Modern neural network (NN) based control schemes have surmounted many of the limitations found in the traditional control approaches. Nevertheless, these modern control techniques have only recently been introduced for use on high-specification Programmable Logic Controllers (PLCs) and usually at a very high cost in terms of the required software and hardware. This ‗intelligent‘ control in the sector of industrial automation, specifically on standard PLCs thus remains an area of study that is open to further research and development. The research documented in this thesis examined the effectiveness of linear traditional control schemes such as Proportional Integral Derivative (PID), Lead and Lead-Lag control, in comparison to non-linear NN based control schemes when applied on a strongly non-linear platform. To this end, a mechatronic-type balancing system, namely, the Ball-on-Wheel (BOW) system was designed, constructed and modelled. Thereafter various traditional and intelligent controllers were implemented in order to control the system. The BOW platform may be taken to represent any single-input, single-output (SISO) non-linear system in use in the real world. The system makes use of current industrial technology including a standard PLC as the digital computational platform, a servo drive and wireless access for remote control. The results gathered from the research revealed that NN based control schemes (i.e. Pure NN and NN-PID), although comparatively slower in response, have greater advantages over traditional controllers in that they are able to adapt to external system changes as well as system non-linearity through a process of learning. These controllers also reduce the guess work that is usually involved with the traditional control approaches where cumbersome modelling, linearization or manual tuning is required. Furthermore, the research showed that online-learning adaptive traditional controllers such as the NN-PID controller which maintains the best of both the intelligent and traditional controllers may be implemented easily and with minimum expense on standard PLCs.

An investigation has been made into symmetry features of patterns as a means by which the patterns are described, or with which they are transformed prior to classification in order to assist a pattern recognition system. There are two main points of departure from existing symmetry use in the pattern recognition domain. The first is the adoption of the theory that patterns can be categorised solely using a map of the symmetry features that exist within the static pattern. The second is the application of symmetry transforms to aid non-trivial recognition in patterns which are not intended to be perfectly symmetrical. An experiment is conducted to classify the reflectional symmetry features of digits, using the Generalised Symmetry Transform to produce the features and Probabilistic Neural Networks to perform the classification. Symmetry feature information is also used to define parameters of affine transformations to achieve improved performance in tolerance to variances in position and orientation. The results lead to an investigation of the role of asymmetry. The Generalised Symmetry Transform is modified to produce two related transforms: the Generalised Asymmetry Transform and the Generalised Asymmetry and Symmetry Transform. Finally, a new symmetry transform is proposed which separates the factors affecting the degree of symmetry in an image into three non-linear functions of corresponding pairs of pixels. These factors are: the colour intensity values; the pixel orientations; and the respective distance between the point and potential reflection plane. The strictness of symmetry, or tolerance to non-symmetrical artifacts, is defined in variable parameters which are set to suit the desired application. This new transform is called the Reflectional Symmetry Transform. The structure of its input and output match those of the Generalised Symmetry Transform, which it is intended to replace.

Anesthetics have been used for nearly two centuries, and have proved to be one of the most important tools in surgical interventions, but their methods of action remain mysterious. Previous research has focused on high-level, low-resolution measurements (average activity of many neurons) or low-level, high-resolution measurements (single neurons). The nematode Caenorhabditis elegans provides an excellent model to bridge the gap between these two scales by measuring the activity of many neurons with single neuron resolution. C. elegans display analogous behaviors to humans under anesthesia. Employing confocal imaging of GCaMP, I measured neuronal activity at different isoflurane levels in C. elegans ganglia and in small behavior-controlling circuits. The activity in C. elegans ganglia is similar to that of human ganglia, as assessed using measures that are similar to EEG. Activity in the small behavior-controlling circuit is disrupted, but not suppressed, when dosed with moderate levels of isoflurane. Neural activity in the circuit is randomized resulting in a loss of coordination between neurons that define behavioral states of the system. As such, the onset of the behaviors of anesthesia appears to be the resultant of randomization rather than suppression of individual neuron activity. Employing light sheet microscopy and automated image analysis for neuronal tracking, I expanded the imaging techniques to measure activity of the majority of neurons in the animal’s head. Expansion of these measurements to the whole head region of the nematode confirms these findings, displaying significant decreases in neuron-to-neuron coordination, as well as randomization of individual neuron signals with the onset of anesthesia. These results reveal a new physiological mechanism of action for anesthetics, and provide an avenue forward for investigating the molecular mechanism including specific genetic mutations known to alter susceptibility to anesthetics. / 2021-05-26T00:00:00Z

Despite the impressive performance of deep neural networks (DNNs) on numerous vision tasks, they still exhibit yet-to-understand uncouth behaviours. One puzzling behaviour is the reaction of DNNs to various noise attacks, where it has been shown that there exist small adversarial noise that can result in a severe degradation in the performance of DNNs. To rigorously treat this, we derive exact analytic expressions for the first and second moments (mean and variance) of a small piecewise linear (PL) network with a single rectified linear unit (ReLU) layer subject to general Gaussian input. We experimentally show that these expressions are tight under simple linearizations of deeper PL-DNNs, especially popular architectures in the literature (e.g. LeNet and AlexNet). Extensive experiments on image classification show that these expressions can be used to study the behaviour of the output mean of the logits for each class, the inter-class confusion and the pixel-level spatial noise sensitivity of the network. Moreover, we show how these expressions can be used to systematically construct targeted and non-targeted adversarial attacks. Then, we proposed a special estimator DNN, named mixture of linearizations (MoL), and derived the analytic expressions for its output mean and variance, as well. We employed these expressions to train the model to be particularly robust against Gaussian attacks without the need for data augmentation. Upon training this network on a loss that is consolidated with the derived output probabilistic moments, the network is not only robust under very high variance Gaussian attacks but is also as robust as networks that are trained with 20 fold data augmentation.

In contrast to large-scale earthquakes which are caused when energy is released as a result of rock failure along a fault, microseismic events are caused when human activities, such as mining or oil and gas production, change the stress distribution or the volume of a rockmass. During such processes, microseismic event location, which aims at estimating source locations accurately, is a vital component of observing, diagnosing and acting upon the dynamic indications in reservoir performance by tracking the fracturing properly. Conventional methods for microseismic event location face considerable drawbacks. For example, traveltime based methods require manual labor in traveltime picking and thus suffer from the heavy workload of human interactions and manmade errors. Migration based and waveform inversion based location methods demand large computational memory and time for simulating the wavefields, especially in face of tens of thousands of microseismic events recorded. In this thesis research, we developed an approach based on a deep CNN for the purpose of microseismic event location, which is completely automatic with no human interactions like traveltime picking and also computationally friendly due to no requirement of wavefield simulations. An example in which the network is well-trained on the synthetic data from the smooth SEAM model and tested on the true SEAM model has shown its accuracy and efficiency. Moreover, we have proved that this approach is not only feasible for the cases with a uniform receiver distribution, but also applicable to cases where the passive seismic data are acquired with an irregular spacing geometry of sensors, which makes this approach more practical in reality.

This thesis aims to contribute to the field of neural networks by improving upon the performance of a state-of-the-art regularization scheme called MixUp, and by contributing to the conceptual understanding of MixUp. MixUp is a data augmentation scheme in which pairs of training samples and their corresponding labels are mixed using linear coefficients. Without label mixing, MixUp becomes a more conventional scheme: input samples are moved but their original labels are retained. Because samples are preferentially moved in the direction of other classes we refer to this method as directional adversarial training, or DAT. We show that under two mild conditions, MixUp asymptotically convergences to a subset of DAT. We define untied MixUp (UMixUp), a superset of MixUp wherein training labels are mixed with different linear coefficients to those of their corresponding samples. We show that under the same mild conditions, untied MixUp converges to the entire class of DAT schemes. Motivated by the understanding that UMixUp is both a generalization of MixUp and a scheme possessing adversarial-training properties, we experiment with different datasets and loss functions to show that UMixUp provides improves performance over MixUp. In short, we present a novel interpretation of MixUp as belonging to a class highly analogous to adversarial training, and on this basis we introduce a simple generalization which outperforms MixUp.

The main purpose of this paper is to demonstrate the data reduction technique of self-organizing maps and to compare it with data reduction techniques in SPSS. Especially, factor analysis and multidimensional scaling (MDS) are chosen. Subsequent to data reduction a cluster analysis was conducted. Due to taking the same cluster algorithm on the base of different data reduction approaches we can compare the final outputs of the cluster algorithm in respect to a target criterion. This is the homogeneity within the groups compared to the homogeneity between the groups. The application example is taken from literature (Backhaus et al. 1994).

Orientador: Rouverson Pereira da Silva / Coorientador: Walter Maldonado Júnior / Glauco de Souza Rolim / Daniel de Bortoli Teixeira / Luis Carlos Cirilo Carvalho / Gabriel Araújo e Silva Ferraz / Resumo: O café é uma commodity cujo preço é ajustado conforme parâmetros de qualidade, sendo a colheita uma operação que está intimamente ligada à qualidade final do produto. Sabemos que frutos no estádio cereja apresentam melhor qualidade de bebida em comparação aos frutos verdes ou secos. Além disso, a planta de café em suas condições naturais, apresenta maturação em todos os estádios na mesma planta. Sendo assim, o café seria uma cultura com potencial que justifique mais pesquisas na colheita. Surge então a hipótese: e se fosse possível colher somente frutos maduros - cereja e seco? Permanecendo os frutos verdes ainda na planta, para completar seu ciclo de maturação fisiológica, sendo colhidos em um segundo momento? E se com um celular em mãos o produtor pudesse tirar fotografias dos grãos, e um estimador digital reconhecesse os frutos e estimasse a produtividade e classificasse os frutos em seus respectivos estádios de maturação. Seria possível ter uma quantidade de informação suficiente de pontos em uma área, para gerar um mapa de produtividade e maturação? De posse do mapa, poderíamos gerar um "projeto de colheita" a ser realizado pela colhedora, buscando colher apenas os frutos de interesse. Para comprovar tal situação dividimos o trabalho em duas etapas: 1) Por meio de imagens identificar os frutos de café digitalmente. Esperamos que por meio de imagens tiradas por celulares ou câmeras possam estimar a produtividade e maturação, utilizando as técnicas de processamento de imag... (Resumo completo, clicar acesso eletrônico abaixo) / Abstract: ABSTRACT: Coffee is a commodity which price is adjusted according to quality parameters, the harvest is an operation that is closely linked to the final quality of the product. We know that fruits in the cherry stage have better quality drink compared to green or dry fruits. In addition, the coffee plant in its natural conditions, shows maturation in all stages in the same plant. Therefore, coffee would be a crop with potential to justify more research at harvesting. The hypothesis then arises: what if it were possible to harvest only ripe fruits - cherry and dry? Remaining the green fruits still in the plant, to complete their physiological maturation cycle, being harvested in a second moment? And if with a cell phone in hand farmers could take pictures of the grains, and a digital estimator would recognize the fruits and estimate the productivity and classify the fruits in their respective stages of maturation. It would be possible to have enough information in the area to generate maps of productivity and maturation. In the possession of the map, we could generate a "harvest project" to be carried out by the harvester, seeking to reap only the fruits of interest. To prove this situation, we divide the work into two stages: 1) With images, digitally identifies coffee fruits. We hope that by means of images taken by cell phones or cameras can estimate the productivity and maturation, using the techniques of image processing by Deep learning; 2) Through maps of productivity a... (Complete abstract click electronic access below) / Doutor

Deep Neural Networks (DNNs) are ubiquitous and span a variety of applications ranging from image classification and facial recognition to medical image analysis and real-time object detection. As DNN models become more sophisticated and complex, the computational cost of training these models becomes a burden. For this reason, outsourcing the training process has been the go-to option for many DNN users. Unfortunately, this comes at the cost of vulnerability to backdoor attacks. These attacks aim at establishing hidden backdoors in the DNN such that it performs well on clean samples but outputs a particular target label when a trigger is applied to the input. Current backdoor attacks generate triggers in the spatial domain; however, as we show in this work, it is not the only domain to exploit and one should always "check the other doors". To the best of our knowledge, this work is the first to propose a pipeline for generating a spatially dynamic (changing) and invisible (low norm) backdoor attack in the frequency domain. We show the advantages of utilizing the frequency domain for creating undetectable and powerful backdoor attacks through extensive experiments on various datasets and network architectures. Unlike most spatial domain attacks, frequency-based backdoor attacks can achieve high attack success rates with low poisoning rates and little to no drop in performance while remaining imperceptible to the human eye. Moreover, we show that the backdoored models (poisoned by our attacks) are resistant to various state-of-the-art (SOTA) defenses, and so we contribute two possible defenses that can successfully evade the attack. We conclude the work with some remarks regarding a network’s learning capacity and the capability of embedding a backdoor attack in the model.

As deep learning continues to push the boundaries with applications previously thought impossible, it has become more important than ever to reduce the associated resource costs. Data is not always abundant, labelling costs valuable human time, and deep models are demanding of computer hardware. In this dissertation, I will examine questions of minimalism in deep learning. I will show that deep learning can learn with fewer measurements, fewer weights, and less information. With minimalism, deep learning can become even more ubiquitous, succeeding in more applications and on more everyday devices.