004 Datenverarbeitung; Informatik
Refine
Year of publication
- 2022 (11) (remove)
Document Type
- Doctoral Thesis (11) (remove)
Has Fulltext
- yes (11)
Is part of the Bibliography
- no (11)
Keywords
- Autonomous Learning (1)
- CBM (1)
- CBM experiment (1)
- Cellular Automaton (1)
- Developmental Robotics (1)
- HPC (1)
- Heavy Ion experiments (1)
- Intrinsic Motivations (1)
- Kalman Filter (1)
- Künstliches Hormonsystem (1)
- Organic Computing (1)
- Prioritäten (1)
- Reinforcement Learning (1)
- Robotics (1)
- STAR (1)
- STAR experiment (1)
- Selbstorganisierende Taskverteilung (1)
- Strange particles (1)
- Tracking (1)
- epilepsy, epileptogenesis, model, neuro-immune, neuroinflammation, blood brain barrier, seizure (1)
Institute
Recent advances in artificial neural networks enabled the quick development of new learning algorithms, which, among other things, pave the way to novel robotic applications. Traditionally, robots are programmed by human experts so as to accomplish pre-defined tasks. Such robots must operate in a controlled environment to guarantee repeatability, are designed to solve one unique task and require costly hours of development. In developmental robotics, researchers try to artificially imitate the way living beings acquire their behavior by learning. Learning algorithms are key to conceive versatile and robust robots that can adapt to their environment and solve multiple tasks efficiently. In particular, Reinforcement Learning (RL) studies the acquisition of skills through teaching via rewards. In this thesis, we will introduce RL and present recent advances in RL applied to robotics. We will review Intrinsically Motivated (IM) learning, a special form of RL, and we will apply in particular the Active Efficient Coding (AEC) principle to the learning of active vision. We also propose an overview of Hierarchical Reinforcement Learning (HRL), an other special form of RL, and apply its principle to a robotic manipulation task.
This thesis presents a first-of-its-kind phenomenological framework that formally describes the development of acquired epilepsy and the role of the neuro-immune axis in this development. Formulated as a system of nonlinear differential equations, the model describes the interaction of processes such as neuroinflammation, blood- brain barrier disruption, neuronal death, circuit remodeling, and epileptic seizures. The model allows for the simulation of epilepsy development courses caused by a variety of neurological injuries. The simulation results are in agreement with ex- perimental findings from three distinct animal models of epileptogenesis. Simula- tions capture injury-specific temporal patterns of seizure occurrence, neuroinflam- mation, blood-brain barrier leakage, and progression of neuronal death. In addition, the model provides insights into phenomena related to epileptogenesis such as the emergence of paradoxically long time scales of disease development after injury, the dose-dependence of epileptogenesis features on injury severity, and the variability of clinical outcomes in subjects exposed to identical injury. Moreover, the developed framework allows for the simulation of therapeutic interventions, which provides insights into the injury-specificity of prominent intervention strategies. Thus, the model can be used as an in silico tool for the generation of testable predictions, which may aid pre-clinical research for the development of epilepsy treatments.
Die Emergenz digitaler Netzwerke ist auf die ständige Entwicklung und Transformation neuer Informationstechnologien zurückzuführen.
Dieser Strukturwandel führt zu äußerst komplexen Systemen in vielen verschiedenen Lebensbereichen.
Es besteht daher verstärkt die Notwendigkeit, die zugrunde liegenden wesentlichen Eigenschaften von realen Netzwerken zu untersuchen und zu verstehen.
In diesem Zusammenhang wird die Netzwerkanalyse als Mittel für die Untersuchung von Netzwerken herangezogen und stellt beobachtete Strukturen mithilfe mathematischer Modelle dar.
Hierbei, werden in der Regel parametrisierbare Zufallsgraphen verwendet, um eine systematische experimentelle Evaluation von Algorithmen und Datenstrukturen zu ermöglichen.
Angesichts der zunehmenden Menge an Informationen, sind viele Aspekte der Netzwerkanalyse datengesteuert und zur Interpretation auf effiziente Algorithmen angewiesen.
Algorithmische Lösungen müssen daher sowohl die strukturellen Eigenschaften der Eingabe als auch die Besonderheiten der zugrunde liegenden Maschinen, die sie ausführen, sorgfältig berücksichtigen.
Die Generierung und Analyse massiver Netzwerke ist dementsprechend eine anspruchsvolle Aufgabe für sich.
Die vorliegende Arbeit bietet daher algorithmische Lösungen für die Generierung und Analyse massiver Graphen.
Zu diesem Zweck entwickeln wir Algorithmen für das Generieren von Graphen mit vorgegebenen Knotengraden, die Berechnung von Zusammenhangskomponenten massiver Graphen und zertifizierende Grapherkennung für Instanzen, die die Größe des Hauptspeichers überschreiten.
Unsere Algorithmen und Implementierungen sind praktisch effizient für verschiedene Maschinenmodelle und bieten sequentielle, Shared-Memory parallele und/oder I/O-effiziente Lösungen.
Antimicrobial resistant infections arise as a consequential response to evolutionary mechanisms within microbes which cause them to be protected from the effects of antimicrobials. The frequent occurrence of resistant infections poses a global public health threat as their control has become challenging despite many efforts. The dynamics of such infections are driven by processes at multiple levels. For a long time, mathematical models have proved valuable for unravelling complex mechanisms in the dynamics of infections. In this thesis, we focus on mathematical approaches to modelling the development and spread of resistant infections at between-host (population-wide) and within-host (individual) levels.
Within an individual host, switching between treatments has been identified as one of the methods that can be employed for the gradual eradication of resistant strains on the long term. With this as motivation, we study the problem using dynamical systems and notions from control theory. We present a model based on deterministic logistic differential equations which capture the general dynamics of microbial resistance inside an individual host. Fundamentally, this model describes the spread of resistant infections whilst accounting for evolutionary mutations observed in resistant pathogens and capturing them in mutation matrices. We extend this model to explore the implications of therapy switching from a control theoretic perspective by using switched systems and developing control strategies with the goal of reducing the appearance of drug resistant pathogens within the host.
At the between-host level, we use compartmental models to describe the transmission of infection between multiple individuals in a population. In particular, we make a case study of the evolution and spread of the novel coronavirus (SARS-CoV-2) pandemic. So far, vaccination remains a critical component in the eventual solution to this public health crisis. However, as with many other pathogens, vaccine resistant variants of the virus have been a major concern in control efforts by governments and all stakeholders. Using network theory, we investigate the spread and transmission of the disease on social networks by compartmentalising and studying the progression of the disease in each compartment, considering both the original virus strain and one of its highly transmissible vaccine-resistant mutant strains. We investigate these dynamics in the presence of vaccinations and other interventions. Although vaccinations are of absolute importance during viral outbreaks, resistant variants coupled with population hesitancy towards vaccination can lead to further spread of the virus.
People can describe spatial scenes with language and, vice versa, create images based on linguistic descriptions. However, current systems do not even come close to matching the complexity of humans when it comes to reconstructing a scene from a given text. Even the ever-advancing development of better and better Transformer-based models has not been able to achieve this so far. This task, the automatic generation of a 3D scene based on an input text, is called text-to-3D scene generation. The key challenge, and focus of this dissertation, now relate to the following topics:
(a) Analyses of how well current language models understand spatial information, how static embeddings compare, and whether they can be improved by anaphora resolution.
(b) Automated resource generation for context expansion and grounding that can help in the creation of realistic scenes.
(c) Creation of a VR-based text-to-3D scene system that can be used as an annotation and active-learning environment, but can also be easily extended in a modular way with additional features to solve more contexts in the future.
(d) Analyze existing practices and tools for digital and virtual teaching, learning, and collaboration, as well as the conditions and strategies in the context of VR.
In the first part of this work, we could show that static word embeddings do not benefit significantly from pronoun substitution. We explain this result by the loss of contextual information, the reduction in the relative occurrence of rare words, and the absence of pronouns to be substituted. But we were able to we have shown that both static and contextualizing language models appear to encode object knowledge, but require a sophisticated apparatus to retrieve it. The models themselves in combination with the measures differ greatly in terms of the amount of knowledge they allow to extract.
Classifier-based variants perform significantly better than the unsupervised methods from bias research, but this is also due to overfitting. The resources generated for this evaluation are later also an important component of point three.
In the second part, we present AffordanceUPT, a modularization of UPT trained on the HICO-DET dataset, which we have extended with Gibsonien/telic annotations. We then show that AffordanceUPT can effectively make the Gibsonian/telic distinction and that the model learns other correlations in the data to make such distinctions (e.g., the presence of hands in the image) that have important implications for grounding images to language.
The third part first presents a VR project to support spatial annotation respectively IsoSpace. The direct spatial visualization and the immediate interaction with the 3D objects should make the labeling more intuitive and thus easier. The project will later be incorporated as part of the Semantic Scene Builder (SeSB). The project itself in turn relies on the Text2SceneVR presented here for generating spatial hypertext, which in turn is based on the VAnnotatoR. Finally, we introduce Semantic Scene Builder (SeSB), a VR-based text-to-3D scene framework using Semantic Annotation Framework (SemAF) as a scheme for annotating semantic relations. It integrates a wide range of tools and resources by utilizing SemAF and UIMA as a unified data structure to generate 3D scenes from textual descriptions and also supports annotations. When evaluating SeSB against another state-of-the-art tool, it was found that our approach not only performed better, but also allowed us to model a wider variety of scenes. The final part reviews existing practices and tools for digital and virtual teaching, learning, and collaboration, as well as the conditions and strategies needed to make the most of technological opportunities in the future.
In the human brain, the incoming light to the retina is transformed into meaningful representations that allow us to interact with the world. In a similar vein, the RGB pixel values are transformed by a deep neural network (DNN) into meaningful representations relevant to solving a computer vision task it was trained for. Therefore, in my research, I aim to reveal insights into the visual representations in the human visual cortex and DNNs solving vision tasks.
In the previous decade, DNNs have emerged as the state-of-the-art models for predicting neural responses in the human and monkey visual cortex. Research has shown that training on a task related to a brain region’s function leads to better predictivity than a randomly initialized network. Based on this observation, we proposed that we can use DNNs trained on different computer vision tasks to identify functional mapping of the human visual cortex.
To validate our proposed idea, we first investigate a brain region occipital place area (OPA) using DNNs trained on scene parsing task and scene classification task. From the previous investigations about OPA’s functions, we knew that it encodes navigational affordances that require spatial information about the scene. Therefore, we hypothesized that OPA’s representation should be closer to a scene parsing model than a scene classification model as the scene parsing task explicitly requires spatial information about the scene. Our results showed that scene parsing models had representation closer to OPA than scene classification models thus validating our approach.
We then selected multiple DNNs performing a wide range of computer vision tasks ranging from low-level tasks such as edge detection, 3D tasks such as surface normals, and semantic tasks such as semantic segmentation. We compared the representations of these DNNs with all the regions in the visual cortex, thus revealing the functional representations of different regions of the visual cortex. Our results highly converged with previous investigations of these brain regions validating the feasibility of the proposed approach in finding functional representations of the human brain. Our results also provided new insights into underinvestigated brain regions that can serve as starting hypotheses and promote further investigation into those brain regions.
We applied the same approach to find representational insights about the DNNs. A DNN usually consists of multiple layers with each layer performing a computation leading to the final layer that performs prediction for a given task. Training on different tasks could lead to very different representations. Therefore, we first investigate at which stage does the representation in DNNs trained on different tasks starts to differ. We further investigate if the DNNs trained on similar tasks lead to similar representations and on dissimilar tasks lead to more dissimilar representations. We selected the same set of DNNs used in the previous work that were trained on the Taskonomy dataset on a diverse range of 2D, 3D and semantic tasks. Then, given a DNN trained on a particular task, we compared the representation of multiple layers to corresponding layers in other DNNs. From this analysis, we aimed to reveal where in the network architecture task-specific representation is prominent. We found that task specificity increases as we go deeper into the DNN architecture and similar tasks start to cluster in groups. We found that the grouping we found using representational similarity was highly correlated with grouping based on transfer learning thus creating an interesting application of the approach to model selection in transfer learning.
During previous works, several new measures were introduced to compare DNN representations. So, we identified the commonalities in different measures and unified different measures into a single framework referred to as duality diagram similarity. This work opens up new possibilities for similarity measures to understand DNN representations. While demonstrating a much higher correlation with transfer learning than previous state-of-the-art measures we extend it to understanding layer-wise representations of models trained on the Imagenet and Places dataset using different tasks and demonstrate its applicability to layer selection for transfer learning.
In all the previous works, we used the task-specific DNN representations to understand the representations in the human visual cortex and other DNNs. We were able to interpret our findings in terms of computer vision tasks such as edge detection, semantic segmentation, depth estimation, etc. however we were not able to map the representations to human interpretable concepts. Therefore in our most recent work, we developed a new method that associates individual artificial neurons with human interpretable concepts.
Overall, the works in this thesis revealed new insights into the representation of the visual cortex and DNNs...
The main task of modern large experiments with heavy ions, such as CBM (FAIR), STAR (BNL) and ALICE (CERN) is a detailed study of the phase diagram of quantum chromodynamics (QCD) in the quark-gluon plasma (QGP), the equation of state of matter at extremely high baryonic densities, and the transition from the hadronic phase of matter to the quark-gluon phase.
In the thesis, the missing mass method is developed for the reconstruction of short-lived particles with neutral particles in their decay products, as well as its implementation in the form of fast algorithms and a set of software for prac- tical application in heavy ion physics experiments. Mathematical procedures implementing the method were developed and implemented within the KF Par- ticle Finder package for the future CBM (FAIR) experiment and subsequently adapted and applied for processing and analysis of real data in the STAR (BNL) experiment.
The KF Particle Finder package is designed to reconstruct most signal particles from the physics program of the CBM experiment, including strange particles, strange resonances, hypernuclei, light vector mesons, charm particles and char- monium. The package includes searches for over a hundred decays of short-lived particles. This makes the KF Particle Finder a universal platform for short-lived particle reconstruction and physics analysis both online and offline.
The missing mass method has been proposed to reconstruct decays of short-lived charged particles when one of the daughter particles is neutral and is not regis- tered in the detector system. The implementation of the missing mass method was integrated into the KF Particle Finder package to search for 18 decays with a neutral daughter particle.
Like all other algorithms of the KF Particle Finder package, the missing mass method is implemented with extensive use of vector (SIMD) instructions and is optimized for parallel operation on modern many-core high performance com- puter clusters, which can include both processors and coprocessors. A set of algorithms implementing the method was tested on computers with tens of cores and showed high speed and practically linear scalability with respect to the num- ber of cores involved.
It is extremely important, especially for the initial stage of the CBM experiment, which is planned for 2025, to demonstrate already now on real data the reliability of the developed approach, as well as the high efficiency of the current implemen- tation of both the entire KF Particle Finder package, and its integral part, the missing mass method. Such an opportunity was provided by the FAIR Phase-0 program, motivating the use in the STAR experiment of software packages orig- inally developed for the CBM experiment.
Application of the method to real data of the STAR experiment shows very good results with a high signal-to-background ratio and a large significance value. The results demonstrate the reliability and high efficiency of the missing mass method in the reconstruction of both charged mother particles and their neutral daughter particles. Being an integral part of the KF Particle Finder package, now the main approach for reconstruction and analysis of short-lived particles in the STAR experiment, the missing mass method will continue to be used for the physics analysis in online and offline modes.
The high quality of the results of the express data analysis has led to their status as preliminary physics results with the right to present them at international physics conferences and meetings on behalf of the STAR Collaboration.
Although everyone is familiar with using algorithms on a daily basis, formulating, understanding and analysing them rigorously has been (and will remain) a challenging task for decades. Therefore, one way of making steps towards their understanding is the formulation of models that are portraying reality, but also remain easy to analyse. In this thesis we take a step towards this way by analyzing one particular problem, the so-called group testing problem. R. Dorfman introduced the problem in 1943. We assume a large population and in this population we find a infected group of individuals. Instead of testing everybody individually, we can test group (for instance by mixing blood samples). In this thesis we look for the minimum number of tests needed such that we can say something meaningful about the infection status. Furthermore we assume various versions of this problem to analyze at what point and why this problem is hard, easy or impossible to solve.
The relevant field of interest in High Energy Physics experiments is shifting to searching and studying extremely rare particles and phenomena. The search for rare probes requires an increase in the number of available statistics by increasing the particle interaction rate. The structure of the events also becomes more complicated, the multiplicity of particles in each event increases, and a pileup appears. Due to technical limitations, such data flow becomes impossible to store fully on available storage devices. The solution to the problem is the correct triggering of events and real-time data processing.
In this work, the issue of accelerating and improving the algorithms for reconstruction of the charged particles' trajectories based on the Cellular Automaton in the STAR experiment is considered to implement them for track reconstruction in real-time within the High-Level Trigger. This is an important step in the preparation of the CBM experiment as part of the FAIR Phase-0 program. The study of online data processing methods in real conditions at similar interaction energies allows us to study this process and determine the possible weaknesses of the approach.
Two versions of the Cellular Automaton based track reconstruction are discussed, which are used, depending on the detecting systems' features. HFT~CA Track Finder, similar to the tracking algorithm of the CBM experiment, has been accelerated by several hundred times, using both algorithm optimization and data-level parallelism. TPC~CA Track Finder has been upgraded to improve the reconstruction quality while maintaining high calculation speed. The algorithm was tuned to work with the new iTPC geometry and provided an additional module for very low momentum track reconstruction.
The improved track reconstruction algorithm for the TPC detector in the STAR experiment was included in the HLT reconstruction chain and successfully tested in the express production for the online real data analysis. This made it possible to obtain important physical results during the experiment runtime without the full offline data processing. The tracker is also being prepared for integration into a standard offline data processing chain, after which it will become the basic track search algorithm in the STAR experiment.
Die allgemein steigende Komplexität technischer Systeme macht sich auch in eingebetteten Systemen bemerkbar. Außerdem schrumpfen die Strukturgrößen der eingesetzten Komponenten, was wiederum die Auftrittswahrscheinlichkeit verschiedener Effekte erhöht, die zu Fehlern und Ausfällen dieser Komponenten und damit der Gesamtsysteme führen können. Da in vielen Anwendungsbereichen ferner Sicherheitsanforderungen eingehalten werden müssen, sind zur Gewährleistung der Zuverlässigkeit flexible Redundanzkonzepte nötig.
Ein Forschungsgebiet, das sich mit Methoden zur Beherrschung der Systemkomplexität befasst, ist das Organic Computing. In dessen Rahmen werden Konzepte erforscht, um in natürlichen Systemen beobachtbare Eigenschaften und Organisationsprinzipien auf technische Systeme zu übertragen. Hierbei sind insbesondere sogenannte Selbst-X-Eigenschaften wie Selbstorganisation, -konfiguration und -heilung von Bedeutung.
Eine konkrete Ausprägung dieses Forschungszweigs ist das künstliche Hormonsystem (artificial hormone system, AHS). Hierbei handelt es sich um eine Middleware für verteilte Systeme, welche es ermöglicht, die Tasks des Systems selbstständig auf seine Prozessorelemente (PEs) zu verteilen und insbesondere Ausfälle einzelner Tasks oder ganzer PEs automatisch zu kompensieren, indem die betroffenen Tasks auf andere PEs migriert werden. Hierbei existiert keine zentrale Instanz, welche die Taskverteilung steuert und somit einen Single-Point-of-Failure darstellen könnte. Entsprechend kann das AHS aufgrund seiner automatischen (Re)konfiguration der Tasks als selbstkonfigurierend und selbstheilend bezeichnet werden, was insbesondere die Zuverlässigkeit des realisierten Systems erhöht. Die Dauer der Selbstkonfiguration und Selbstheilung unterliegt zudem harten Zeitschranken, was den Einsatz des AHS auch in Echtzeitsystemen erlaubt.
Das AHS nimmt jedoch an, dass alle Tasks gleichwertig sind, zudem werden alle Tasks beim Systemstart in einer zufälligen Reihenfolge auf die einzelnen PEs verteilt. Häufig sind die in einem System auszuführenden Tasks jedoch für das Gesamtsystem von unterschiedlicher Wichtigkeit oder müssen gar in einer bestimmten Reihenfolge gestartet werden.
Um den genannten Eigenschaften Rechnung zu tragen, liefert diese Dissertation gegenüber dem aktuellen Stand der Forschung folgende Beiträge:
Zunächst werden die bisher bekannten Zeitschranken des AHS genauer betrachtet und verfeinert.
Anschließend wird das AHS durch die Einführung von Zuteilungsprioritäten erweitert: Mithilfe dieser Prioritäten kann eine Reihenfolge definiert werden, in welcher die Tasks beim Start des Systems auf die PEs verteilt beziehungsweise in welcher betroffene Tasks nach einem Ausfall auf andere PEs migriert werden.
Die Zeitschranken dieser AHS-Erweiterung werden im Detail analysiert.
Durch die Priorisierung von Tasks ist es möglich, implizit Teilmengen von Tasks zu definieren, die ausgeführt werden sollen, falls die Rechenkapazitäten des Systems nach einer bestimmten Anzahl von PE-Ausfällen nicht mehr ausreichen, um alle Tasks auszuführen: Die im Rahmen dieser Dissertation entwickelten Erweiterungen erlauben es in solchen Überlastsituationen, das System automatisch und kontrolliert zu degradieren, sodass die wichtigsten Systemfunktionalitäten lauffähig bleiben.
Überlastsituationen werden daher im Detail betrachtet und analysiert. In solchen müssen gegebenenfalls Tasks niedriger Priorität gestoppt werden, um auf den funktionsfähig verbleibenden PEs hinreichend viel Rechenkapazität zu schaffen, um Tasks höherer Priorität ausführen zu können und das System so in einen wohldefinierten Zustand zu überführen. Die Entscheidung, in welcher Reihenfolge hierbei Tasks gestoppt werden, wird von einer Task-Dropping-Strategie getroffen, die entsprechend einen großen Einfluss auf die Dauer einer solchen Selbstheilung nimmt.
Es werden zwei verschiedene Task-Dropping-Strategien entwickelt und im Detail analysiert: die naive Task-Dropping-Strategie, welche alle niedrigprioren Tasks auf einmal stoppt, sowie das Eager Task Dropping, das in mehreren Phasen jeweils höchstens eine Task pro PE stoppt. Im Vergleich zeigt sich, dass von letzterem fast immer weniger Tasks gestoppt werden als von der naiven Strategie, was einen deutlich schnelleren Abschluss der Selbstheilung ermöglicht. Lediglich in wenigen Sonderfällen ist die naive Strategie überlegen.
Es wird detailliert gezeigt, dass die entwickelte AHS-Erweiterung auch in Überlastsituationen die Einhaltung bestimmter harter Zeitschranken garantieren kann, was den Einsatz des erweiterten AHS in Echtzeitsystemen erlaubt.
Alle theoretisch hergeleiteten Zeitschranken werden durch umfassende Evaluationen vollumfänglich bestätigt.
Abschließend wird das erweiterte, prioritätsbasierten AHS mit verschiedenen verwandten Konzepten verglichen, um dessen Vorteile gegenüber dem Stand der Forschung herauszuarbeiten sowie zukünftige vertiefende Forschung zu motivieren.
Machine learning (ML) techniques have evolved rapidly in recent years and have shown impressive capabilities in feature extraction, pattern recognition, and causal inference. There has been an increasing attention to applying ML to medical applications, such as medical diagnosis, drug discovery, personalized medicine, and numerous other medical problems. ML-based methods have the advantage of processing vast amounts of data.
With an ever increasing amount of medical data collection and large, inter-subject variability in the medical data, automated data processing pipelines are very much desirable since it is laborious, expensive, and error-prone to rely solely on human processing. ML methods have the potential to uncover interesting patterns, unravel correlations between complex features, learn patient-specific representations, and make accurate predictions. Motivated by these promising aspects, in this thesis, I present studies where I have implemented deep neural networks for the early diagnosis of epilepsy based on electroencephalography (EEG) data and brain tumor detection based on magnetic resonance spectroscopy (MRS) data.
In the project for early diagnosis of epilepsy, we are dealing with one of the most common neurological disorders, epilepsy, which is characterized by recurrent unprovoked seizures. It can be triggered by a variety of initial brain injuries and manifests itself after a time window which is called the latent period. During this period, a cascade of structural and functional brain alterations takes place leading to an increased seizure susceptibility.
The development and extension of brain tissue capable of generating spontaneous seizures is defined as epileptogenesis (EPG).
Detecting the presence of EPG provides a precious opportunity for targeted early medical interventions and, thus, can slow down or even halt the disease progression. In order to study brain signals in this latent window, animal epilepsy models are used to provide valuable data as it is extremely difficult to obtain this data from human patients. The aim of this study is to discover biomarkers of EPG using animal models and then to find the equivalent and counterparts in human patients' data. However, the EEG features for EPG are not well-understood and there is not a sufficiently large amount of annotated data for ML-based algorithms. To approach this problem, firstly, I utilized the timestamp information of the recorded EEG from an animal epilepsy model where epilepsy is induced by an electrical stimulation. The timestamp serves as a form of weak supervision, i.e., before and after the stimulation. Secondly, I implemented a deep residual neural network and trained it with a binary classification task to distinguish the EEG signals from these two phases. After obtaining a high discriminative ability on the binary classification task, I proposed to divide further the time span after the stimulation for a three-class classification, aiming to detect possible stages of the progression of the latent EPG phase. I have shown that the model can distinguish EEG signals at different stages of EPG with high accuracy and generalization ability. I have also demonstrated that some of the learned features from the network are clinically relevant.
In the task of detecting brain tumors based on MRS data, I first proposed to apply a deep neural network on the MRS data collected from over 400 patients for a binary classification task. To combat the challenge of noisy labeling, I developed a distillation step to filter out relatively ``cleanly'' labeled samples. A mixing-based data augmentation method was also implemented to expand the size of the training set. All the experiments were designed to be conducted with a leave-patient-out scheme to ensure the generalization ability of the model. Averaged across all leave-patient-out cross-validation sets, the proposed method performed on par with human neuroradiologists, while outperforming other baseline methods. I have demonstrated the distillation effect on the MNIST data set with manually-introduced label noise as well as providing visualization of the input influences on the final classification through a class activation map method.
Moreover, I have proposed to aggregate information at the subject level, which could provide more information and insights. This is inspired by the concept of multiple instance learning, where instance-level labels are not required and which is more tolerant to noisy labeling. I have proposed to generate data bags consisting of instances from each patient and also proposed two modules to ensure permutation invariance, i.e., an attention module and a pooling module. I have compared the performance of the network in different cases, i.e., with and without permutation-invariant modules, with and without data augmentation, single-instance-based and multiple-instance-based learning and have shown that neural networks equipped with the proposed attention or pooling modules can outperform human experts.