Fundamentals (PDF)

Machine learning is part of Artificial Intelligence since its beginning. Certainly, not learning would only allow the perfect being to show intelligent behavior. All others, be it humans or machines, need to learn in order to enhance their capabilities. In the eighties of the last century, learning from examples and modeling human learning strategies have been investigated in concert. The formal statistical basis of many learning methods has been put forward later on and is still an integral part of machine learning. Neural networks have always been in the toolbox of methods. Integrating all the pre-processing, exploitation of kernel functions, and transformation steps of a machine learning process into the architecture of a deep neural network increased the performance of this model type considerably. Modern machine learning is challenged on the one hand by the amount of data and on the other hand by the demand of real-time inference. This leads to an interest in computing architectures and modern processors. For a long time, the machine learning research could take the von-Neumann architecture for granted. All algorithms were designed for the classical CPU. Issues of implementation on a particular architecture have been ignored. This is no longer possible. The time for independently investigating machine learning and computational architecture is over.

Computing architecture has experienced a similarly rampant development from mainframe or personal computers in the last century to now very large compute clusters on the one hand and ubiquitous computing of embedded systems in the Internet of Things on the other hand. Cyber-physical systems' sensors produce a huge amount of streaming data which need to be stored and analyzed. Their actuators need to react in real-time. This clearly establishes a close connection with machine learning. Cyber-physical systems and systems in the Internet of Things consist of diverse components, heterogeneous both in hard- and software. Modern multi-core systems, graphic processors, memory technologies and hardware-software codesign offer opportunities for better implementations of machine learning models.

Machine learning and embedded systems together now form a field of research which tackles leading edge problems in machine learning, algorithm engineering, and embedded systems. Machine learning today needs to make the resource demands of learning and inference meet the resource constraints of used computer architecture and platforms. A large variety of algorithms for the same learning method and, moreover, diverse implementations of an algorithm for particular computing architectures optimize learning with respect to resource efficiency while keeping some guarantees of accuracy. The trade-off between a decreased energy consumption and an increased error rate, to just give an example, needs to be theoretically shown for training a model and the model inference. Pruning and quantization are ways of reducing the resource requirements by either compressing or approximating the model. In addition to memory and energy consumption, timeliness is an important issue, since many embedded systems are integrated into large products that interact with the physical world. If the results are delivered too late, they may have become useless. As a result, real-time guarantees are needed for such systems. To efficiently utilize the available resources, e.g., processing power, memory, and accelerators, with respect to response time, energy consumption, and power dissipation, different scheduling algorithms and resource management strategies need to be developed.

This book series addresses machine learning under resource constraints as well as the application of the described methods in various domains of science and engineering. Turning big data into smart data requires many steps of data analysis: methods for extracting and selecting features, filtering and cleaning the data, joining heterogeneous sources, aggregating the data, and learning predictions need to scale up. The algorithms are challenged on the one hand by high-throughput data, gigantic data sets like in astrophysics, on the other hand by high dimensions like in genetic data. Resource constraints are given by the relation between the demands for processing the data and the capacity of the computing machinery. The resources are runtime, memory, communication, and energy. Novel machine learning algorithms are optimized with regard to minimal resource consumption. Moreover, learned predictions are applied to program executions in order to save resources.