Aktuelle Publikationen

This paper gives a comprehensive overview over classic shallow and novel deep approaches to anomaly detection. We identify the common underlying principles and provide an empirical assessment of major existing methods that are enriched by the use of recent explainability techniques. We present specific worked-through examples together with practical advice and discuss open challenges.

This paper proposes a novel criterion for CNN pruning inspired by neural network interpretability: The most relevant units, i.e. weights or filters, are automatically found using their relevance scores obtained from concepts of explainable AI (XAI). By exploring this idea, we connect the lines of interpretability and model compression research.

Time-domain spectroscopy with terahertz frequencies typically requires complex and bulky systems. Here, the authors present an optoelectronics-based, frequency-domain terahertz sensing technique which offers competitive measurement performance in a much simpler system.

Externally modulated DFB lasers (EML) and vertically illuminated photodetectors are presented. Because of their excellent high-speed behavior and operation wavelength of 1310 nm, the devices are of interest for intra-data center communication. Since the EML and the photodetector chips are compatible with current systems, these devices are candidates for upgrading existing transceivers to higher baud rates. Therefore, a proof of concept for 100 GBaud data transmission with the presented components is demonstrated. Even without predistor! tion, the experiments show clearly open eye diagrams.

This paper proposes a novel, simple yet effective defense strategy for adversarial attacks on deep learning models. Our algorithm, called MALA for DEfense (MALADE), is applicable to any existing classifier, providing robust defense as well as off-manifold sample detection. In our experiments, MALADE exhibited state-of-the-art performance against various elaborate attacking strategies.

Inverse design is rapidly gaining popularity for automated design of photonic components. Two methods to improve it for large passive waveguide structures are developed: Adaptive Threshold Binarization and Hybrid Optimization. To demonstrate their capability, inverse design is applied to an InP waveguide platform for the first time. As an example, a polarizer with a PER of -19.4 dB is presented.

In this paper we present an example-based approach for visual speech synthesis. We combine the advantages of deep generative models and classical animation approaches to create a real-time capable facial animation framework based on volumetric captures.

Second harmonic generation is an efficient way to create coherent radiation at wavelengths that are not accessible with standard laser sources. In this work we demonstrate second harmonic generation from 1550 nm to 775 nm in a polymer photonic integrated circuit via the hybrid integration of a periodically poled lithium niobate crystal. The bulk crystal is inserted in an on-chip free-space section between two waveguide couple! d GRIN lenses. Fiber to fiber conversion efficiencies were 0.03 %/W for a continuous wave laser source and 100 %/W for a femtosecond laser source. Furthermore, third and fourth harmonic light at 517 nm and 388 nm was observed.

We demonstrate the benefits of a federated learning framework for (re)training of global ML models over geo-distributed data sources. The demonstration is carried out on a live edge computing enabled optical networking test-bed. In this demonstration, we perform real-time training of a QoT classifier by exploiting data of three different Domain Managers (DM), representing a multi-vendor ecosystem, without sharing any data with the Network Management System (NMS) in order to avoid transporting any data t! o a central location and to protect the privacy of different vendors while offering their knowledge to train a global ML model.

In autonomous driving, safety-related connected applications will coexist with infotainment services for passenger entertainment. We propose a multi-cell anticipatory networking framework with interference coordination based on Interference Calculus to serve diverse QoS requirements. The iterative approach optimizes packet transmission times leveraging service properties and channel distribution information.

Using the concept of a “THz-Wireless Fiber Extender” it is possible to combine the flexibility of wireless networks with the high capacity of fiber-optical networks. In this article, we report on a real-time short-range demonstration of a 100 Gb/s fiber extender and discuss the potential of long-range data transmission at 300 GHz using a 500-meter-long wireless link in Berlin, Germany.

To address the capacity crunch in optical metropolitan networks caused by the roll out of innovations in the context of 5G and beyond innovative approaches are required to increase the achievable throughput. Multi band systems are an interesting solution to address this issue. In this contribution, we investigate the capacity limits of such networks.

Using a coherent laboratory setup with up to 768 Gb/s data rate (64-GBd DP-64QAM), we experimentally show that multilevel coding (MLC) provides superior physical layer security (i.e. smaller security gaps) as compared to conventional bit-interleaved coded modulation (BICM) . MLC offers a flexible trade-off between security and net secure data rate.

Besides the biological and epidemiological factors, a multitude of social and economic criteria also govern the extent of the coronavirus disease spread within a population. Consequently, there is an active debate regarding the critical socio-economic determinants that contribute to the impact of the resulting pandemic. Here, we leverage Bayesian model averaging techniques and country level data to investigate the potential of 31 determinants, describing a diverse set of socio-economic characteristics, in explaining the outcome of the first wave of the coronavirus pandemic. We show that the true empirical model behind the coronavirus outcome is constituted only of few determinants. To understand the relationship between the potential determinants in the specification of the true model, we develop the coronavirus determinants Jointness space. The extent to which each determinant is able to provide a credible explanation varies between countries due to their heterogeneous socio-economic characteristics. In this aspect, the obtained Jointness map acts as a bridge between theoretical investigations and empirical observations and offers an alternate view for the joint importance of the socio-economic determinants when used for developing policies aimed at preventing future epidemic crises.

We demonstrate a fully packaged photodetector module based on surface-illuminated modified uni-traveling-carrier (MUTC) photodiode (PD) and investigate its DC and RF characteristics. It has a very a low dark current below 2 nA at the operational reverse bias of 4 V and a responsivity of 0.49 A/W at 1550 nm. The module shows a high f3dB bandwidth of 38 GHz for low optical input powers and excellent linearity with RF output power levels of 11.7 dBm at 40 GHz. The power conversion efficiency ? PCE was 6.7% at 40 GHz making it suitable for RF signal generation.

We propose two algorithms that use both models and datasets to estimate angular power spectra from channel covariance matrices in massive MIMO systems. The first algorithm is an iterative fixed-point method that solves a hierarchical problem. It uses model knowledge to narrow down candidate angular power spectra to a set that is consistent with a measured covariance matrix. Then, from this set, the algorithm selects the angular power spectrum with minimum distance to its expected value with respect to a Hilbertian metric learned from data. The second algorithm solves an alternative optimization problem with a single application of a solver for nonnegative least squares programs. By fusing information obtained from datasets and models, both algorithms can outperform existing approaches based on models, and they are also robust against environmental changes and small datasets.

Digital contact tracing approaches based on Bluetooth low energy (BLE) have the potential to efficiently contain and delay outbreaks of infectious diseases such as the ongoing SARS-CoV- 2 pandemic. In this work we propose an approach to reliably detect subjects that have spent enough time in close proximity to be at risk of being infected. Our study is an important proof of concept that will aid the battery of epidemiological policies.

We present photoconductive terahertz (THz) emitters based on rhodium (Rh) doped InGaAs for time-domain spectroscopy (TDS). The emitters feature a record high THz power of 637 µW. In combination with InGaAs:Rh receivers, a 6.5 THz bandwidth and a record peak dynamic range of 111 dB can be achieved. These improvements enable layer thickness measurement systems with unprecedented resolution and accuracy.

This paper puts forward first benchmarking results for the PTB-XL dataset, covering a variety of tasks from different ECG statement prediction tasks over age and gender prediction to signal quality assessment. We find that convolutional neural networks show the strongest performance across all tasks outperforming feature-based algorithms by a large margin.

Federated Learning comes at the cost of a significant communication overhead during training. In this work, we propose Sparse Ternary Compression (STC), a new compression framework that is specifically designed to meet the requirements of the Federated Learning environment. Our experiments on different tasks demonstrate that STC distinctively outperforms Federated Averaging in common scenarios.

Optical feedback has an impact on the tunability of lasers. We created a model of a tunable distributed Bragg reflector (DBR) laser describing the effect of optical feedback from a constant reflector distance on the wavelength tuning. Theoretical and experimental results are in good agreement. A further discussion of the model sheds light on design rules to reduce the effect of optical feedback on the tuning behavior. We introduced a new parameter called mode loss difference (MLD) as a metric for the feedback tolerance of the tuning behavior. A large MLD indicates higher tolerance of the laser to cavity length variations.

A morphed face image is a fusion of two face images and represents biometrics of two different subjects. Embedded in an official document, it can cause immense damage, since both subjects can claim its ownership and thus share an identity. In this paper, we propose and compare different neural network training schemes based on alternations of training data to obtain accurate and robust detectors for such kind of fraud. In addition, we use layer-wise relevance propagation (LRP) to analyze the differently trained networks in depth.

Federated Learning (FL) is currently the most widely adopted framework for collaborative training of (deep) machine learning models under privacy constraints. However, FL yields suboptimal results if the local clients’ data distributions diverge. The proposed Clustered FL approach tackles the problem by identifying diverging clients and grouping them into separate clusters.

Polarization division multiplexing doubles the transmission capacity of optical communication systems. For such systems, splitters separating the TE from the TM mode are indispensable components. We design and manufacture an integrated InP-based polarization rotator-splitter. The device is simplified in manufacturing and has a polarization extinction ratio of 17 dB over 60 nm optical bandwidth.

Electrocardiography (ECG) is increasingly supported by algorithms based on machine learning. We put forward PTB-XL, the to-date largest freely accessible clinical 12-lead ECG-waveform dataset comprising 21837 records from 18885 patients of 10 seconds length. The dataset covers a broad range of diagnostic classes including, in particular, a large fraction of healthy records.

AI solutions have not by large entered routine dental practice, mainly due to (1) limited data availability, accessibility, structure and comprehensiveness, (2) lacking methodological rigor and standards in their development, (3) and practical questions around the value and usefulness of these solutions, but also ethics and responsibility. This paper describes the chances of AI in medicine and dentistry.

We present an end-to-end pipeline for the creation of high-quality animatable volumetric video content of human performances. Going beyond the application of free-viewpoint volumetric video, we allow re-animation of an actor’s performance through (i) the enrichment of the captured data with semantics and animation properties and (ii) hybrid geometry- and video-based animation combined with neural infilling.

This work shows the application of explainable AI (XIA) methods to resolve common challenges encountered in deep learning-based digital histopathology analyses. We investigate three types of biases and show that XAI techniques are helpful and highly relevant tool for the development and the deployment phases of real-world applications in digital pathology.

Coherent comb lasers may serve as a source for multiwavelength modulators in short reach transmission, or for phase controlled OFDM channels in long reach. We explore and compare quantum dot (QD) and quantum well (QW) lasers with more than 33 channels in the DWDM 50 GHz grid, thus enabling > 1 Tb/s optical transmission. In addition, the mode-locked devices can be applied as pulse sources with < 500 fs pulses by using a simple SMF.

Deep models are also being increasingly applied in distributed settings, where the data are separated by limited communication channels and privacy constraints. To address the challenges, a wide range of training and evaluation schemes have been developed, which require the communication of neural network parametrizations. This paper gives an overview over the recent advancements and challenges.