In this thesis, our focus is on low-complexity, high-performance detection algorithms for multi-antenna communication receivers. A key contribution in this thesis is the demonstration that efficient algorithms from metaheuristics and machine learning can be gainfully adapted for signal detection in multi- antenna communication receivers. We first investigate a popular metaheuristic known as the reactive tabu search (RTS), a combinatorial optimization technique, to decode the transmitted signals in large-dimensional communication systems. A basic version of the RTS algorithm is shown to achieve near-optimal performance for 4-QAM in large dimensions. We then propose a method to obtain a lower bound on the BER performance of the optimal detector. This lower bound is tight at moderate to high SNRs and is useful in situations where the performance of optimal detector is needed for comparison, but cannot be obtained due to very high computational complexity. To improve the performance of the basic RTS algorithm for higher-order modulations, we propose variants of the basic RTS algorithm using layering and multiple explorations. These variants are shown to achieve near-optimal performance in higher-order QAM as well. Next, we propose a new receiver called linear regression of minimum mean square error (MMSE) residual receiver (referred to as LRR receiver). The proposed LRR receiver improves the MMSE receiver by learning a linear regression model for the error of the MMSE receiver. The LRR receiver uses pilot data to estimate the channel, and then uses locally generated training data (not transmitted over the channel) to find the linear regression parameters. The LRR receiver is suitable for applications where the channel remains constant for a long period (slow-fading channels) and performs well. Finally, we propose a receiver that uses a committee of linear receivers, whose parameters are estimated from training data using a variant of the AdaBoost algorithm, a celebrated supervised classification algorithm in ma- chine learning. We call our receiver boosted MMSE (B-MMSE) receiver. We demonstrate that the performance and complexity of the proposed B-MMSE receiver are quite attractive for multi-antenna communication receivers.

Machine learning algorithms have been used for price prediction within several application areas. Examples include real estate, the stock market, tourist accommodation, electricity, art, cryptocurrencies, and fine wine. Common approaches in studies are to evaluate the accuracy of predictions and compare different algorithms, such as Linear Regression or Neural Networks. There is a thriving global second-hand market for vinyl records, but the research of price prediction within the area is very limited. The purpose of this project was to expand on existing knowledge within price prediction in general to evaluate some aspects of price prediction of vinyl records. That included investigating the possible level of accuracy and comparing the efficiency of algorithms. A dataset of 37000 samples of vinyl records was created with data from the Discogs website, and multiple machine learning algorithms were utilized in a controlled experiment. Among the conclusions drawn from the results was that the Random Forest algorithm generally generated the strongest results, that results can vary substantially between different artists or genres, and that a large part of the predictions had a good accuracy level, but that a relatively small amount of large errors had a considerable effect on the general results.

Accurate estimation of voltage drop (IR drop), in Application-Specific Integrated Circuits (ASICs) is a critical challenge, which impacts their performance and power consumption. As technology advances and die sizes shrink, predicting IR drop fast and accurate becomes increasingly challenging. This thesis focuses on exploring the application of Machine Learning (ML) algorithms, including Extreme Gradient Boosting (XGBoost), Convolutional Neural Network (CNN) and Graph Neural Network (GNN), to address this problem. Traditional methods of estimating IR drop using commercial tools are time consuming, especially for complex designs with millions of transistors. To overcome that, ML algorithms are investigated for their ability to provide fast and accurate IR drop estimation. This thesis utilizes electrical, timing and physical features of the ASIC design as input to train the ML models. The scalability of the selected features allows for their effective application across various ASIC designs with very few adjustments. Experimental results demonstrate the advantages of ML models over commercial tools, offering significant improvements in prediction speed. Notably, GNNs, such as Graph Convolutional Network (GCN) models showed promising performance with low prediction errors in voltage drop estimation. The incorporation of graph-structures models opens new fields of research for accurate IR drop prediction. The conclusions drawn emphasize the effectiveness of ML algorithms in accurately estimating IR drop, thereby optimizing ASIC design efficiency. The application of ML models enables faster predictions and noticeably reducing calculation time. This contributes to enhancing energy efficiency and minimizing environmental impact through optimised power circuits. Future work can focus on exploring the scalability of the models by training on a smaller portion of the circuit and extrapolating predictions to the entire design seems promising for more efficient and accurate IR drop estimation in complex ASIC designs. These advantages present new opportunities in the field and extend the capabilities of ML algorithms in the task of IR drop prediction. / Noggrann uppskattning av spänningsfallet (IR-fall), i ASIC är en kritisk utmaning som påverkar deras prestanda och strömförbrukning. När tekniken går framåt och formstorlekarna krymper, blir det allt svårare att förutsäga IR-fall snabbt och exakt. Denna avhandling fokuserar på att utforska tillämpningen av ML-algoritmer, inklusive XGBoost, CNN och GNN, för att lösa detta problem. Traditionella metoder för att uppskatta IR-fall med kommersiella verktyg är tidskrävande, särskilt för komplexa konstruktioner med miljontals transistorer. För att övervinna det undersöks ML-algoritmer för deras förmåga att ge snabb och exakt IR-falluppskattning. Denna avhandling använder elektriska, timing och fysiska egenskaper hos ASIC-designen som input för att träna ML-modellerna. Skalbarheten hos de valda funktionerna möjliggör deras effektiva tillämpning över olika ASIC-designer med mycket få justeringar. Experimentella resultat visar fördelarna med ML-modeller jämfört med kommersiella verktyg, och erbjuder betydande förbättringar i förutsägelsehastighet. Noterbart är att GNNs, såsom GCN-modeller, visade lovande prestanda med låga prediktionsfel vid uppskattning av spänningsfall. Införandet av grafstrukturmodeller öppnar nya forskningsfält för exakt IRfallförutsägelse. De slutsatser som dras betonar effektiviteten hos MLalgoritmer för att noggrant uppskatta IR-fall, och därigenom optimera ASICdesigneffektiviteten. Tillämpningen av ML-modeller möjliggör snabbare förutsägelser och märkbart minskad beräkningstid. Detta bidrar till att förbättra energieffektiviteten och minimera miljöpåverkan genom optimerade kraftkretsar. Framtida arbete kan fokusera på att utforska skalbarheten hos modellerna genom att träna på en mindre del av kretsen och att extrapolera förutsägelser till hela designen verkar lovande för mer effektiv och exakt IR-falluppskattning i komplexa ASIC-designer. Dessa fördelar ger nya möjligheter inom området och utökar kapaciteten hos ML-algoritmer i uppgiften att förutsäga IR-fall.

[ES] Esta tesis aborda el problema de fusión de datos en el ámbito de la neurociencia. El objetivo principal de este estudio es la fusión de modalidades, con énfasis en la fusión bimodal de señales biomédicas fMRI+EEG y de ECG+EEG. Las técnicas de fusión de datos tienen como objetivo alcanzar la exactitud y precisión en la toma de decisiones que sería más difícil con una sola modalidad. Hemos hecho una extensa revisión bibliográfica que contempla la fusión temprana y la fusión tardía de la siguiente manera: fusión temprana a nivel de sensores; fusión temprana a nivel de características; fusión tardía a nivel de scores; y fusión tardía a nivel de decisiones. En cada uno de esos apartados se presenta una tabla comparativa con las debilidades y fortalezas de cada método, así como los trabajos más citados. También hemos hecho aportes teóricos en esta área abordando el tema de la comparación entre la fusión temprana y la fusión tardía (soft y hard) para un problema multimodal de dos clases, dando elementos sobre la opción más adecuada a la hora de seleccionar la fusión temprana o tardía. Para este análisis hemos asumido inicialmente el conocimiento de los modelos utilizados., para después considerar modelos donde hay que estimar una serie de parámetros a partir de un conjunto de entrenamiento. El análisis se ha hecho para datos incorrelados y se ha extendido a datos con matrices de covarianza arbitrarias. Hemos realizado un estudio experimental como complemento del capítulo teórico. A partir de cuatro experimentos diferentes se destaca la efectividad de la fusión de datos multimodales para la mejora del rendimiento de los clasificadores. Los métodos de fusión y los clasificadores probados mostraron consistentemente un rendimiento superior en términos de métricas como el F1 score, la precisión, AUC y APR, en comparación con el uso de una sola modalidad de datos. Los resultados logrados subrayan la importancia de la fusión de datos en aplicaciones neurocientíficas y abren nuevas posibilidades para el desarrollo de sistemas de diagnóstico más precisos y robustos. / [CA] Aquesta tesi aborda el problema de la fusió de dades en l'àmbit de la neurociència. L'objectiu principal d'aquest estudi és la fusió de modalitats, amb èmfasi en la fusió bimodal de senyals biomèdiques fMRI+EEG i d'ECG+EEG. Les tècniques de fusió de dades tenen com a objectiu assolir l'exactitud i precisió en la presa de decisions que seria més difícil amb una sola modalitat. Hem fet una extensa revisió bibliogràfica que contempla la fusió primerenca i la fusió tardana de la següent manera: fusió primerenca a nivell de sensors; fusió primerenca a nivell de característiques; fusió tardana a nivell de puntuacions; i fusió tardana a nivell de decisions. En cadascun d'aquests apartats es presenta una taula comparativa amb les debilitats i fortaleses de cada mètode, així com els treballs més citats. També hem fet aportacions teòriques en aquesta àrea abordant el tema de la comparació entre la fusió primerenca i la fusió tardana (suau i dura) per a un problema multimodal de dues classes, donant elements sobre l'opció més adequada a l'hora de seleccionar la fusió primerenca o tardana. Per a aquesta anàlisi, hem assumit inicialment el coneixement dels models utilitzats, per després considerar models on cal estimar una sèrie de paràmetres a partir d'un conjunt d'entrenament. L'anàlisi s'ha fet per a dades incorrelades i s'ha estès a dades amb matrius de covariància arbitràries. Hem realitzat un estudi experimental com a complement del capítol teòric. A partir de quatre experiments diferents es destaca l'efectivitat de la fusió de dades multimodals per a la millora del rendiment dels classificadors. Els mètodes de fusió i els classificadors provats van mostrar constantment un rendiment superior en termes de mètriques com el F1 score, la precisió, AUC i APR, en comparació amb l'ús d'una sola modalitat de dades. Els resultats obtinguts subratllen la importància de la fusió de dades en aplicacions neurocientífiques i obrin noves possibilitats per al desenvolupament de sistemes de diagnòstic més precisos i robusts. / [EN] This thesis addresses the problem of data fusion in the field of neuroscience. The main objective of this study is to explore multimodal fusion, with an emphasis on bimodal fusion of biomedical signals such as fMRI+EEG and ECG+EEG. Data fusion techniques aim to achieve accuracy and precision in decision-making that would be more challenging with a single modality. We have conducted an extensive literature review covering early fusion and late fusion, as follows: early fusion at the sensor level, early fusion at the feature level, late fusion at the score level, and late fusion at the decision level. In each of these sections, we present a comparative table outlining the strengths and weaknesses of each method, as well as the most cited works. We have also made theoretical contributions to this area by addressing the comparison between early and late fusion (both soft and hard) for a two-class multimodal problem, providing insights into the most suitable choice between early and late fusion. For this analysis, we initially assumed knowledge of the models used, then considered scenarios where a series of parameters must be estimated from a training set. The analysis was conducted for uncorrelated data and extended to data with arbitrary covariance matrices. We conducted an experimental study to complement the theoretical chapter. Based on four different experiments, the effectiveness of multimodal data fusion in enhancing classifier performance was highlighted. The tested fusion methods and classifiers consistently demonstrated superior performance in terms of metrics such as F1 score, precision, AUC, and APR compared to using a single data modality. The results emphasize the importance of data fusion in neuroscientific applications and open up new possibilities for developing more accurate and robust diagnostic systems. / Pereira González, LM. (2024). Nuevas contribuciones en aplicaciones de fusión multimodal de bioseñales [Tesis doctoral]. Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/213614

In forensic autopsies, accurately estimating the postmortem interval (PMI) is crucial. Traditional methods, relying on physical parameters and police data, often lack precision, particularly after approximately two days have passed since the person's death. New methods are increasingly focusing on analyzing postmortem metabolomics in biological systems, acting as a 'fingerprint' of ongoing processes influenced by internal and external molecules. By carefully analyzing these metabolomic profiles, which span a diverse range of information from events preceding death to postmortem changes, there is potential to provide more accurate estimates of the PMI. The limitation of available real human data has hindered comprehensive investigation until recently. Large-scale metabolomic data collected by the National Board of Forensic Medicine (RMV, Rättsmedicinalverket) presents a unique opportunity for predictive analysis in forensic science, enabling innovative approaches for improving  PMI estimation. However, the metabolomic data appears to be large, complex, and potentially nonlinear, making it difficult to interpret. This underscores the importance of effectively employing machine learning algorithms to manage metabolomic data for the purpose of PMI predictions, the primary focus of this project.  In this study, a dataset consisting of 4,866 human samples and 2,304 metabolites from the RMV was utilized to train a model capable of predicting the PMI. Random Forest (RF) and Artificial Neural Network (ANN) models were then employed for PMI prediction. Furthermore, feature selection and incorporating sex and age into the model were explored to improve the neural network's performance.  This master's thesis shows that ANN consistently outperforms RF in PMI estimation, achieving an R2 of 0.68 and an MAE of 1.51 days compared to RF's R2 of 0.43 and MAE of 2.0 days across the entire PMI-interval. Additionally, feature selection indicates that only 35% of total metabolites are necessary for comparable results with maintained predictive accuracy. Furthermore, Principal Component Analysis (PCA) reveals that these informative metabolites are primarily located within a specific cluster on the first and second principal components (PC), suggesting a need for further research into the biological context of these metabolites.  In conclusion, the dataset has proven valuable for predicting PMI. This indicates significant potential for employing machine learning models in PMI estimation, thereby assisting forensic pathologists in determining the time of death. Notably, the model shows promise in surpassing current methods and filling crucial gaps in the field, representing an important step towards achieving accurate PMI estimations in forensic practice. This project suggests that machine learning will play a central role in assisting with determining time since death in the future.

No description available.