Petia Georgieva | University of Aveiro (original) (raw)

Papers by Petia Georgieva

A summary is provided of the widely used control paradigms and their novel counterparts applied i... more A summary is provided of the widely used control paradigms and their novel counterparts applied in a variety of process industries, including the dairy industry. Though basic dairy processes have changed little in the past decade, the general demands of lower cost, higher product quality, and more environmentally friendly solutions lead to the necessity of improving the traditional control structures. This review provides a general idea about the main practical and research lines in dairy process control. It summarizes the proportional integral derivative (PID) controller as a control engineering practice that has been established for decades, various statistical process control (SPC) techniques widely implemented in batch and fed-batch dairy plants, and recent intelligent control (IC) alternatives. Fuzzy logic control systems (FLCSs) and artificial neural network (ANN)-based model predictive control (MPC) are the main IC paradigms described.

Review of Scientific Instruments, 2021

Aquaculture is a fundamental sector of the food industry nowadays. However, to become a sustainab... more Aquaculture is a fundamental sector of the food industry nowadays. However, to become a sustainable and more profitable industry, it is necessary to monitor several associated parameters, such as temperature, salinity, ammonia, potential of hydrogen, nitrogen dioxide, bromine, among others. Their regular and simultaneous monitoring is expected to predict and avoid catastrophes, such as abnormal fish mortality rates. In this paper, we propose a novel anomaly detection approach for the early prediction of high fish mortality based on a multivariate Gaussian probability model. The goal of this approach is to determine the correlation between the number of daily registered physicochemical parameters of the fish tank water and the fish mortality. The proposed machine learning model was fitted with data from the weaning and pre-fattening phases of Senegalese sole (Solea senegalensis) collected over 2018, 2019, and 2020. This approach is suitable for real-time tracking and successful predi...

Technologies, 2021

Floor-cleaning robots are becoming increasingly more sophisticated over time and with the additio... more Floor-cleaning robots are becoming increasingly more sophisticated over time and with the addition of digital cameras supported by a robust vision system they become more autonomous, both in terms of their navigation skills but also in their capabilities of analyzing the surrounding environment. This document proposes a vision system based on the YOLOv5 framework for detecting dirty spots on the floor. The purpose of such a vision system is to save energy and resources, since the cleaning system of the robot will be activated only when a dirty spot is detected and the quantity of resources will vary according to the dirty area. In this context, false positives are highly undesirable. On the other hand, false negatives will lead to a poor cleaning performance of the robot. For this reason, a synthetic data generator found in the literature was improved and adapted for this work to tackle the lack of real data in this area. This synthetic data generator allows for large datasets with ...

Pattern Recognition and Image Analysis, 2019

Games have been used as an effective way to measure the advancement of artificial intelligence. C... more Games have been used as an effective way to measure the advancement of artificial intelligence. Chess, Atari2600 and Go, are some of the most mediatic demonstrations where AI computer programs defeated human players. In this paper we add the popular Flappy Bird game in the list of games to quantify the performance of an AI player. Based on Q-Reinforcement Learning and Neuroevolution (neural network fitted by genetic algorithm), artificial agents were trained to take the most favorable action at each game instant. The Neuroevolution agent outperformed by far the Reinforcement Learning agent (111 points average result) and achieved on average super-human performance of impressive score of 28700 points.

Metals

Accurate numerical simulations require constitutive models capable of providing precise material ... more Accurate numerical simulations require constitutive models capable of providing precise material data. Several calibration methodologies have been developed to improve the accuracy of constitutive models. Nevertheless, a model’s performance is always constrained by its mathematical formulation. Machine learning (ML) techniques, such as artificial neural networks (ANNs), have the potential to overcome these limitations. Nevertheless, the use of ML for material constitutive modelling is very recent and not fully explored. Difficulties related to data requirements and training are still open problems. This work explores and discusses the use of ML techniques regarding the accuracy of material constitutive models in metal plasticity, particularly contributing (i) a parameter identification inverse methodology, (ii) a constitutive model corrector, (iii) a data-driven constitutive model using empirical known concepts and (iv) a general implicit constitutive model using a data-driven learn...

FABULOUS 2019 International Conference on Future Access Enablers of Ubiquitous and Intelligent Infrastructures, 2019

Modern air quality monitoring systems are characterised by high complexity and costs. The expensi... more Modern air quality monitoring systems are characterised by high complexity and costs. The expensive embedded units such as sensor arrays, processors, power blocks, displays and communication units make them less appropriate for small indoor spaces.

2018 International Joint Conference on Neural Networks (IJCNN), 2018

Despite numerous successful applications of Deep Learning (DL) to large-scale image, video, speec... more Despite numerous successful applications of Deep Learning (DL) to large-scale image, video, speech and text data, they remain relatively unexplored in brain imaging field. In this paper, we make an overview of recent DL architectures for recognizing cognitive brain activities from Electroencephalogram (EEG) data with particular emphasis on Brain Computer Interface(BCI) technologies and Affective Neurocomputing. We discuss the use of convolutional, recurrent neural nets, as well as deep belief networks, echo-state networks, reservoir computing, and denoising auto encoder models. A major challenge in modeling brain cognitive activity from EEG data is finding representations that are invariant to inter- and intra-subject differences, as well as the inherent noise in the EEG recordings. The reviewed studies reveal the great potential of DL to decode human intentions in BCI applications and to find the invariant descriptors of human emotions across subjects in Affective Neurocomputing applications. Many of the DL models prove to be more accurate and efficient than traditional machine learning models.

2016 IEEE 8th International Conference on Intelligent Systems (IS), 2016

Certain professions rely on the ability to maintain attention constant throughout long periods of... more Certain professions rely on the ability to maintain attention constant throughout long periods of time, like truck drivers, air traffic controllers, health professionals, among others. These could greatly benefit from the development of a real-time alerting system that will call subjects back to task even before lapses occur or shortly after they happened. Attention levels have been shown to relate to the properties of the electroencephalogram (EEG). In this paper, we propose for the first time a regression approach to detect fluctuating levels of attention, based on spatiotemporal patterns extracted from EEG recordings. Previous studies have shown that reaction time is related to the level of task related attention. Moment-to-moment fluctuations in attention level are paralleled by moment-to-moment fluctuations in reaction time (faster reaction times are related to high attention allocation). We took advantage of this parallel and used reaction time data obtained during a repetitive visuomotor task as a proxy for task related attention level. Furthermore, instead of defining high attention versus low attention periods, we labeled each moment according to a continuum based on each trial's reaction time. In order to determine if it is possible to predict attention level from EEG features, we developed regression models between the extracted features and the subject's reaction time.

Lecture Notes of the Institute for Computer Sciences, Social Informatics and Telecommunications Engineering, 2015

The use of general descriptive names, registered names, trademarks, service marks, etc. in this p... more The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use.

2015 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), 2015

Localization of the brain neural generators that create Electroencephalographs (EEGs) has been an... more Localization of the brain neural generators that create Electroencephalographs (EEGs) has been an important problem in clinical, research and technological applications related to the brain. The active regions in the brain are modeled as equivalent current dipoles, and the positions and moments of these dipoles or brain sources are estimated. So far, the brain dipoles are assumed to be fixed or time-invariant. However, recent neurological studies are showing that brain sources are not static but vary (in terms of location and moment) depending on various internal and external stimuli. This paper presents a shift in the current paradigm of brain source localization by considering dynamic sources in the brain. We formulate the brain source estimation problem from EEG measurements as a (nonlinear) state-space model. We use the Particle Filter (PF), essentially a sequential Monte Carlo method, to track the trajectory of the moving dipoles in the brain. We further address the "curse of dimensionality," issue of the PF by taking advantage of the structure of the EEG state-space model, and marginalizing out the linearly evolving states. A Kalman Filter is used to optimally estimate the linear elements, whereas the PF is used to track only the non-linear components. This technique reduces the dimension of the problem; thus exponentially reducing the computational cost. Our simulation results show that, where the PF fails, the Marginalized PF is able to successfully track two dipoles in the brain with only 500 particles.

Neural Networks, 2016

The aim of this paper is to identify the common neural signatures based on which the positive and... more The aim of this paper is to identify the common neural signatures based on which the positive and negative valence of human emotions across multiple subjects can be reliably discriminated. The brain activity is observed via Event Related Potentials (ERPs). ERPs are transient components in the Electroencephalography (EEG) generated in response to a stimulus. ERPs were collected while subjects were viewing images with positive or negative emotional content. Building inter-subject discrimination models is a challenging problem due to the high ERPs variability between individuals. We propose to solve this problem with the aid of the Echo State Networks (ESN) as a general framework for extracting the most relevant discriminative features between multiple subjects. The original feature vector is mapped into the reservoir feature space defined by the number of the reservoir equilibrium states. The dominant features are extracted iteratively from low dimensional combinations of reservoir states. The relevance of the new feature space was validated by experiments with standard supervised and unsupervised machine learning techniques. From one side this proof of concept application enhances the usability context of the reservoir computing for high dimensional static data representations by lowdimensional feature transformation as functions of the reservoir states. From other side, the proposed solution for emotion valence detection across subjects is suitable for brain studies as a complement to statistical methods. This problem is important because such decision making systems constitute "virtual sensors" of hidden emotional states, which are useful in psychology science research and clinical applications.

Citeseer

Abstract: This paper is focused on a comprehensive study of neural network (NN) model based predi... more Abstract: This paper is focused on a comprehensive study of neural network (NN) model based predictive control (MPC), as an operation strategy for a fed-batch sugar crystallizer. The process is divided into four subsequent control loops and for each of them an individual NN-based ...

2021 International Conference on INnovations in Intelligent SysTems and Applications (INISTA), 2021

Recent years have seen significant advances in automated diagnosis systems for medical imaging ta... more Recent years have seen significant advances in automated diagnosis systems for medical imaging tasks aimed to support the decision-making process. More specifically, Convolutional neural networks (CNN) show remarkable performance in tasks such as multi-class skin lesion classification using images. However, concerns remain about the deployment of such models, as real-world test data distribution can significantly differ from the distribution of the training data. In other words, models can classify unknown samples as known classes with high confidence, which could lead to catastrophic mistakes. In line with these concerns, this paper focuses on accessing the current methods to detect out-of-training distribution samples in the context of skin lesion classification. The results contribute towards the understanding of the effectiveness of out-of-distribution detection methods.

Annals of Mathematics and Artificial Intelligence, 2019

Deep Learning methods have been rising in popularity in the past few years, and are now used as a... more Deep Learning methods have been rising in popularity in the past few years, and are now used as a fundamental component in various application domains such as computer vision, natural language processing, bioinformatics. Supervised learning with Convolutional Neural Networks has become the state of the art approach in many image related works. However, despite the great success of deep learning methods in other areas they remain relatively unexplored in the brain imaging field. In this paper we make an overview of recent achievements of Deep Learning to automatically extract features from brain signals that enable building Brain-Machine Interfaces (BMI). Major challenge in the BMI research is to find common subject-independent neural signatures due to the high brain data variability across multiple subjects. To address this problem we propose a Deep Neural Autoencoder with sparsity constraint as a promising approach to extract hidden features from Electroencephalogram data (in-dept feature learning) and build a subject-independent noninvasive BMI in the affective neuro computing framework. Future direction for research are also outlined.

$Research paper thumbnail of Antimicrobial and immunological activity of ethanol extracts and fractions from Isopyrum thalictroides$

Journal of Ethnopharmacology, 1996

The antimicrobial and immunological properties of ethanol extracts, non-alkaloid, tertiary alkalo... more The antimicrobial and immunological properties of ethanol extracts, non-alkaloid, tertiary alkaloid and quaternary alkaloid fractions, obtained from roots and aerial parts of Isopyrum thalictroides were examined. The non-alkaloid fraction from aerial parts inhibited the growth of seven test microorganisms and was the most effective suppresser of classical pathway (CP) complement activity in normal human serum (NHS) and guinea pig serum (GPS). The alkaloid fractions, containing quaternary alkaloids expressed suppressive effect on mitogen-induced splenocyte proliferation. The in vitro antibody response against sheep red blood cells (anti-SRBC) was inhibited by ethanol extracts and quaternary alkaloid fraction. The intraperitoneal (i.p.) application of ethanol extract and tertiary alkaloid fraction from aerial parts showed that they possess in vivo effect on alternative pathway (APt complement activity, anti-SRBC response and delayed type hypersensitivity (DTH).

Computers & Chemical Engineering, 2010

A supply chain (SC) model incorporating business strategic decision components is an important to... more A supply chain (SC) model incorporating business strategic decision components is an important tool for gaining a competitive edge in todays global market. Enterprise models of this type must encompass not only the SC, but also the demand chain since understanding the market is crucial for developing good business policies. To operate effectively, marketing activities must be coordinated with other

This paper discusses a linear time invariant (LTI) systems approach to signal enhancement via pro... more This paper discusses a linear time invariant (LTI) systems approach to signal enhancement via projective subspace techniques. It provides closed form expressions for the frequency response of data adaptive finite impulse response eigenfilters. An illustrative example ...

This paper is focused on issues of nonlinear dynamic process modeling and model-based predictive ... more This paper is focused on issues of nonlinear dynamic process modeling and model-based predictive control of a fed-batch sugar crystallization process applying the concept of artificial neural networks as computational tools. The control objective is to force the operation into following optimal supersaturation trajectory. It is achieved by manipulating the feed flow rate of sugar liquor/syrup, considered as the control input. A feed forward neural network (FFNN) model of the process is first built as part of the controller structure to predict the process response over a specified (prediction) horizon. The predictions are supplied to an optimization procedure to determine the values of the control action over a specified (control) horizon that minimizes a predefined performance index. The control task is rather challenging due to the strong nonlinearity of the process dynamics and variations in the crystallization kinetics. However, the simulation results demonstrated smooth behavio...