Refine
Document Type
- Conference Proceeding (19) (remove)
Conference Type
- Konferenzartikel (18)
- Konferenzband (1)
Keywords
- Neural networks (2)
- efficient training (2)
- 5G (1)
- Blockchain (1)
- Car-to-Car-(C2C)-Communication (1)
- Cyber Physical Systems, (1)
- Edge AI (1)
- Eingebettetes System (1)
- Embedded AI (1)
- Embedded Systems (1)
Institute
- ivESK - Institut für verlässliche Embedded Systems und Kommunikationselektronik (19) (remove)
Open Access
- Open Access (19) (remove)
Training deep neural networks using backpropagation is very memory and computationally intensive. This makes it difficult to run on-device learning or fine-tune neural networks on tiny, embedded devices such as low-power micro-controller units (MCUs). Sparse backpropagation algorithms try to reduce the computational load of on-device learning by training only a subset of the weights and biases. Existing approaches use a static number of weights to train. A poor choice of this so-called backpropagation ratio limits either the computational gain or can lead to severe accuracy losses. In this paper we present TinyProp, the first sparse backpropagation method that dynamically adapts the back-propagation ratio during on-device training for each training step. TinyProp induces a small calculation overhead to sort the elements of the gradient, which does not significantly impact the computational gains. TinyProp works particularly well on fine-tuning trained networks on MCUs, which is a typical use case for embedded applications. For typical datasets from three datasets MNIST, DCASE2020 and CIFAR10, we are 5 times faster compared to non-sparse training with an accuracy loss of on average 1%. On average, TinyProp is 2.9 times faster than existing, static sparse backpropagation algorithms and the accuracy loss is reduced on average by 6 % compared to a typical static setting of the back-propagation ratio.
The importance of machine learning has been increasing dramatically for years. From assistance systems to production optimisation to support the health sector, almost every area of daily life and industry comes into contact with machine learning. Besides all the benefits that ML brings, the lack of transparency and the difficulty in creating traceability pose major risks. While there are solutions that make the training of machine learning models more transparent, traceability is still a major challenge. Ensuring the identity of a model is another challenge. Unnoticed modification of a model is also a danger when using ML. One solution is to create an ML birth certificate and an ML family tree secured by blockchain technology. Important information about training and changes to the model through retraining can be stored in a blockchain and accessed by any user to create more security and traceability about an ML model.
Spatially Distributed Wireless Networks (SDWN) are one of the basic technologies for the Internet of Things (IoT) and (Industrial) Internet of Things (IIoT) applications. These SDWN for many of these applications has strict requirements such as low cost, simple installation and operations, and high potential flexibility and mobility. Among the different Narrowband Wireless Wide Area Networking (NBWWAN) technologies, which are introduced to address these categories of wireless networking requirements, Narrowband Internet of Things (NB-IoT) is getting more traction due to attractive system parameters, energy-saving mode of operation with low data rates and bandwidth, and its applicability in 5G use cases. Since several technologies are available and because the underlying use cases come with various requirements, it is essential to perform a systematic comparative analysis of competing technologies to choose the right technology. It is also important to perform testing during different phases of the system development life cycle. This paper describes the systematic test environment for automated testing of radio communication and systematic measurements of the performance of NB-IoT.
In this paper, we study the runtime performance of symmetric cryptographic algorithms on an embedded ARM Cortex-M4 platform. Symmetric cryptographic algorithms can serve to protect the integrity and optionally, if supported by the algorithm, the confidentiality of data. A broad range of well-established algorithms exists, where the different algorithms typically have different properties and come with different computational complexity. On deeply embedded systems, the overhead imposed by cryptographic operations may be significant. We execute the algorithms AES-GCM, ChaCha20-Poly1305, HMAC-SHA256, KMAC, and SipHash on an STM32 embedded microcontroller and benchmark the execution times of the algorithms as a function of the input lengths.
In recent years, the topic of embedded machine learning has become very popular in AI research. With the help of various compression techniques such as pruning, quantization and others compression techniques, it became possible to run neural networks on embedded devices. These techniques have opened up a whole new application area for machine learning. They range from smart products such as voice assistants to smart sensors that are needed in robotics. Despite the achievements in embedded machine learning, efficient algorithms for training neural networks in constrained domains are still lacking. Training on embedded devices will open up further fields of applications. Efficient training algorithms would enable federated learning on embedded devices, in which the data remains where it was collected, or retraining of neural networks in different domains. In this paper, we summarize techniques that make training on embedded devices possible. We first describe the need and requirements for such algorithms. Then we examine existing techniques that address training in resource-constrained environments as well as techniques that are also suitable for training on embedded devices, such as incremental learning. At the end, we also discuss which problems and open questions still need to be solved in these areas.
MPC-Workshop Juli 2015
(2015)
To demonstrate how deep learning can be applied to industrial applications with limited training data, deep learning methodologies are used in three different applications. In this paper, we perform unsupervised deep learning utilizing variational autoencoders and demonstrate that federated learning is a communication efficient concept for machine learning that protects data privacy. As an example, variational autoencoders are utilized to cluster and visualize data from a microelectromechanical systems foundry. Federated learning is used in a predictive maintenance scenario using the C-MAPSS dataset.
The Metering Bus, also known as M-Bus, is a European standard EN13757-3 for reading out metering devices, like electricity, water, gas, or heat meters. Although real-life M-Bus networks can reach a significant size and complexity, only very simple protocol analyzers are available to observe and maintain such networks. In order to provide developers and installers with the ability to analyze the real bus signals easily, a web-based monitoring tool for the M-Bus has been designed and implemented. Combined with a physical bus interface it allows for measuring and recording the bus signals. For this at first a circuit has been developed, which transforms the voltage and current-modulated M-Bus signals to a voltage signal that can be read by a standard ADC and processed by an MCU. The bus signals and packets are displayed using a web server, which analyzes and classifies the frame fragments. As an additional feature an oscilloscope functionality is included in order to visualize the physical signal on the bus. This paper describes the development of the read-out circuit for the Wired M-Bus and the data recovery.
The CAN bus still is an important fieldbus in various domains, e.g. for in-car communication or automation applications. To counter security threats and concerns in such scenarios we design, implement, and evaluate the use of an end-to-end security concept based on the Transport Layer Security protocol. It is used to establish authenticated, integrity-checked, and confidential communication channels between field devices connected via CAN. Our performance measurements show that it is possible to use TLS at least for non time-critical applications, as well as for generic embedded networks.
The paper describes the methodology and experimental results for revealing similarities in thermal dependencies of biases of accelerometers and gyroscopes from 250 inertial MEMS chips (MPU-9250). Temperature profiles were measured on an experimental setup with a Peltier element for temperature control. Classification of temperature curves was carried out with machine learning approach.
A perfect sensor should not have thermal dependency at all. Thus, only sensors inside the clusters with smaller dependency (smaller total temperature slopes) might be pre-selected for production of high accuracy inertial navigation modules. It was found that no unified thermal profile (“family” curve) exists for all sensors in a production batch. However, obviously, sensors might be grouped according to their parameters. Therefore, the temperature compensation profiles might be regressed for each group. 12 slope coefficients on 5 degrees temperature intervals from 0°C to +60°C were used as the features for the k-means++ clustering algorithm.
The minimum number of clusters for all sensors to be well separated from each other by bias thermal profiles in our case is 6. It was found by applying the elbow method. For each cluster a regression curve can be obtained.
Environmental Monitoring is an attractive application field for Wireless Sensor Network (WSN). Water Level Monitoring helps to increase the efficiency of water distribution and management. In Pakistan, the world’s largest irrigation system covers 90.000 km of channels which needs to be monitored and managed on different levels. Especially the sensor systems for the small distribution channels need to be low energy and low cost. The distribution presents a technical solution for a communication system which is developed in a research project being co-funded by German Academic Exchange Service (DAAD). The communication module is based on IEEE-802.15.4 transceivers which are enhanced through Wake-On-Radio (WOR) to combine low-energy and real-time behavior. On higher layers, IPv6 (6LoWPAN) and corresponding routing protocols like Routing Protocol for Low power and Lossy Networks (RPL) can extend range of the network. The data are stored in a database and can be viewed online via a web interface. Of course, also automatic data analysis can be performed.
Institute of Reliable Embedded Systems and Communication Electronics, Offenburg University of Applied Sciences, Germany has developed an automated testing environment, Automated Physical TestBeds (APTB), for analyzing the performance of wireless systems and its supporting protocols. Wireless physical networking nodes can connect to this APTB and the antenna output of this attaches with the RF waveguides. To model the RF environment this RF waveguides then establish wired connection among RF elements like splitters, attenuators and switches. In such kind of set up it’s well possible to vary the path characteristics by altering the attenuators and switches. The major advantage of using APTB is the possibility of isolated, well controlled, repeatable test environment in various conditions to run statistical analysis and even to execute regression tests. This paper provides an overview of the design and implementation of APTB, demonstrates its ability to automate test cases, and its efficiency.
The Thread protocol is a recent development based on 6LoWPAN (IPv6 over IEEE 802.15.4), but with extensions regarding a more media independent approach, which – additionally – also promises true interoperability. To evaluate and analyse the operation of a Thread network a given open source 6LoWPAN stack for embedded devices (emb::6) has been extended in order to comply with the Thread specification. The implementation covers Mesh Link Establishment (MLE) and network layer functionality as well as 6LoWPAN mesh under routing mechanism based on MAC short addresses. The development has been verified on a virtualization platform and allows dynamical establishment of network topologies based on Thread's partitioning algorithm.
OPC UA (Open Platform Communications Unified Architecture) is already a well-known concept used widely in the automation industry. In the area of factory automation, OPC UA models the underlying field devices such as sensors and actuators in an OPC UA server to allow connecting OPC UA clients to access device-specific information via a standardized information model. One of the requirements of the OPC UA server to represent field device data using its information model is to have advanced knowledge about the properties of the field devices in the form of device descriptions. The international standard IEC 61804 specifies EDDL (Electronic Device Description Language) as a generic language for describing the properties of field devices. In this paper, the authors describe a possibility to dynamically map and integrate field device descriptions based on EDDL into OPCUA.
Legacy industrial communication protocols are proved robust and functional. During the last decades, the industry has invented completely new or advanced versions of the legacy communication solutions. However, even with the high adoption rate of these new solutions, still the majority industry applications run on legacy, mostly fieldbus related technologies. Profibus is one of those technologies that still keep on growing in the market, albeit a slow in market growth in recent years. A retrofit technology that would enable these technologies to connect to the Internet of Things, utilize the ever growing potential of data analysis, predictive maintenance or cloud-based application, while at the same time not changing a running system is fundamental.
Die neueste Generation von programmierbaren Logikbausteinen verfügt neben den konfigurierbaren Logikzellen über einen oder mehrere leistungsfähige Mikroprozessoren. In dieser Arbeit wird gezeigt, wie ein bestehendes Zwei-Chip-System auf einen Xilinx Zynq 7000 mit zwei ARM A9-Cores migriert wird. Bei dem System handelt es sich um das „GPS-gestützte Kreisel-system ADMA“ des Unternehmens GeneSys. Die neue Lösung verbessert den Datenaustausch zwischen dem ersten Mikroprozessor zur digitalen Signalverarbeitung und dem zweiten Prozessor zur Ablaufsteuerung durch ein Shared Memory. Für die schnelle und echtzeitfähige Datenübertragung werden zahlreiche hochbitratige Schnittstellengenutzt.
Wireless communication systems more and more become part of our daily live. Especially with the Internet of Things (IoT) the overall connectivity increases rapidly since everyday objects become part of the global network. For this purpose several new wireless protocols have arisen, whereas 6LoWPAN (IPv6 over Low power Wireless Personal Area Networks) can be seen as one of the most important protocols within this sector. Originally designed on top of the IEEE802.15.4 standard it is a subject to various adaptions that will allow to use 6LoWPAN over different technologies; e.g. DECT Ultra Low Energy (ULE). Although this high connectivity offers a lot of new possibilities, there are several requirements and pitfalls coming along with such new systems. With an increasing number of connected devices the interoperability between different providers is one of the biggest challenges, which makes it necessary to verify the functionality and stability of the devices and the network. Therefore testing becomes one of the key components that decides on success or failure of such a system. Although there are several protocol implementations commonly available; e.g., for IoT based systems, there is still a lack of according tools and environments as well as for functional and conformance testing. This article describes the architecture and functioning of the proposed test framework based on Testing and Test Control Notation Version 3 (TTCN-3) for 6LoWPAN over ULE networks.
The Internet of Things (IoT), ubiquitous computing and ubiquitous connectivity, Cyber Physical Systems (CPS), ambient intelligence, Machine-to-Machine communication (M2M) or Car-to-Car (C2C)-communication, smart metering, smart grid, telematics, telecare, telehealth – there are many buzzwords around current developments related to the Internet.
This contribution gives an overview on such IoT-applications, as they are already used today to improve the availability of information, increase efficiency, push system limits and extend the value chain. At a closer look, the economic and technical development can be separated into different phases. It is interesting that we are currently at the threshold to a new phase, with decentralized and cooperative communication and control nodes as cornerstones. Thus, embedded systems and their connectivity are in the middle of the scene.
This recent development is described along with some example projects from the author’s team which are used in industrial automation, energy supply and distribution (home automation and smart metering), traffic engineering (cooperative driver assistance systems), and in telehealth and telecare.
Covert and Side-Channels have been known for a long time due to their versatile forms of appearance. For nearly every technical improvement or change in technology, such channels have been (re-)created or known methods have been adapted. For example the introduction of hyperthreading technology has introduced new possibilities for covert communication between malicious processes because they can now share the arithmetic logical unit (ALU) as well as the L1 and L2 cache which enables establishing multiple covert channels. Even virtualization which is known for its isolation of multiple machines is prone to covert and side-channel attacks due to the sharing of resources. Therefore itis not surprising that cloud computing is not immune to this kind of attacks. Even more, cloud computing with multiple, possibly competing users or customers using the same shared resources may elevate the risk of unwanted communication. In such a setting the ”air gap” between physical servers and networks disappears and only the means of isolation and virtual separation serve as a barrier between adversary and victim. In the work at hand we will provide a survey on weak spots an adversary trying to exfiltrate private data from target virtual machines could exploit in a cloud environment. We will evaluate the feasibility of example attacks and point out possible mitigation solutions if they exist.