Refine
Year of publication
Document Type
- Doctoral Thesis (55) (remove)
Language
- English (55) (remove)
Has Fulltext
- yes (55)
Is part of the Bibliography
- no (55)
Keywords
- ALICE (2)
- FPGA (2)
- Abfrageverarbeitung (1)
- Abstraction (1)
- Agent <Künstliche Intelligenz> (1)
- Agenten (1)
- Agents (1)
- Analog (1)
- Analog Circuits (1)
- Analog Verification (1)
Institute
- Informatik (55) (remove)
A Large Ion Collider Experiment (ALICE) is a high-energy physics experiment, designed to study heavy ion collisions at the European Organization for Nuclear Research (CERN)Large Hadron Collider (LHC). ALICE is built to study the fundamental properties of matter as it existed shortly after the big bang. This requires reading out millions of sensors with high frequency, enabling high statistics for physics analysis, resulting in a considerable computing demand concerning network throughput and processing power. With the ALICE Run 3 upgrade [14], requirements for a High Throughput Computing
(HTC) online processing cluster increased significantly, due to more than an order of magnitude more data than in Run 2, resulting in a processing input rate of up to 900 GB/s. Online (real-time) event reconstruction allows for the compression of the data stream to 130 GB/s, which is stored on disk for physics analysis.
This thesis presents the implementation of the ALICE Event Processing Node (EPN) compute farm, to cope with the Run 3 online computing challenges. Building a Data Centre tailored to ALICE requirements for the Run 3 and Run 4 EPN farm. Providing the operational conditions for a dynamic compute environment of a High Performance Computing (HPC) cluster, with significant load changes in a short time span, when starting or stopping a data-taking run. EPN servers provide the required computing resources for online reconstruction and data compression. The farm includes network connectivity towards First Level Processors (FLPs), requiring reliable throughput of 900 GB/s between FLPs and EPNs and connectivity from the internal InfiniBand network to the CERN Exabyte Object Storage (EOS) Ethernet network, with more than 100 GB/s.
The results of operating the EPN computing infrastructure during the first year of Run 3 LHC collisions are described in the context of the ALICE experiment. The EPN farm was delivering the expected performance for ALICE data-taking. Data Centre environmental conditions remained stable during the last more than two years, in particular during starting and stopping runs, which include significant changes in IT load. Several unforeseen external circumstances lead to increasing demands for the Online Offline System (O2). Higher data rates than anticipated required network performance to exceed the initial design specifications, for the throughput between FLPs and EPNs. In particular, the high throughput from an internal EPN InfiniBand network towards the storage Ethernet network was one of the challenges to overcome.
A central concern in genetics is to identify mechanisms of transcriptional regulation. The aim is to unravel the mapping between the DNA sequence and gene expression. However, it turned out that this is extremely complex. Gene regulation is highly cell type-specific and even moderate changes in gene ex- pression can have functional consequences.
Important contributors to gene regulation are transcription factors (TFs), that are able to directly interact with the DNA. Often, a first step in understanding the effect of a TF on the gene’s regulation is to identify the genomic regions a TF binds to. Therefore, one needs to be aware of the TF’s binding preferences, which are commonly summarized in TF binding motifs. Although for many TFs the binding motif is experimentally validated, there is still a large number of TFs where no binding motif is known. There exist many tools that link TF binding motifs to TFs. We developed the method Massif that improves the performance of such tools by incorporating a domain score that uses the DNA binding domain of the studied TF as additional information.
TF binding sites are often enriched in regulatory elements (REMs) such as promoters or enhancers, where the latter can be located megabases away from its target gene. However, to understand the regulation of a gene it is crucial to know where the REMs of a gene are located. We introduced the EpiRegio webserver that holds REMs associated to target genes predicted across many cell types and tissues using STITCHIT, a previously established method. Our publicly available webserver enables to query for REMs associated to genes (gene query) and REMs overlapping genomic regions (region query). We illus- trated the usefulness of EpiRegio by pointing to a TF that occurs enriched in the REMs of differential expressed genes in circPLOD2 depleted pericytes. Further, we highlighted genes, which are affected by CRISPR-Cas induced mutations in non-coding genomic regions using EpiRegio’s region query. Non-coding genetic variants within REMs may alter gene expression by modifying TF binding sites, which can lead to various kinds of traits or diseases. To understand the underlying molecular mechanisms, one aims to evaluate the effect of such genetic variations on TF binding sites. We developed an accurate and fast statistical approach, that can assess whether a single nucleotide polymorphism (SNP) is regulatory. Further, we combined this approach with epigenetic data and additional analyses in our Sneep workflow. For instance, it enables to identify TFs whose binding preferences are affected by the analyzed SNPs, which is illustrated on eQTL datasets for different cell types. Additionally, we used our Sneep workflow to highlight cardiovascular disease genes using regulatory SNPs and REM-gene interactions.
Overall, the described results allow a better understanding of REM-gene interactions and their interplay with TFs on gene regulation.
With the rise of digitalization and ubiquity of media use, both opportunities and challenges emerge for academic learning. One prevalent challenge is media multitasking, which can become distracting and hinder learning success. This thesis investigates two facets of this issue: the enhancement of data tracking, and the exploration of digital interventions that support self-control.
The first paper focuses on digital tracking of media use, as a comprehensive understanding of digital distractions requires careful data collection to avoid misinterpretations. The paper presents a tracking system where media use is linked to learning activities. An annotation dashboard enabled the enrichment of the log data with self-reports. The efficacy of this system was evaluated in a 14-day online course taken by 177 students, with results confirming the initial assumptions about media tracking.
The second paper tackles the recognition of whether a text was thoroughly read, an issue brought on by the tendency of students to skip lengthy and demanding texts. A method utilizing scroll data and time series classification algorithms is presented and tested, showing promising results for early recognition and intervention.
The third paper presents the results of a systematic literature review on the effectiveness of digital self-control tools in academic learning. The paper identifies gaps in existing research and outlines a roadmap for further research on self-control tools.
The fourth paper shares findings from a survey of 273 students, exploring the practical use and perceived helpfulness of DSCTs. The study highlights the challenge of balancing between too restrictive and too lenient DSCTs, particularly for platforms offering both learning content and entertainment. The results also show a special role of media use that is highly habitual.
The fifth paper of this work investigates facets of app-based habit building. In a study over 27 days, 106 school-aged children used the specially developed PROMPT-app. The children carried out one of three digital activities each day, each of which was supposed to promote a deeper or more superficial processing of plans. Significant differences regarding the processing of plans emerged between the three activities, and the results suggest that a child-friendly planning application needs to be personalized to be effective.
Overall, this work offers a comprehensive insight into the complexity and potentials of dealing with distracting media usage and shows ways for future research and interventions in this fascinating and ever more important field.
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.
Efficient algorithms for object recognition are crucial for the newly robotics and computer vision applications that demand real-time and on-line methods. Some examples are autonomous systems, navigating robots, autonomous driving. In this work, we focus on efficient semantic segmentation, which is the problem of labeling each pixel of an image with a semantic class.
Our aim is to speed-up all of the parts of the semantic segmentation pipeline. We also aim at delivering a labeling solution on a time budget, that can be decided on-the-fly. For this purpose, we analyze all the components of the semantic segmentation pipeline, and identify the computational bottleneck of each of them. The different components of the pipeline are over-segmenting the image with local regions, extracting features and classify the local regions, and the final inference of the image labeling with semantic classes. We focus on each of these steps.
First, we introduce a new superpixel algorithm to over-segment the image. Our superpixel method runs in real-time and can deliver a solution at any time budget. Then, for feature extraction, we focus on the framework that computes descriptors and encodes them, followed by a pooling step. We see that the encoding step is the bottleneck, for computational efficiency and performance. We present a novel assignment-based encoding formulation, that allows for the design of a new, very efficient, encoding. Finally, the image labeling output is obtained modeling the dependencies with a Conditional Random Field (CRF). In semantic image segmentation, the computational cost of instantiating the potentials is much higher than MAP inference. We introduce Active MAP inference to on-the-fly select a subset of potentials to be instantiated in the energy function, leaving the rest as unknown, and to estimate the MAP labeling from such incomplete energy function.
We perform experiments on all proposed methods for the different parts of the semantic segmentation pipeline. We show that our superpixel extraction achieves higher accuracy than state-of-the-art on standard superpixel benchmark, while it runs in real-time. We test our feature encoding on standard image classification and segmentation benchmarks, and we show that our method achieves competitive results with the state-of-the-art, and requires less time and memory. Finally, results for semantic segmentation benchmark show that Active MAP inference achieves similar levels of accuracy but with major efficiency gains.
Multi-view microscopy techniques are used to increase the resolution along the optical axis for 3D imaging. Without this, the resolution is insufficient to resolve subcellular events. In addition, parts of the images of opaque specimens are often highly degraded or masked. Both problems motivate scientists to record the same specimen from multiple directions. The images, then have to be digitally fused into a single high-quality image. Selective-plane illumination microscopy has proven to be a powerful imaging technique due to its unsurpassed acquisition speed and gentle optical sectioning. However, even in the case of multi view imaging techniques that illuminate and image the sample from multiple directions, light scattering inside tissues often severely impairs image contrast.
Here we show that for c-elegans embryos multi view registration can be achieved based on segmented nuclei. However, segmentation of nuclei in high density distribution like c-elegans embryo is challenging. We propose a method which uses 3D Mexican hat filter for preprocessing and 3D Gaussian curvature for the post-processing step to separate nuclei. We used this method successfully on 3 data sets of c-elegans embryos in 3 different views. The result of segmentation outperforms previous methods. Moreover, we provide a simple GUI for manual correction and adjusting the parameters for different data.
We then proposed a method that combines point and voxel registration for an accurate multi view reg- istration of c-elegans embryo, which does not need any special experimental preparation. We demonstrate the performance of our approach on data acquired from fixed embryos of c-elegans worms. This multi step approach is successfully evaluated by comparison to different methods and also by using synthetic data. The proposed method could overcome the typically low resolution along the optical axis and enable stitching to- gether the different parts of the embryo available through the different views. A tool for running the code and analyzing the results is developed.
In the last two decades, our understanding of human gene regulation has improved tremendously. There are plentiful computational methods which focus on integrative data analysis of humans, and model organisms, like mouse and drosophila. However, these tools are not directly employable by researchers working on non-model organisms to answer fundamental biological, and evolutionary questions. We aimed to develop new tools, and adapt existing software for the analysis of transcriptomic and epigenomic data of one such non-model organism, Paramecium tetraurelia, an unicellular eukaryote. Paramecium contains two diploid (2n) germline micronuclei (MIC) and a polyploid (800n) somatic macronuclei (MAC). The transcriptomic and epigenomic regulatory landscape of the MAC genome, which has 80% protein-coding genes and short intergenic regions, is poorly understood.
We developed a generic automated eukaryotic short interfering RNA (siRNA) analysis tool, called RAPID. Our tool captures diverse siRNA characteristics from small RNA sequencing data and provides easily navigable visualisations. We also introduced a normalisation technique to facilitate comparison of multiple siRNA-based gene knockdown studies. Further, we developed a pipeline to characterise novel genome-wide endogenous short interfering RNAs (endo-siRNAs). In contrary to many organisms, we found that the endo-siRNAs are not acting in cis, to silence their parent mRNA. We also predicted phasing of siRNAs, which are regulated by the RNA interference (RNAi) pathway.
Further, using RAPID, we investigated the aberrations of endo-siRNAs, and their respective transcriptomic alterations caused by an RNAi pathway triggered by feeding small RNAs against a target gene. We find that the small RNA transcriptome is altered, even if a gene unrelated to RNAi pathway is targeted. This is important in the context of investigations of genetically modified organisms (GMOs). We suggest that future studies need to distinguish transcriptomic changes caused by RNAi inducing techniques and actual regulatory changes.
Subsequently, we adapted existing epigenomics analysis tools to conduct the first comprehensive epigenomic characterisation of nucleosome positioning and histone modifications of the Paramecium MAC. We identified well positioned nucleosomes shifted downstream of the transcription start site. GC content seems to dictate, in cis, the positioning of nucleosomes, histone marks (H3K4me3, H3K9ac, and H3K27me3), and Pol II in the AT-rich Paramecium genome. We employed a chromatin state segmentation approach, on nucleosomes and histone marks, which revealed genes with active, repressive, and bivalent chromatin states. Further, we constructed a regulatory association network of all the aforementioned data, using the sparse partial correlation network technique. Our analysis revealed subsets of genes, whose expression is positively associated with H3K27me3, different to the otherwise reported negative association with gene expression in many other organisms.
Further, we developed a Random Forests classifier to predict gene expression using genic (gene length, intron frequency, etc.) and epigenetic features. Our model has a test performance (PR-AUC) of 0.83. Upon evaluating different feature sets, we found that genic features are as predictive, of gene expression, as the epigenetic features. We used Shapley local feature explanation values, to suggest that high H3K4me3, high intron frequency, low gene length, high sRNA, and high GC content are the most important elements for determining gene expression status.
In this thesis, we developed novel tools, and employed several bioinformatics and machine learning methods to characterise the regulatory landscape of the Paramecium’s (epi)genome.
High-energy physics experiments aim to deepen our understanding of the fundamental structure of matter and the governing forces. One of the most challenging aspects of the design of new experiments is data management and event selection. The search for increasingly rare and intricate physics events asks for high-statistics measurements and sophisticated event analysis. With progressively complex event signatures, traditional hardware-based trigger systems reach the limits of realizable latency and complexity. The Compressed Baryonic Matter experiment (CBM) employs a novel approach for data readout and event selection to address these challenges. Self-triggered, free-streaming detectors push all data to a central compute cluster, called First-level Event Selector (FLES), for software-based event analysis and selection. While this concept solves many issues present in classical architectures, it also sets new challenges for the design of the detector readout systems and online event selection.
This thesis presents an efficient solution to the data management challenges presented by self-triggered, free-streaming particle detectors. The FLES must receive asynchronously streamed data from a heterogeneous detector setup at rates of up to 1 TB/s. The real-time processing environment implies that all components have to deliver high performance and reliability to record as much valuable data as possible. The thesis introduces a time-based data model to partition the input streams into containers of fixed length in experiment time for efficient data management. These containers provide all necessary metadata to enable generic, detector-subsystem-agnostic data distribution across the entire cluster. An analysis shows that the introduced data overhead is well below 1 % for a wide range of system parameters.
Furthermore, a concept and the implementation of a detector data input interface for the CBM FLES, optimized for resource-efficient data transport, are presented. The central element of the architecture is an FPGA-based PCIe extension card for the FLES entry nodes. The hardware designs developed in the thesis enable interfacing with a diverse set of detector systems. A custom, high-throughput DMA design structures data in a way that enables low-overhead access and efficient software processing. The ability to share the host DMA buffers with other devices, such as an InfiniBand HCA, allows for true zero-copy data distribution between the cluster nodes. The discussed FLES input interface is fully implemented and has already proven its reliability in production operation in various physics experiments.
In this dissertation the formal abstraction and verification of analog circuit is examined. An approach is introduced that automatically abstracts a transistor level circuit with full Spice accuracy into a hybrid automaton (HA) in various output languages. The generated behavioral model exhibits a significant simulation speed-up compared to the original netlist, while maintaining an acceptable accuracy, and can be therefore used in various verification and validation routines. On top of that, the generated models can be formally verified against their Spice netlists, making the obtained models correct by construction.
The generated abstract models can be extended to enclose modeling as well as technology dependent parameter variations with little over approximations. As these models enclose the various behaviors of the sampled netlists, the obtained models are of significant importance as they can replace several simulations with just a single reachability analysis or symbolic simulation. Moreover, these models can be as well be used in different verification routines as demonstrated in this dissertation.
As the obtained models are described by HAs with linear behaviors in the locations, the abstract models can be as well compositionally linked, allowing thereby the abstraction of complex analog circuits.
Depending on the specified modeling settings, including for example the number of locations of the HA and the description of the system behavior, the accuracy, speedup, and various additional properties of the HA can be influenced. This is examined in detail in this dissertation. The underlying abstraction process is first covered in detail. Several extensions are then handled including the modeling of the HAs with parameter variations. The obtained models are then verified using various verification methodologies. The accuracy and speed-up of the abstraction methodology is finally evaluated on several transistor level circuits ranging from simple operational amplifiers up to a complex circuits.
Human readers have the ability to infer knowledge from text, even if that particular information is not explicitly stated. In this thesis, we address the phenomena of text-level implicit information and outline novel automated methods for its recovery.
The main focus of this work is on two types of unexpressed content that arises between sentences (implicit discourse relations) and within sentences (implicit semantic roles).
Traditional approaches mostly rely on costly rich linguistic features, e.g., sentiment or frame-based lexicons, and require heuristics or manual feature engineering.
As an improvement, we propose a collection of generic resource-lean methods, implemented in the form of statistical background knowledge or by means of neural architectures.
Our models are largely language-independent and produce state-of-the-art performance, e.g., in the classification of Chinese implicit discourse relations, or the detection of locally covert predicative arguments in free texts.
In novel experiments, we quantitatively demonstrate that both types of implicit information are mutually dependent insofar as, for instance, some implicit roles directly correlate with implicit discourse relations of similar properties.
We show that implicit information processing further benefits downstream applications and demonstrate its applicability to the higher-level task of narrative story understanding.
In the conclusion of the dissertation, we argue for the need of implicit information processing in order to realize the goal of true natural language understanding.