on recent advances in applying deep neural networks to computer vision tasks, while also making extensive use of openly available libraries, models and data. Training a vehicle recognition system over three classes, the paper reports on two experiments that use different architectures and strategies to overcome the challenges of working with limited training data: data augmentation and transfer learning. The results show that transfer learning outperforms data augmentation, achieving an average accuracy of 95.18% using 10-fold cross-validation, while also generalizing well on a separate testing set consisting of social media content.
  • Enwald, Joel (Helsingin yliopisto, 2020)
    Mammography is used as an early detection system for breast cancer, which is one of the most common types of cancer, regardless of one’s sex. Mammography uses specialised X-ray machines to look into the breast tissue for possible tumours. Due to the machine’s set-up as well as to reduce the radiation patients are exposed to, the number of X-ray measurements collected is very restricted. Reconstructing the tissue from this limited information is referred to as limited angle tomography. This is a complex mathematical problem and ordinarily leads to poor reconstruction results. The aim of this work is to investigate how well a neural network whose structure utilizes pre-existing models and known geometry of the problem performs at this task. In this preliminary work, we demonstrate the results on simulated two-dimensional phantoms and discuss the extension of the results to 3-dimensional patient data.
  • Serra, Angela; Fratello, Michele; Cattelani, Luca; Liampa, Irene; Melagraki, Georgia; Kohonen, Pekka; Nymark, Penny; Federico, Antonio; Kinaret, Pia Anneli Sofia; Jagiello, Karolina; Ha, My Kieu; Choi, Jang-Sik; Sanabria, Natasha; Gulumian, Mary; Puzyn, Tomasz; Yoon, Tae-Hyun; Sarimveis, Haralambos; Grafström, Roland; Afantitis, Antreas; Greco, Dario (2020)
    Transcriptomics data are relevant to address a number of challenges in Toxicogenomics (TGx). After careful planning of exposure conditions and data preprocessing, the TGx data can be used in predictive toxicology, where more advanced modelling techniques are applied. The large volume of molecular profiles produced by omics-based technologies allows the development and application of artificial intelligence (AI) methods in TGx. Indeed, the publicly available omics datasets are constantly increasing together with a plethora of different methods that are made available to facilitate their analysis, interpretation and the generation of accurate and stable predictive models. In this review, we present the state-of-the-art of data modelling applied to transcriptomics data in TGx. We show how the benchmark dose (BMD) analysis can be applied to TGx data. We review read across and adverse outcome pathways (AOP) modelling methodologies. We discuss how network-based approaches can be successfully employed to clarify the mechanism of action (MOA) or specific biomarkers of exposure. We also describe the main AI methodologies applied to TGx data to create predictive classification and regression models and we address current challenges. Finally, we present a short description of deep learning (DL) and data integration methodologies applied in these contexts. Modelling of TGx data represents a valuable tool for more accurate chemical safety assessment. This review is the third part of a three-article series on Transcriptomics in Toxicogenomics.
  • Nezami, Somayeh; Khoramshahi, Ehsan; Nevalainen, Olli; Pölönen, Ilkka; Honkavaara, Eija (2020)
    Interest in drone solutions in forestry applications is growing. Using drones, datasets can be captured flexibly and at high spatial and temporal resolutions when needed. In forestry applications, fundamental tasks include the detection of individual trees, tree species classification, biomass estimation, etc. Deep neural networks (DNN) have shown superior results when comparing with conventional machine learning methods such as multi-layer perceptron (MLP) in cases of huge input data. The objective of this research is to investigate 3D convolutional neural networks (3D-CNN) to classify three major tree species in a boreal forest: pine, spruce, and birch. The proposed 3D-CNN models were employed to classify tree species in a test site in Finland. The classifiers were trained with a dataset of 3039 manually labelled trees. Then the accuracies were assessed by employing independent datasets of 803 records. To find the most efficient set of feature combination, we compare the performances of 3D-CNN models trained with hyperspectral (HS) channels, Red-Green-Blue (RGB) channels, and canopy height model (CHM), separately and combined. It is demonstrated that the proposed 3D-CNN model with RGB and HS layers produces the highest classification accuracy. The producer accuracy of the best 3D-CNN classifier on the test dataset were 99.6%, 94.8%, and 97.4% for pines, spruces, and birches, respectively. The best 3D-CNN classifier produced ~5% better classification accuracy than the MLP with all layers. Our results suggest that the proposed method provides excellent classification results with acceptable performance metrics for HS datasets. Our results show that pine class was detectable in most layers. Spruce was most detectable in RGB data, while birch was most detectable in the HS layers. Furthermore, the RGB datasets provide acceptable results for many low-accuracy applications.