@article{SchloerRingHotho2020,
  author    = {Schl{\"o}r, Daniel and Ring, Markus and Hotho, Andreas},
  title     = {iNALU: Improved Neural Arithmetic Logic Unit},
  series = {Frontiers in Artificial Intelligence},
  volume    = {3},
  journal   = {Frontiers in Artificial Intelligence},
  issn      = {2624-8212},
  doi       = {10.3389/frai.2020.00071},
  url       = {http://nbn-resolving.de/urn:nbn:de:bvb:20-opus-212301},
  year      = {2020},
  abstract  = {Neural networks have to capture mathematical relationships in order to learn various tasks. They approximate these relations implicitly and therefore often do not generalize well. The recently proposed Neural Arithmetic Logic Unit (NALU) is a novel neural architecture which is able to explicitly represent the mathematical relationships by the units of the network to learn operations such as summation, subtraction or multiplication. Although NALUs have been shown to perform well on various downstream tasks, an in-depth analysis reveals practical shortcomings by design, such as the inability to multiply or divide negative input values or training stability issues for deeper networks. We address these issues and propose an improved model architecture. We evaluate our model empirically in various settings from learning basic arithmetic operations to more complex functions. Our experiments indicate that our model solves stability issues and outperforms the original NALU model in means of arithmetic precision and convergence.},
  language  = {en}
}
@article{HoeserBachoferKuenzer2020,
  author    = {Hoeser, Thorsten and Bachofer, Felix and Kuenzer, Claudia},
  title     = {Object detection and image segmentation with deep learning on Earth Observation data: a review — part II: applications},
  series = {Remote Sensing},
  volume    = {12},
  journal   = {Remote Sensing},
  number    = {18},
  issn      = {2072-4292},
  doi       = {10.3390/rs12183053},
  url       = {http://nbn-resolving.de/urn:nbn:de:bvb:20-opus-213152},
  year      = {2020},
  abstract  = {In Earth observation (EO), large-scale land-surface dynamics are traditionally analyzed by investigating aggregated classes. The increase in data with a very high spatial resolution enables investigations on a fine-grained feature level which can help us to better understand the dynamics of land surfaces by taking object dynamics into account. To extract fine-grained features and objects, the most popular deep-learning model for image analysis is commonly used: the convolutional neural network (CNN). In this review, we provide a comprehensive overview of the impact of deep learning on EO applications by reviewing 429 studies on image segmentation and object detection with CNNs. We extensively examine the spatial distribution of study sites, employed sensors, used datasets and CNN architectures, and give a thorough overview of applications in EO which used CNNs. Our main finding is that CNNs are in an advanced transition phase from computer vision to EO. Upon this, we argue that in the near future, investigations which analyze object dynamics with CNNs will have a significant impact on EO research. With a focus on EO applications in this Part II, we complete the methodological review provided in Part I.},
  language  = {en}
}
@article{HoeserKuenzer2020,
  author    = {Hoeser, Thorsten and Kuenzer, Claudia},
  title     = {Object detection and image segmentation with deep learning on Earth observation data: a review-part I: evolution and recent trends},
  series = {Remote Sensing},
  volume    = {12},
  journal   = {Remote Sensing},
  number    = {10},
  issn      = {2072-4292},
  doi       = {10.3390/rs12101667},
  url       = {http://nbn-resolving.de/urn:nbn:de:bvb:20-opus-205918},
  year      = {2020},
  abstract  = {Deep learning (DL) has great influence on large parts of science and increasingly established itself as an adaptive method for new challenges in the field of Earth observation (EO). Nevertheless, the entry barriers for EO researchers are high due to the dense and rapidly developing field mainly driven by advances in computer vision (CV). To lower the barriers for researchers in EO, this review gives an overview of the evolution of DL with a focus on image segmentation and object detection in convolutional neural networks (CNN). The survey starts in 2012, when a CNN set new standards in image recognition, and lasts until late 2019. Thereby, we highlight the connections between the most important CNN architectures and cornerstones coming from CV in order to alleviate the evaluation of modern DL models. Furthermore, we briefly outline the evolution of the most popular DL frameworks and provide a summary of datasets in EO. By discussing well performing DL architectures on these datasets as well as reflecting on advances made in CV and their impact on future research in EO, we narrow the gap between the reviewed, theoretical concepts from CV and practical application in EO.},
  language  = {en}
}
@article{MostosiSchindelinKollmannsbergeretal.2020,
  author    = {Mostosi, Philipp and Schindelin, Hermann and Kollmannsberger, Philip and Thorn, Andrea},
  title     = {Haruspex: A Neural Network for the Automatic Identification of Oligonucleotides and Protein Secondary Structure in Cryo-Electron Microscopy Maps},
  series = {Angewandte Chemie International Edition},
  volume    = {59},
  journal   = {Angewandte Chemie International Edition},
  number    = {35},
  doi       = {10.1002/anie.202000421},
  url       = {http://nbn-resolving.de/urn:nbn:de:bvb:20-opus-214763},
  pages     = {14788 -- 14795},
  year      = {2020},
  abstract  = {In recent years, three-dimensional density maps reconstructed from single particle images obtained by electron cryo-microscopy (cryo-EM) have reached unprecedented resolution. However, map interpretation can be challenging, in particular if the constituting structures require de-novo model building or are very mobile. Herein, we demonstrate the potential of convolutional neural networks for the annotation of cryo-EM maps: our network Haruspex has been trained on a carefully curated set of 293 experimentally derived reconstruction maps to automatically annotate RNA/DNA as well as protein secondary structure elements. It can be straightforwardly applied to newly reconstructed maps in order to support domain placement or as a starting point for main-chain placement. Due to its high recall and precision rates of 95.1 \% and 80.3 \%, respectively, on an independent test set of 122 maps, it can also be used for validation during model building. The trained network will be available as part of the CCP-EM suite.},
  language  = {en}
}
@article{AnkenbrandLohrSchloetelburgetal.2021,
  author    = {Ankenbrand, Markus Johannes and Lohr, David and Schl{\"o}telburg, Wiebke and Reiter, Theresa and Wech, Tobias and Schreiber, Laura Maria},
  title     = {Deep learning-based cardiac cine segmentation: Transfer learning application to 7T ultrahigh-field MRI},
  series = {Magnetic Resonance in Medicine},
  volume    = {86},
  journal   = {Magnetic Resonance in Medicine},
  number    = {4},
  doi       = {10.1002/mrm.28822},
  url       = {http://nbn-resolving.de/urn:nbn:de:bvb:20-opus-257604},
  pages     = {2179-2191},
  year      = {2021},
  abstract  = {Purpose Artificial neural networks show promising performance in automatic segmentation of cardiac MRI. However, training requires large amounts of annotated data and generalization to different vendors, field strengths, sequence parameters, and pathologies is limited. Transfer learning addresses this challenge, but specific recommendations regarding type and amount of data required is lacking. In this study, we assess data requirements for transfer learning to experimental cardiac MRI at 7T where the segmentation task can be challenging. In addition, we provide guidelines, tools, and annotated data to enable transfer learning approaches by other researchers and clinicians. Methods A publicly available segmentation model was used to annotate a publicly available data set. This labeled data set was subsequently used to train a neural network for segmentation of left ventricle and myocardium in cardiac cine MRI. The network is used as starting point for transfer learning to 7T cine data of healthy volunteers (n = 22; 7873 images) by updating the pre-trained weights. Structured and random data subsets of different sizes were used to systematically assess data requirements for successful transfer learning. Results Inconsistencies in the publically available data set were corrected, labels created, and a neural network trained. On 7T cardiac cine images the model pre-trained on public imaging data, acquired at 1.5T and 3T, achieved DICE\(_{LV}\) = 0.835 and DICE\(_{MY}\) = 0.670. Transfer learning using 7T cine data and ImageNet weight initialization improved model performance to DICE\(_{LV}\) = 0.900 and DICE\(_{MY}\) = 0.791. Using only end-systolic and end-diastolic images reduced training data by 90\%, with no negative impact on segmentation performance (DICE\(_{LV}\) = 0.908, DICE\(_{MY}\) = 0.805). Conclusions This work demonstrates and quantifies the benefits of transfer learning for cardiac cine image segmentation. We provide practical guidelines for researchers planning transfer learning projects in cardiac MRI and make data, models, and code publicly available.},
  language  = {en}
}
@article{AnkenbrandShainbergHocketal.2021,
  author    = {Ankenbrand, Markus J. and Shainberg, Liliia and Hock, Michael and Lohr, David and Schreiber, Laura M.},
  title     = {Sensitivity analysis for interpretation of machine learning based segmentation models in cardiac MRI},
  series = {BMC Medical Imaging},
  volume    = {21},
  journal   = {BMC Medical Imaging},
  number    = {1},
  doi       = {10.1186/s12880-021-00551-1},
  url       = {http://nbn-resolving.de/urn:nbn:de:bvb:20-opus-259169},
  pages     = {27},
  year      = {2021},
  abstract  = {Background Image segmentation is a common task in medical imaging e.g., for volumetry analysis in cardiac MRI. Artificial neural networks are used to automate this task with performance similar to manual operators. However, this performance is only achieved in the narrow tasks networks are trained on. Performance drops dramatically when data characteristics differ from the training set properties. Moreover, neural networks are commonly considered black boxes, because it is hard to understand how they make decisions and why they fail. Therefore, it is also hard to predict whether they will generalize and work well with new data. Here we present a generic method for segmentation model interpretation. Sensitivity analysis is an approach where model input is modified in a controlled manner and the effect of these modifications on the model output is evaluated. This method yields insights into the sensitivity of the model to these alterations and therefore to the importance of certain features on segmentation performance. Results We present an open-source Python library (misas), that facilitates the use of sensitivity analysis with arbitrary data and models. We show that this method is a suitable approach to answer practical questions regarding use and functionality of segmentation models. We demonstrate this in two case studies on cardiac magnetic resonance imaging. The first case study explores the suitability of a published network for use on a public dataset the network has not been trained on. The second case study demonstrates how sensitivity analysis can be used to evaluate the robustness of a newly trained model. Conclusions Sensitivity analysis is a useful tool for deep learning developers as well as users such as clinicians. It extends their toolbox, enabling and improving interpretability of segmentation models. Enhancing our understanding of neural networks through sensitivity analysis also assists in decision making. Although demonstrated only on cardiac magnetic resonance images this approach and software are much more broadly applicable.},
  language  = {en}
}
@article{HepbasliGredyUllrichetal.2021,
  author    = {Hepbasli, Denis and Gredy, Sina and Ullrich, Melanie and Reigl, Amelie and Abeßer, Marco and Raabe, Thomas and Schuh, Kai},
  title     = {Genotype- and Age-Dependent Differences in Ultrasound Vocalizations of SPRED2 Mutant Mice Revealed by Machine Deep Learning},
  series = {Brain Sciences},
  volume    = {11},
  journal   = {Brain Sciences},
  number    = {10},
  issn      = {2076-3425},
  doi       = {10.3390/brainsci11101365},
  url       = {http://nbn-resolving.de/urn:nbn:de:bvb:20-opus-248525},
  year      = {2021},
  abstract  = {Vocalization is an important part of social communication, not only for humans but also for mice. Here, we show in a mouse model that functional deficiency of Sprouty-related EVH1 domain-containing 2 (SPRED2), a protein ubiquitously expressed in the brain, causes differences in social ultrasound vocalizations (USVs), using an uncomplicated and reliable experimental setting of a short meeting of two individuals. SPRED2 mutant mice show an OCD-like behaviour, accompanied by an increased release of stress hormones from the hypothalamic-pituitary-adrenal axis, both factors probably influencing USV usage. To determine genotype-related differences in USV usage, we analyzed call rate, subtype profile, and acoustic parameters (i.e., duration, bandwidth, and mean peak frequency) in young and old SPRED2-KO mice. We recorded USVs of interacting male and female mice, and analyzed the calls with the deep-learning DeepSqueak software, which was trained to recognize and categorize the emitted USVs. Our findings provide the first classification of SPRED2-KO vs. wild-type mouse USVs using neural networks and reveal significant differences in their development and use of calls. Our results show, first, that simple experimental settings in combination with deep learning are successful at identifying genotype-dependent USV usage and, second, that SPRED2 deficiency negatively affects the vocalization usage and social communication of mice.},
  language  = {en}
}
@article{PerniaAndradeWengerEspositoetal.2021,
  author    = {Pern{\´i}a-Andrade, Alejandro J. and Wenger, Nikolaus and Esposito, Maria S. and Tovote, Philip},
  title     = {Circuits for State-Dependent Modulation of Locomotion},
  series = {Frontiers in Human Neuroscience},
  volume    = {15},
  journal   = {Frontiers in Human Neuroscience},
  issn      = {1662-5161},
  doi       = {10.3389/fnhum.2021.745689},
  url       = {http://nbn-resolving.de/urn:nbn:de:bvb:20-opus-249995},
  year      = {2021},
  abstract  = {Brain-wide neural circuits enable bi- and quadrupeds to express adaptive locomotor behaviors in a context- and state-dependent manner, e.g., in response to threats or rewards. These behaviors include dynamic transitions between initiation, maintenance and termination of locomotion. Advances within the last decade have revealed an intricate coordination of these individual locomotion phases by complex interaction of multiple brain circuits. This review provides an overview of the neural basis of state-dependent modulation of locomotion initiation, maintenance and termination, with a focus on insights from circuit-centered studies in rodents. The reviewed evidence indicates that a brain-wide network involving excitatory circuit elements connecting cortex, midbrain and medullary areas appears to be the common substrate for the initiation of locomotion across different higher-order states. Specific network elements within motor cortex and the mesencephalic locomotor region drive the initial postural adjustment and the initiation of locomotion. Microcircuits of the basal ganglia, by implementing action-selection computations, trigger goal-directed locomotion. The initiation of locomotion is regulated by neuromodulatory circuits residing in the basal forebrain, the hypothalamus, and medullary regions such as locus coeruleus. The maintenance of locomotion requires the interaction of an even larger neuronal network involving motor, sensory and associative cortical elements, as well as defined circuits within the superior colliculus, the cerebellum, the periaqueductal gray, the mesencephalic locomotor region and the medullary reticular formation. Finally, locomotor arrest as an important component of defensive emotional states, such as acute anxiety, is mediated via a network of survival circuits involving hypothalamus, amygdala, periaqueductal gray and medullary premotor centers. By moving beyond the organizational principle of functional brain regions, this review promotes a circuit-centered perspective of locomotor regulation by higher-order states, and emphasizes the importance of individual network elements such as cell types and projection pathways. The realization that dysfunction within smaller, identifiable circuit elements can affect the larger network function supports more mechanistic and targeted therapeutic intervention in the treatment of motor network disorders.},
  language  = {en}
}
@article{SteiningerAbelZiegleretal.2023,
  author    = {Steininger, Michael and Abel, Daniel and Ziegler, Katrin and Krause, Anna and Paeth, Heiko and Hotho, Andreas},
  title     = {ConvMOS: climate model output statistics with deep learning},
  series = {Data Mining and Knowledge Discovery},
  volume    = {37},
  journal   = {Data Mining and Knowledge Discovery},
  number    = {1},
  issn      = {1384-5810},
  doi       = {10.1007/s10618-022-00877-6},
  url       = {http://nbn-resolving.de/urn:nbn:de:bvb:20-opus-324213},
  pages     = {136-166},
  year      = {2023},
  abstract  = {Climate models are the tool of choice for scientists researching climate change. Like all models they suffer from errors, particularly systematic and location-specific representation errors. One way to reduce these errors is model output statistics (MOS) where the model output is fitted to observational data with machine learning. In this work, we assess the use of convolutional Deep Learning climate MOS approaches and present the ConvMOS architecture which is specifically designed based on the observation that there are systematic and location-specific errors in the precipitation estimates of climate models. We apply ConvMOS models to the simulated precipitation of the regional climate model REMO, showing that a combination of per-location model parameters for reducing location-specific errors and global model parameters for reducing systematic errors is indeed beneficial for MOS performance. We find that ConvMOS models can reduce errors considerably and perform significantly better than three commonly used MOS approaches and plain ResNet and U-Net models in most cases. Our results show that non-linear MOS models underestimate the number of extreme precipitation events, which we alleviate by training models specialized towards extreme precipitation events with the imbalanced regression method DenseLoss. While we consider climate MOS, we argue that aspects of ConvMOS may also be beneficial in other domains with geospatial data, such as air pollution modeling or weather forecasts.},
  subject      = {Klima},
  language  = {en}
}