Spiking neural networks are a new generation of neural networks that use neuronal models that are more biologically plausible than the typically used perceptron model. They do not use analog values to perform computations, as is the case in regular neural networks, but rely on spatio-temporal information encoded into sequences of delta-functions known as spike trains. Spiking neural networks are highly energy efficient compared to regular neural networks which makes them highly attractive in certain applications. This thesis implements two approaches for training spiking neural networks. The first approach uses surrogate gradient descent to deal with the issues of non-differentiability that arise with training spiking neural networks. The second approach is based on Bayesian probability theory and uses variational inference for parameter estimation and leads to a Bayesian spiking neural network. The two methods are tested on two datasets from the spiking neural network literature and limited hyperparameter studies are performed. The results indicate that both training methods work on the two datasets but that the Bayesian implementation yields a lower accuracy on test data. Moreover, the Bayesian implementation appear to be robust to the choice of prior parameter distribution. / <p>Sekretess</p>

Prediction of stock prices is an important financial problem that is receiving increased attention in the field of artificial intelligence. Many different neural network and hybrid models for obtaining accurate prediction results have been proposed during the last few years in an attempt to outperform the traditional linear and nonlinear approaches. This study evaluates the performance of three different hybrid neural network models used for one-day stock close price prediction; a pre-processed evolutionary Levenberg-Marquardt neural network, Bayesian regularized artificial neural network and neural network with technical- and fractal analysis. It was also determined which of the three outperformed the others. The performance evaluation and comparison of the models are done using statistical error measures for accuracy; mean square error, symmetric mean absolute percentage error and point of change in direction. The results indicate good performance values for the Bayesian regularized artificial neural network, and varied performance for the others. Using the Friedman test, one model clearly is different in its performance relative to the others, probably the above mentioned model. The results for two of the models showed a large standard deviation of the error measurements which indicates that the results are not entirely reliable.

The amount of water flowing through streams and rivers changes through time. The seasonality and duration of these changes can have profound impacts on human freshwater availability, aquatic habitat, and biogeochemical cycling. Numerous factors are thought to influence streamflow regime, including drainage basin area, temperature, precipitation, and land cover. Few of these qualities have remained untouched, either directly or indirectly, by expanding human activities. Altered climate, sweeping changes to large portions of the earth's surface, and the construction of dams and other infrastructure have fundamentally altered streamflows worldwide. Understanding the nature of these changes, both globally and regionally in the Western United States, is the subject of this thesis. In chapter 1 we explore ideal metric spaces for describing streamflow regime. The representation of information in concise terms is usually preliminary to developing an understanding of any system, and streamflow regime, which has been described with over 600 unique variables, is no exception. We demonstrate the efficacy of dimensionality reduction techniques, as well as frequency decompositions, in succinctly capturing much of the information previously described with hundreds of variables. We use this succinct language to gain key insights into major drivers of streamflow regime and present a new hypothesis about the mechanisms mediating flow variability. In chapter 2, we use frequency decompositions and several machine learning approaches to characterize streamflow regimes around the world and to understand how they are changing through time. Finally, in chapter 3, we analyze the effect that wildfire has had on the timing, amount, and variability of flow in the western US in recent decades. The work presented here demonstrates the power that advances in data science, particularly in time series analysis methods and machine learning, can have when coupled with large datasets in revealing insights into global and regional phenomena in hydrology.

Neural networks have long been known as universal function approximators and have more recently been shown to be powerful and versatile in practice. But it can be extremely challenging to find the right set of parameters and hyperparameters. Model training is both expensive and difficult due to the large number of parameters and sensitivity to hyperparameters such as learning rate and architecture. Hyperparameter searches are notorious for requiring tremendous amounts of processing power and human resources. This thesis provides an analytic approach to estimating the optimal value of one of the key hyperparameters in neural networks, the learning rate. Where possible, the analysis is computed exactly, and where necessary, approximations and assumptions are used and justified. The result is a method that estimates the optimal learning rate for a certain type of network, a fully connected CReLU network.

El texto completo de este trabajo no está disponible en el Repositorio Académico UPC por restricciones de la casa editorial donde ha sido publicado. / This article presents the design of an unmanned aerial vehicle manufactured in aramid, through the use of sensors and actuators for flight stabilization, capturing the images through a thermal imager and its wireless transmission for ground processing for application in the social security area used in fire accidents. The work shows that it is feasible to use the aramid material for the construction of the prototype, since it is a high temperature resistant material, also the integration of neural networks for semi-automatic flight control. The results of this research will serve to develop more advanced control devices, with simple components and controls so that people with technological limitations can use it, so that they can save lives in danger, that of their colleagues or themselves. / Revisión por pares

The construction and operation of central wastewater treatment plants started around the 20th century. With the advent of rigorous membrane research and development in the middle of the 20th century, more and more wastewater plants started incorporating a Membrane BioReactor, MBR, in their design. The MBR system however is far from perfect. Membrane systems continuously foul, and if fouling is incurred for a long period of time, maintenance and cleaning costs will rise in proportion. A Fouling monitoring and prediction tool has been designed in MATLAB\Simulink. The model takes states related to membrane fouling, and calculates the membrane total resistance based on deterministic and stochastic models. The tool is capable of predicting future TMP cycles based on older TMP performance via an artificial neural network algorithm. TMP data have been synthetically generated from a validated mathematical model. Finally, an artificial neural network controller is implemented to control temperature and MLSS around their desired setpoints. The controller is able to minimize disturbances in both states in a narrow band around their desired setpoints.

The development of machine learning research has provided statistical innovations and further developments within the field of time series analysis. This study seeks to investigate two different approaches on artificial neural network models based on different learning techniques, and answering how well the neural network approach compares with a basic autoregressive approach, as well as how the artificial neural network models compare to each other. The models were compared and analyzed in regards to the univariate forecast accuracy on 20 randomly drawn time series from two different time frequencies from the M4 competition dataset. Forecasting was made dependent on one time lag (t-1) and forecasted three and six steps ahead respectively. The artificial neural network models outperformed the baseline Autoregressive model, showing notably lower mean average percentage error overall. The Multilayered perceptron models performed better than the Long short-term memory model overall, whereas the Long short-term memory model showed improvement on longer prediction time dimensions. As the training were done univariately  on a limited set of time steps, it is believed that the one layered-approach gave a good enough approximation on the data, whereas the added layer couldn’t fully utilize its strengths of processing power. Likewise, the Long short-term memory model couldn’t fully demonstrate the advantagements of recurrent learning. Using the same dataset, further studies could be made with another approach to data processing. Implementing an unsupervised approach of clustering the data before analysis, the same models could be tested with multivariate analysis on models trained on multiple time series simultaneously.

Biometric identification has already been applied to society today, as today’s mobile phones use fingerprints and other methods like iris and the face itself. With growth for technologies like computer vision, the Internet of Things, Artificial Intelligence, The use of face recognition as a biometric identification on ordinary doors has become increasingly common. This thesis studies is looking into the possibility of replacing regular door locks with face recognition or supplement the locks to increase security by using a pre-trained state-of-the-art face recognition method based on a convolution neural network. A subsequent investigation concluded that a networks based face recognition are is highly vulnerable to attacks in the form of presentation attacks. This study investigates protection mechanisms against these forms of attack by developing a presentation attack detection and analyzing its performance. The obtained results from the proof of concept  showed that local binary patterns histograms as a presentation attack detection could help the state of art face recognition to avoid attacks up to 88\% of the attacks the convolution neural network approved without the presentation attack detection. However, to replace traditional locks, more work must be done to detect more attacks in form of both higher percentage of attacks blocked by the system and the types of attack that can be done. Nevertheless, as a supplement face recognition represents a promising technology to supplement traditional door locks, enchaining their security by complementing the authorization with biometric authentication. So the main contributions is that  by using simple older methods LBPH can help modern state of the art face regognition to detect presentation attacks according to the results of the tests. This study also worked to adapt this PAD to be suitable for low end edge devices to be able to adapt in an environment where modern solutions are used, which LBPH have.

The dissertation proposed an approach, named "Signal Optimization Under Rail Crossing sAfety cOnstraints"(SOURCAO), to the traffic signal control near a highway rail grade crossing (HRGC). SOURCAO targets two objectives: HRGC safety improvement (a high priority national transportation goal) and highway traffic delay reduction (a common desire for virtually all of us). Communication and data availability from ITS and the next generation train control are assumed available in SOURCAO. The first step in SOURCAO is to intelligently choose a proper preemption phase sequence to promote HRGC safety. An inference engine is designed in place of traditional traffic signal preemption calls to prevent the queue from backing onto HRGC. The potential hazard is dynamically examined as to whether any queuing vehicle stalls on railroad tracks. The inference engine chooses the appropriate phase sequence to eliminate the hazardous situation. The second step in SOURCAO is to find the optimized phase length. The optimization process uses the network traffic delay (close to the control delay) at the intersections within HRGC vicinities as an objective function. The delay function is approximated and represented by multilayer perceptron neural network (off-line). After the function was trained and obtained, an optimization algorithm named Successive Quadratic Programming (SQP) searches the length of phases (on-line) by minimizing the delay function. The inference engine and proposed delay model in optimization take the on-line surveillance detector data and HRGC closure information as input. By integrating artificial intelligence and optimization technologies, the independent simulation evaluation of SOURCAO by TSIS/CORSIM demonstrated that the objectives are reached. The average network delay for 20 runs of simulation evaluation is reduced over eight percent by a t-test while the safety of HRGC is promoted. The sensitivity tests demonstrate that SOURCAO works efficiently under light and heavy traffic conditions, as well as a wide range of HRGC closure times. / Ph. D.

No description available.