<p>Deep Neural Networks (DNNs) have greatly advanced the state-of-the-art in a wide range of machine learning tasks involving image, video, speech and text analytics, and are deployed in numerous widely-used products and services. Improvements in the capabilities of hardware platforms such as Graphics Processing Units (GPUs) and specialized accelerators have been instrumental in enabling these advances as they have allowed more complex and accurate networks to be trained and deployed. However, the enormous computational and memory demands of DNNs continue to increase with growing data size and network complexity, posing a continuing challenge to computing system designers. For instance, state-of-the-art image recognition DNNs require hundreds of millions of parameters and hundreds of billions of multiply-accumulate operations while state-of-the-art language models require hundreds of billions of parameters and several trillion operations to process a single input instance. Another major obstacle in the adoption of DNNs, despite their impressive accuracies on a range of datasets, has been their lack of robustness. Specifically, recent efforts have demonstrated that small, carefully-introduced input perturbations can force a DNN to behave in unexpected and erroneous ways, which can have to severe consequences in several safety-critical DNN applications like healthcare and autonomous vehicles. In this dissertation, we explore approximate computing as an avenue to improve the speed and energy efficiency of DNNs, as well as their robustness to input perturbations.</p> <p> </p> <p>Approximate computing involves executing selected computations of an application in an approximate manner, while generating favorable trade-offs between computational efficiency and output quality. The intrinsic error resilience of machine learning applications makes them excellent candidates for approximate computing, allowing us to achieve execution time and energy reductions with minimal effect on the quality of outputs. This dissertation performs a comprehensive analysis of different approximate computing techniques for improving the execution efficiency of DNNs. Complementary to generic approximation techniques like quantization, it identifies approximation opportunities based on the specific characteristics of three popular classes of networks - Feed-forward Neural Networks (FFNNs), Recurrent Neural Networks (RNNs) and Spiking Neural Networks (SNNs), which vary considerably in their network structure and computational patterns.</p> <p> </p> <p>First, in the context of feed-forward neural networks, we identify sparsity, or the presence of zero values in the data structures (activations, weights, gradients and errors), to be a major source of redundancy and therefore, an easy target for approximations. We develop lightweight micro-architectural and instruction set extensions to a general-purpose processor core that enable it to dynamically detect zero values when they are loaded and skip future instructions that are rendered redundant by them. Next, we explore LSTMs (the most widely used class of RNNs), which map sequences from an input space to an output space. We propose hardware-agnostic approximations that dynamically skip redundant symbols in the input sequence and discard redundant elements in the state vector to achieve execution time benefits. Following that, we consider SNNs, which are an emerging class of neural networks that represent and process information in the form of sequences of binary spikes. Observing that spike-triggered updates along synaptic connections are the dominant operation in SNNs, we propose hardware and software techniques to identify connections that can be minimally impact the output quality and deactivate them dynamically, skipping any associated updates.</p> <p> </p> <p>The dissertation also delves into the efficacy of combining multiple approximate computing techniques to improve the execution efficiency of DNNs. In particular, we focus on the combination of quantization, which reduces the precision of DNN data-structures, and pruning, which introduces sparsity in them. We observe that the ability of pruning to reduce the memory demands of quantized DNNs decreases with precision as the overhead of storing non-zero locations alongside the values starts to dominate in different sparse encoding schemes. We analyze this overhead and the overall compression of three different sparse formats across a range of sparsity and precision values and propose a hybrid compression scheme that identifies that optimal sparse format for a pruned low-precision DNN.</p> <p> </p> <p>Along with improved execution efficiency of DNNs, the dissertation explores an additional advantage of approximate computing in the form of improved robustness. We propose ensembles of quantized DNN models with different numerical precisions as a new approach to increase robustness against adversarial attacks. It is based on the observation that quantized neural networks often demonstrate much higher robustness to adversarial attacks than full precision networks, but at the cost of a substantial loss in accuracy on the original (unperturbed) inputs. We overcome this limitation to achieve the best of both worlds, i.e., the higher unperturbed accuracies of the full precision models combined with the higher robustness of the low precision models, by composing them in an ensemble.</p> <p> </p> <p><br></p><p>In summary, this dissertation establishes approximate computing as a promising direction to improve the performance, energy efficiency and robustness of neural networks.</p>

This paper deals with classifying objects using deep neural networks. Whole scene segmentation was used as main algorithm for the classification purpose which works with video sequences and obtains information between two video frames. Optical flow was used for getting information from the video frames, based on which features maps of a~neural network are warped. Two neural network architectures were adjusted to work with videos and experimented with. Results of the experiments show, that using videos for image segmentation improves accuracy (IoU) compared to the same architecture working with images.

The goal of this work is to bring automatic defect detection to the manufacturing process of plastic cards. A card is considered defective when it is contaminated with a dust particle or a hair. The main challenges I am facing to accomplish this task are a very few training data samples (214 images), small area of target defects in context of an entire card (average defect area is 0.0068 \% of the card) and also very complex background the detection task is performed on. In order to accomplish the task, I decided to use Mask R-CNN detection algorithm combined with augmentation techniques such as synthetic dataset generation. I trained the model on the synthetic dataset consisting of 20 000 images. This way I was able to create a model performing 0.83 AP at 0.1 IoU on the original data test set.

The aim of this work is to increase the accuracy of the transcription of text documents. This work is mainly focused on texts printed on degraded materials such as newspapers or old books. To solve this problem, the current method and problems associated with text recognition are analyzed. Based on the acquired knowledge, the implemented method based on GAN network architecture is chosen. Experiments are a performer on these networks in order to find their appropriate size and their learning parameters. Subsequently, testing is performed to compare different learning methods and compare their results. Both training and testing is a performer on an artificial data set. Using implemented trained networks increases the transcription accuracy from 65.61 % for the raw damaged text lines to 93.23 % for lines processed by this network.

This work deals with automatic chord recognition using neural networks. The problem was separated into two subproblems. The first subproblem aims to experimental finding of most suitable solution for a acoustic model and the second one aims to experimental finding of most suitable solution for a language model. The problem was solved by iterative method. First a suboptimal solution of the first subproblem was found and then the second one. A total of 19 acoustic and 12 language models were made. Ten training datasets was created for acoustic models and three for language models. In total, over 200 models were trained. The best results were achieved on acoustic models represented by convolutional networks together with language models represented by recurent networks with LSTM modules.

První část této diplomové práci se zabývá teoretickým rozborem principů neuronových sítí, včetně možnosti jejich použití v oblasti rozpoznávání řeči. Práce pokračuje popisem viceúkolových neuronových sítí a souvisejících experimentů. Praktická část práce obsahovala změny software pro trénování neuronových sítí, které umožnily viceúkolové trénování. Je rovněž popsáno připravené prostředí, včetně několika dedikovaných skriptů. Experimenty představené v této diplomové práci ověřují použití artikulačních characteristik řeči pro viceúkolové trénování. Experimenty byly provedeny na dvou řečových databázích lišících se kvalitou a velikostí a representujících různé jazyky - angličtinu a vietnamštinu. Artikulační charakteristiky byly také kombinovány s jinými sekundárními úkoly, například kontextem, s záměrem ověřit jejich komplementaritu. Porovnaní je provedeno s neuronovými sítěmi různých velikostí tak, aby byl popsán vztah mezi velikostí neuronových sítí a efektivitou viceúkolového trénování. Závěrem provedených experimentů je, že viceúkolové trénování s použitím artikulačnich charakteristik jako sekundárních úkolů vede k lepšímu trénování neuronových sítí a výsledkem tohoto trénování může být přesnější rozpoznávání fonémů. V závěru práce jsou viceúkolové neuronové sítě testovány v systému rozpoznávání řeči jako extraktor příznaků.

While browsing an online map server, continuous photographs of certain places can be browsed as well. When the map service takes pictures of a public space, there are some personal data captured as well (i.e. faces, car licence plates). The goal of this thesis is the design of automated car licence plates anonymization system, optimized for the Panorama service provided by the Seznam.cz a.s. corporation. In this thesis, the process of car licence plate anonymization is divided into two parts: the first one solves a detection of cars and the second solves a car licence plate localization in the selected image. The car detection is based on the deep neural network approach, the car licence plate localization is solved by using a fully connected neural network performing a regression task. The goal of this thesis is to get over the disadvantages of commercial solution used nowadays. These are false posititive results and high computational complexity. Results of this thesis are not as good as expected. The reason could be a dataset provided by Seznam.cz a.s. corporation, which seemed to be robust enough in the beginning, but in the end it showed up to be not suffice enough to train the neural network.

<p>Automatsko prepoznavanje govora je tehnologija koja računarima<br />omogućava pretvaranje izgovorenih reči u tekst. Ona se može<br />primeniti u mnogim savremenim sistemima koji uključuju komunikaciju<br />između čoveka i mašine. U ovoj disertaciji detaljno je opisana jedna<br />od dve glavne komponente sistema za prepoznavanje govora, a to je<br />jezički model, koji specificira rečnik sistema, kao i pravila prema<br />kojim se pojedinačne reči mogu povezati u rečenicu. Srpski jezik spada<br />u grupu visoko inflektivnih i morfološki bogatih jezika, što znači<br />da koristi veći broj različitih završetaka reči za izražavanje<br />željene gramatičke, sintaksičke ili semantičke funkcije date reči.<br />Ovakvo ponašanje često dovodi do velikog broja grešaka sistema za<br />prepoznavanje govora kod kojih zbog dobrog akustičkog poklapanja<br />prepoznavač pogodi osnovni oblik reči, ali pogreši njen završetak.<br />Taj završetak može da označava drugu morfološku kategoriju, na<br />primer, padež, rod ili broj. U radu je predstavljen novi alat za<br />modelovanje jezika, koji uz identitet reči u modelu može da koristi<br />dodatna leksička i morfološka obeležja reči, čime je testirana<br />hipoteza da te dodatne informacije mogu pomoći u prevazilaženju<br />značajnog broja grešaka prepoznavača koje su posledica<br />inflektivnosti srpskog jezika.</p> / <p>Automatic speech recognition is a technology that allows computers to<br />convert spoken words into text. It can be applied in various areas which<br />involve communication between humans and machines. This thesis primarily<br />deals with one of two main components of speech recognition systems - the<br />language model, that specifies the vocabulary of the system, as well as the<br />rules by which individual words can be linked into sentences. The Serbian<br />language belongs to a group of highly inflective and morphologically rich<br />languages, which means that it uses a number of different word endings to<br />express the desired grammatical, syntactic, or semantic function of the given<br />word. Such behavior often leads to a significant number of errors in speech<br />recognition systems where due to good acoustic matching the recognizer<br />correctly guesses the basic form of the word, but an error occurs in the word<br />ending. This word ending may indicate a different morphological category, for<br />example, word case, grammatical gender, or grammatical number. The<br />thesis presents a new language modeling tool which, along with the word<br />identity, can also model additional lexical and morphological features of the<br />word, thus testing the hypothesis that this additional information can help<br />overcome a significant number of recognition errors that result from the high<br />inflectivity of the Serbian language.</p>

abstract: Object detection is an interesting computer vision area that is concerned with the detection of object instances belonging to specific classes of interest as well as the localization of these instances in images and/or videos. Object detection serves as a vital module in many computer vision based applications. This work focuses on the development of object detection methods that exhibit increased robustness to varying illuminations and image quality. In this work, two methods for robust object detection are presented. In the context of varying illumination, this work focuses on robust generic obstacle detection and collision warning in Advanced Driver Assistance Systems (ADAS) under varying illumination conditions. The highlight of the first method is the ability to detect all obstacles without prior knowledge and detect partially occluded obstacles including the obstacles that have not completely appeared in the frame (truncated obstacles). It is first shown that the angular distortion in the Inverse Perspective Mapping (IPM) domain belonging to obstacle edges varies as a function of their corresponding 2D location in the camera plane. This information is used to generate object proposals. A novel proposal assessment method based on fusing statistical properties from both the IPM image and the camera image to perform robust outlier elimination and false positive reduction is also proposed. In the context of image quality, this work focuses on robust multiple-class object detection using deep neural networks for images with varying quality. The use of Generative Adversarial Networks (GANs) is proposed in a novel generative framework to generate features that provide robustness for object detection on reduced quality images. The proposed GAN-based Detection of Objects (GAN-DO) framework is not restricted to any particular architecture and can be generalized to several deep neural network (DNN) based architectures. The resulting deep neural network maintains the exact architecture as the selected baseline model without adding to the model parameter complexity or inference speed. Performance results provided using GAN-DO on object detection datasets establish an improved robustness to varying image quality and a higher object detection and classification accuracy compared to the existing approaches. / Dissertation/Thesis / Doctoral Dissertation Electrical Engineering 2020

Quality inspections have seen varying degrees of automation depending on the complexity of the task and the environment. Especially in later phases of multi-stage manufacturing processes, such as final assembly in automotive industries, quality inspections are largely manual to this day. Today, emerging technologies offer both pressures and tools to increase automation. However, the current state of the research field is lacking in studies that help guide companies toward implementation. Thus, quality managers at final assembly for Scania's truck coachwork factory in Oskarshamn (MC) stipulated a thesis assignment to explore how inspections in their final assembly workshop could be automated. This assignment constitutes the purpose of this thesis project - to provide an exploratory study into existing and emerging technologies that enable automation of quality inspections at MC. This was eventually delimited to exploring automated visual inspection technologies. In order to better understand Scania's inspection and manufacturing system, a series of interviews and shadowings were undertaken with appropriate respondents. From these, we were able to extract seven inspection system requirements, most important were the ability to (1) handle high variability, (2) add new inspections fast, (3) inspect in direct flow and (4) inspect inside and outside of the truck coach without disassembly. Then, a thorough and comprehensive review of 559 active inspections allowed us to categorize and map the nature of inspections at MC. In our literature review, a model for a general quality inspections was found, which was used to guide and ground our proposals and recommendations as well as provided intuitive illustration. Further, two paradigms emerged as most interesting for this project: machine vision and deep learning. A theoretical comparison of the two suggested that the more traditional, rule-based machine vision algorithms would struggle in accommodating the requirements previously found. However, we could infer that deep learning would be highly suitable with respect to MC's requirements and inspections. A prototype deep learning inspection system gave further validation toward our speculations that deep learning offered the greatest potential for automation in complex environments such as MC's. Although this thesis was created for Scania as a primary customer, important theoretical and practical contributions were developed for a more general audience. Firstly, the exploration into new avenues for automation that overcome their traditional limitations were provided; something that is of high current import given the trends toward more complex manufacturing settings. Practically, we provide some guidance to industries that find themselves in similar situations to Scania - employing complex manufacturing systems or having complex products - where our findings can give insights in regards to modern automation challenges and solutions. / Kvalitetsinspektioner har automatiserats i variarande grad beroende på uppgiftens och omgivningens komlexitet. I synnerhet i de senare stadierna av flerstegsproduktioner, exempelvis slutmontering i fordonstillverkningsindustrin, består manuella inspektioner i stor utsträckning. Den snabba tekniska utvecklingen som har skett nyligen avger både ett tryck och skapar verktyg för att utöka automatiseringen. Dessvärre erbjuder dagslägets forskning föga stöd till företag gällande storskalig implementering av automatiserade kvalitetsinspektionssystem. Därför skapade kvalitetschefer på Scanias lastbilshyttmonteringsfabrik i Oskarshamn (MC) ett uppdrag att utforska hur deras inspektioner skulle kunna automatiseras. Detta uppdrag utgjorde syftet i vårt examensarbete: att utföra en explorativ studie inom befintliga och nya tekniker som möjliggör automatisering av MCs kvalitetsinpspektioner, vilket senare avgränsades till undersökandet av visuella kvalitetsinspektioner. För att tillgodogöra oss en djupare förståelse för Scanias inspektions- och produktionssytem utfördes en serie intervjuer och skuggningar med kunniga respondenter. Datan som erhölls utgjorde grunden i en nulägesanalys, från vilken sju systemkrav för ett inspektionssystem på MC kunde extraheras. De viktigaste av dessa var förmågan att (1) klara av hög variation, (2) addera nya inspektionspunkter snabbt, (3) kontrollera i direktflödet och (4) kontrollera innan- och utanför lastbilshytten. Vidare gjordes en omfattande genomgång av 559 aktiva inspektionspunkter vilket resulterade i en kategorisering och kartläggning av inspektioner på MC. I vår genomgång av relevant vetenskaplig litteratur hittades en generell modell för kvalitetskontroll som användes för att illustrera och teroretiskt förankra rekommendationer för ett automatiskt inspektionssystem. Vidare urskiljdes två intressanta områden i forskningen, machine vision och deep learning. En teoretisk jämförelse av traditionella regelbaserade machine vision algoritmer med deep learning erhöll att den förstnämnda är mindre lämpad för Scania med hänsyn till de krav som tagits fram. Deep learning å andra sidan, erbjuder många fördelar i relation till dessa. Genom en relativt enkel process kunde en deep learning baserad prototyp utvecklas. Prototypen påvisade goda resultat och gav vidare validering av vår spekulation att deep learning är ett lämpligt verktyg för automatisering i komplexa miljöer.Trots att detta examensarbete hade Scania som huvudsaklig uppdragsgivare så gjordes viktiga teoretiska och praktiska bidrag. En utforskning av i nya möjligheter för automatisering som kan överkomma begränsningarna av traditionell automatisering framtogs, vilket anses som både aktuellt och av vikt för samtiden där trender går mot mer dynamiska produktionssystem. Vad gäller praktiska bidrag så utgör denna rapport en sammanställning av råd till företag som befinner sig i liknande sitser som Scania - som använder komplexa produktionssystem eller har komplexa produkter - där våra resultat kan ge insikt gällande svårigheter och lösningar för modern automatisering.