Refine
Year of publication
Document Type
- Preprint (759)
- Article (402)
- Working Paper (119)
- Doctoral Thesis (93)
- Diploma Thesis (47)
- Conference Proceeding (41)
- Book (37)
- Bachelor Thesis (36)
- diplomthesis (28)
- Report (25)
Has Fulltext
- yes (1619)
Is part of the Bibliography
- no (1619)
Keywords
Institute
- Informatik (1619) (remove)
Electrocardiograms (ECG) record the heart activity and are the most common and reliable method to detect cardiac arrhythmias, such as atrial fibrillation (AFib). Lately, many commercially available devices such as smartwatches are offering ECG monitoring. Therefore, there is increasing demand for designing deep learning models with the perspective to be physically implemented on these small portable devices with limited energy supply. In this paper, a workflow for the design of small, energy-efficient recurrent convolutional neural network (RCNN) architecture for AFib detection is proposed. However, the approach can be well generalized to every type of long time series. In contrast to previous studies, that demand thousands of additional network neurons and millions of extra model parameters, the logical steps for the generation of a CNN with only 114 trainable parameters are described. The model consists of a small segmented CNN in combination with an optimal energy classifier. The architectural decisions are made by using the energy consumption as a metric in an equally important way as the accuracy. The optimization steps are focused on the software which can be embedded afterwards on a physical chip. Finally, a comparison with some previous relevant studies suggests that the widely used huge CNNs for similar tasks are mostly redundant and unessentially computationally expensive.
Electrocardiograms (ECG) record the heart activity and are the most common and reliable method to detect cardiac arrhythmias, such as atrial fibrillation (AFib). Lately, many commercially available devices such as smartwatches are offering ECG monitoring. Therefore, there is increasing demand for designing deep learning models with the perspective to be physically implemented on these small portable devices with limited energy supply. In this paper, a workflow for the design of small, energy-efficient recurrent convolutional neural network (RCNN) architecture for AFib detection is proposed. However, the approach can be well generalized to every type of long time series. In contrast to previous studies, that demand thousands of additional network neurons and millions of extra model parameters, the logical steps for the generation of a CNN with only 114 trainable parameters are described. The model consists of a small segmented CNN in combination with an optimal energy classifier. The architectural decisions are made by using the energy consumption as a metric in an equally important way as the accuracy. The optimisation steps are focused on the software which can be embedded afterwards on a physical chip. Finally, a comparison with some previous relevant studies suggests that the widely used huge CNNs for similar tasks are mostly redundant and unessentially computationally expensive.
Non-coding variations located within regulatory elements may alter gene expression by modifying Transcription Factor (TF) binding sites and thereby lead to functional consequences like various traits or diseases. To understand these molecular mechanisms, different TF models are being used to assess the effect of DNA sequence variations, such as Single Nucleotide Polymorphisms (SNPs). However, few statistical approaches exist to compute statistical significance of results but they often are slow for large sets of SNPs, such as data obtained from a genome-wide association study (GWAS) or allele-specific analysis of chromatin data.
Results We investigate the distribution of maximal differential TF binding scores for general computational models that assess TF binding. We find that a modified Laplace distribution can adequately approximate the empirical distributions. A benchmark on in vitro and in vivo data sets showed that our new approach improves on an existing method in terms of performance and speed. In applications on large sets of eQTL and GWAS SNPs we could illustrate the usefulness of the novel statistic to highlight cell type specific regulators and TF target genes.
Conclusions Our approach allows the evaluation of DNA changes that induce differential TF binding in a fast and accurate manner, permitting computations on large mutation data sets. An implementation of the novel approach is freely available at https://github.com/SchulzLab/SNEEP.
A key competence for open-ended learning is the formation of increasingly abstract representations useful for driving complex behavior. Abstract representations ignore specific details and facilitate generalization. Here we consider the learning of abstract representations in a multi-modal setting with two or more input modalities. We treat the problem as a lossy compression problem and show that generic lossy compression of multimodal sensory input naturally extracts abstract representations that tend to strip away modalitiy specific details and preferentially retain information that is shared across the different modalities. Furthermore, we propose an architecture to learn abstract representations by identifying and retaining only the information that is shared across multiple modalities while discarding any modality specific information.
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.
Human functional brain connectivity can be temporally decomposed into states of high and low cofluctuation, defined as coactivation of brain regions over time. Rare states of particularly high cofluctuation have been shown to reflect fundamentals of intrinsic functional network architecture and to be highly subject-specific. However, it is unclear whether such network-defining states also contribute to individual variations in cognitive abilities – which strongly rely on the interactions among distributed brain regions. By introducing CMEP, a new eigenvector-based prediction framework, we show that as few as 16 temporally separated time frames (< 1.5% of 10min resting-state fMRI) can significantly predict individual differences in intelligence (N = 263, p < .001). Against previous expectations, individual’s network-defining time frames of particularly high cofluctuation do not predict intelligence. Multiple functional brain networks contribute to the prediction, and all results replicate in an independent sample (N = 831). Our results suggest that although fundamentals of person-specific functional connectomes can be derived from few time frames of highest connectivity, temporally distributed information is necessary to extract information about cognitive abilities. This information is not restricted to specific connectivity states, like network-defining high-cofluctuation states, but rather reflected across the entire length of the brain connectivity time series.
The cortical networks that underlie behavior exhibit an orderly functional organization at local and global scales, which is readily evident in the visual cortex of carnivores and primates1-6. Here, neighboring columns of neurons represent the full range of stimulus orientations and contribute to distributed networks spanning several millimeters2,7-11. However, the principles governing functional interactions that bridge this fine-scale functional architecture and distant network elements are unclear, and the emergence of these network interactions during development remains unexplored. Here, by using in vivo wide-field and 2-photon calcium imaging of spontaneous activity patterns in mature ferret visual cortex, we find widespread and specific modular correlation patterns that accurately predict the local structure of visually-evoked orientation columns from the spontaneous activity of neurons that lie several millimeters away. The large-scale networks revealed by correlated spontaneous activity show abrupt ‘fractures’ in continuity that are in tight register with evoked orientation pinwheels. Chronic in vivo imaging demonstrates that these large-scale modular correlation patterns and fractures are already present at early stages of cortical development and predictive of the mature network structure. Silencing feed-forward drive through either retinal or thalamic blockade does not affect network structure suggesting a cortical origin for this large-scale correlated activity, despite the immaturity of long-range horizontal network connections in the early cortex. Using a circuit model containing only local connections, we demonstrate that such a circuit is sufficient to generate large-scale correlated activity, while also producing correlated networks showing strong fractures, a reduced dimensionality, and an elongated local correlation structure, all in close agreement with our empirical data. These results demonstrate the precise local and global organization of cortical networks revealed through correlated spontaneous activity and suggest that local connections in early cortical circuits may generate structured long-range network correlations that underlie the subsequent formation of visually-evoked distributed functional networks.
The fundamental structure of cortical networks arises early in development prior to the onset of sensory experience. However, how endogenously generated networks respond to the onset of sensory experience, and how they form mature sensory representations with experience remains unclear. Here we examine this ‘nature-nurture transform’ using in vivo calcium imaging in ferret visual cortex. At eye-opening, visual stimulation evokes robust patterns of cortical activity that are highly variable within and across trials, severely limiting stimulus discriminability. Initial evoked responses are distinct from spontaneous activity of the endogenous network. Visual experience drives the development of low-dimensional, reliable representations aligned with spontaneous activity. A computational model shows that alignment of novel visual inputs and recurrent cortical networks can account for the emergence of reliable visual representations.
The fundamental structure of cortical networks arises early in development prior to the onset of sensory experience. However, how endogenously generated networks respond to the onset of sensory experience, and how they form mature sensory representations with experience remains unclear. Here we examine this "nature-nurture transform" using in vivo calcium imaging in ferret visual cortex. At eye-opening, visual stimulation evokes robust patterns of cortical activity that are highly variable within and across trials, severely limiting stimulus discriminability. Initial evoked responses are distinct from spontaneous activity of the endogenous network. Visual experience drives the development of low-dimensional, reliable representations aligned with spontaneous activity. A computational model shows that alignment of novel visual inputs and recurrent cortical networks can account for the emergence of reliable visual representations.
The human brain achieves visual object recognition through multiple stages of linear and nonlinear transformations operating at a millisecond scale. To predict and explain these rapid transformations, computational neuroscientists employ machine learning modeling techniques. However, state-of-the-art models require massive amounts of data to properly train, and to the present day there is a lack of vast brain datasets which extensively sample the temporal dynamics of visual object recognition. Here we collected a large and rich dataset of high temporal resolution EEG responses to images of objects on a natural background. This dataset includes 10 participants, each with 82,160 trials spanning 16,740 image conditions. Through computational modeling we established the quality of this dataset in five ways. First, we trained linearizing encoding models that successfully synthesized the EEG responses to arbitrary images. Second, we correctly identified the recorded EEG data image conditions in a zero-shot fashion, using EEG synthesized responses to hundreds of thousands of candidate image conditions. Third, we show that both the high number of conditions as well as the trial repetitions of the EEG dataset contribute to the trained models’ prediction accuracy. Fourth, we built encoding models whose predictions well generalize to novel participants. Fifth, we demonstrate full end-to-end training of randomly initialized DNNs that output EEG responses for arbitrary input images. We release this dataset as a tool to foster research in visual neuroscience and computer vision.
The human brain achieves visual object recognition through multiple stages of nonlinear transformations operating at a millisecond scale. To predict and explain these rapid transformations, computational neuroscientists employ machine learning modeling techniques. However, state-of-the-art models require massive amounts of data to properly train, and to the present day there is a lack of vast brain datasets which extensively sample the temporal dynamics of visual object recognition. Here we collected a large and rich dataset of high temporal resolution EEG responses to images of objects on a natural background. This dataset includes 10 participants, each with 82,160 trials spanning 16,740 image conditions. Through computational modeling we established the quality of this dataset in five ways. First, we trained linearizing encoding models that successfully synthesized the EEG responses to arbitrary images. Second, we correctly identified the recorded EEG data image conditions in a zero-shot fashion, using EEG synthesized responses to hundreds of thousands of candidate image conditions. Third, we show that both the high number of conditions as well as the trial repetitions of the EEG dataset contribute to the trained models’ prediction accuracy. Fourth, we built encoding models whose predictions well generalize to novel participants. Fifth, we demonstrate full end-to-end training of randomly initialized DNNs that output M/EEG responses for arbitrary input images. We release this dataset as a tool to foster research in visual neuroscience and computer vision.
Human lymph nodes play a central part of immune defense against infection agents and tumor cells. Lymphoid follicles are compartments of the lymph node which are spherical, mainly filled with B cells. B cells are cellular components of the adaptive immune systems. In the course of a specific immune response, lymphoid follicles pass different morphological differentiation stages. The morphology and the spatial distribution of lymphoid follicles can be sometimes associated to a particular causative agent and development stage of a disease. We report our new approach for the automatic detection of follicular regions in histological whole slide images of tissue sections immuno-stained with actin. The method is divided in two phases: (1) shock filter-based detection of transition points and (2) segmentation of follicular regions. Follicular regions in 10 whole slide images were manually annotated by visual inspection, and sample surveys were conducted by an expert pathologist. The results of our method were validated by comparing with the manual annotation. On average, we could achieve a Zijbendos similarity index of 0.71, with a standard deviation of 0.07.
Grasping the meaning of everyday visual events is a fundamental feat of human intelligence that hinges on diverse neural processes ranging from vision to higher-level cognition. Deciphering the neural basis of visual event understanding requires rich, extensive, and appropriately designed experimental data. However, this type of data is hitherto missing. To fill this gap, we introduce the BOLD Moments Dataset (BMD), a large dataset of whole-brain fMRI responses to over 1,000 short (3s) naturalistic video clips and accompanying metadata. We show visual events interface with an array of processes, extending even to memory, and we reveal a match in hierarchical processing between brains and video-computable deep neural networks. Furthermore, we showcase that BMD successfully captures temporal dynamics of visual events at second resolution. BMD thus establishes a critical groundwork for investigations of the neural basis of visual event understanding.
Visual scene perception is mediated by a set of cortical regions that respond preferentially to images of scenes, including the occipital place area (OPA) and parahippocampal place area (PPA). However, the differential contribution of OPA and PPA to scene perception remains an open research question. In this study, we take a deep neural network (DNN)-based computational approach to investigate the differences in OPA and PPA function. In a first step we search for a computational model that predicts fMRI responses to scenes in OPA and PPA well. We find that DNNs trained to predict scene components (e.g., wall, ceiling, floor) explain higher variance uniquely in OPA and PPA than a DNN trained to predict scene category (e.g., bathroom, kitchen, office). This result is robust across several DNN architectures. On this basis, we then determine whether particular scene components predicted by DNNs differentially account for unique variance in OPA and PPA. We find that variance in OPA responses uniquely explained by the navigation-related floor component is higher compared to the variance explained by the wall and ceiling components. In contrast, PPA responses are better explained by the combination of wall and floor, that is scene components that together contain the structure and texture of the scene. This differential sensitivity to scene components suggests differential functions of OPA and PPA in scene processing. Moreover, our results further highlight the potential of the proposed computational approach as a general tool in the investigation of the neural basis of human scene perception.
The human visual cortex enables visual perception through a cascade of hierarchical computations in cortical regions with distinct functionalities. Here, we introduce an AI-driven approach to discover the functional mapping of the visual cortex. We related human brain responses to scene images measured with functional MRI (fMRI) systematically to a diverse set of deep neural networks (DNNs) optimized to perform different scene perception tasks. We found a structured mapping between DNN tasks and brain regions along the ventral and dorsal visual streams. Low-level visual tasks mapped onto early brain regions, 3-dimensional scene perception tasks mapped onto the dorsal stream, and semantic tasks mapped onto the ventral stream. This mapping was of high fidelity, with more than 60% of the explainable variance in nine key regions being explained. Together, our results provide a novel functional mapping of the human visual cortex and demonstrate the power of the computational approach.
Dendritic spines are considered a morphological proxy for excitatory synapses, rendering them a target of many different lines of research. Over recent years, it has become possible to image simultaneously large numbers of dendritic spines in 3D volumes of neural tissue. In contrast, currently no automated method for spine detection exists that comes close to the detection performance reached by human experts. However, exploiting such datasets requires new tools for the fully automated detection and analysis of large numbers of spines. Here, we developed an efficient analysis pipeline to detect large numbers of dendritic spines in volumetric fluorescence imaging data. The core of our pipeline is a deep convolutional neural network, which was pretrained on a general-purpose image library, and then optimized on the spine detection task. This transfer learning approach is data efficient while achieving a high detection precision. To train and validate the model we generated a labelled dataset using five human expert annotators to account for the variability in human spine detection. The pipeline enables fully automated dendritic spine detection and reaches a near human-level detection performance. Our method for spine detection is fast, accurate and robust, and thus well suited for large-scale datasets with thousands of spines. The code is easily applicable to new datasets, achieving high detection performance, even without any retraining or adjustment of model parameters.
Active efficient coding explains the development of binocular vision and its failure in amblyopia
(2020)
The development of vision during the first months of life is an active process that comprises the learning of appropriate neural representations and the learning of accurate eye movements. While it has long been suspected that the two learning processes are coupled, there is still no widely accepted theoretical framework describing this joint development. Here, we propose a computational model of the development of active binocular vision to fill this gap. The model is based on a formulation of the active efficient coding theory, which proposes that eye movements as well as stimulus encoding are jointly adapted to maximize the overall coding efficiency. Under healthy conditions, the model self-calibrates to perform accurate vergence and accommodation eye movements. It exploits disparity cues to deduce the direction of defocus, which leads to coordinated vergence and accommodation responses. In a simulated anisometropic case, where the refraction power of the two eyes differs, an amblyopia-like state develops in which the foveal region of one eye is suppressed due to inputs from the other eye. After correcting for refractive errors, the model can only reach healthy performance levels if receptive fields are still plastic, in line with findings on a critical period for binocular vision development. Overall, our model offers a unifying conceptual framework for understanding the development of binocular vision.
Future operation of the CBM detector requires ultra-fast analysis of the continuous stream of data from all subdetector systems. Determining the inter-system time shifts among individual detector systems in the existing prototype experiment mCBM is an essential step for data processing and in particular for stable data taking. Based on the input of raw measurements from all detector systems, the corresponding time correlations can be obtained at digital level by evaluating the differences in time stamps. If the relevant systems are stable during data taking and sufficient digital measurements are available, the distribution of time differences should display a clear peak. Up to now, the outcome of the processed time differences is stored in histograms and the maximum peak is considered, after the evaluation of all timeslices of a run leading to significant run times. The results presented here demonstrate the stability of the synchronicity of mCBM systems. Furthermore it is illustrated that relatively small amounts of raw measurements are sufficient to evaluate corresponding time correlations among individual mCBM detectors, thus enabling fast online monitoring of them in future online data processing.
Im Fachbereich der Computerlinguistik ist die automatische Generierung von Szenen aus, in natürlicher Sprache verfassten, Text seit bereits vielen Jahrzehnten ein wichtiger Bestandteil der Forschung, welche in der "Kunst", "Lehre" und "Robotik" Verwendung finden. Mit Hilfe von neuen Technologien im Bereich der Künstlichen Intelligenzen (KI), werden neue Entwicklungen möglich, welche diese Generierungen vereinfachen, allerdings auch undurchsichtige interne vom Modell getroffene Entscheidungen fördern.
Ziel der vorgeschlagenen Lösung „ARES: Annotation von Relationen und Eigenschaften zur Szenengenerierung“ ist es, ein modulares System zu entwerfen, wobei einzelne Prozesse für den Benutzer verständlich bleiben. Außerdem sollen Möglichkeiten geboten werden, neue Entitäten und Relationen, welche über die Textanalyse bereitgestellt werden, auch in die Szenengenerierung im dreidimensionalen Raum einzupflegen, ohne dass hierfür Code zwingend notwendig wird.
Der Fokus liegt auf der syntaktisch korrekten Darstellung der Elemente im Raum. Dagegen lässt sich die semantische Korrektheit durch weitere manuelle Anpassungen, welche für spätere Generierungen gespeichert werden erhöhen. Letztlich soll die Menge der zur Darstellung benötigten Annotationen möglichst gering bleiben und neue szenenbezogene Annotationen durch die implementierten Annotationstools hinzugefügt werden.