Today’s advances in Artificial Intelligence (AI) have been primarily driven by deep learning and have led to astounding progress in several tasks such as image classification, multiple object detection, language translation, speech recognition and even in the ability to play strategic games. However, the AI systems of today have several limitations. Specifically, the hardware infrastructure is limited to high-power and large-scale processing systems that are based on the von Neumann computing paradigm. Moreover, there is a growing demand for applications with cognitive functionality that will be able to operate in real time and in an autonomous manner in the field. The limitations of contemporary AI systems are in stark contrast to the capabilities of the brain which is able to learn and adapt very quickly consuming just about 20W of power.
Neuromorphic computing, inspired by neuroscience, is a promising path towards the next-generation AI systems. The research focuses on different levels of the design stack, i.e. the computing model, the architecture and the learning algorithms. The computing model is based on the so-called Spiking Neural Networks (SNNs), which possess more biologically-realistic neuronal dynamics compared to those of Artificial Neural Networks (ANNs). Recent advances have demonstrated the SNN/ANN duality, but so far SNN’s applicability to large scale tasks is rather limited compared to the spectacular success of deep ANNs. At the architectural level, mimicking the neuronal structure by combining memory and processing leads to the so-called in-memory computing. The concept of in-memory computing, which recently demonstrated substantial acceleration for ANNs, is well suited for SNN hardware realizations. Finally, at the algorithmic level, neuro-inspired learning paradigms are based on the insight that the brain continuously processes incoming information and is able to adapt to changing conditions. Thus, novel biologically-inspired algorithms such as online learning, learning-to-learn and unsupervised learning have become essential for low-power, accurate and reliable operation of neuromorphic computing systems.
This Research Topic aims to provide a comprehensive overview of the recent advances in the different aspects of neuromorphic computing: the computing model, the architecture and the learning algorithms. Specific themes of interest for this Research Topic include but are not limited to:
Neuronal models inspired by neuroscience and their impact on AI applications
Neuromorphic hardware realizations, including analog in-memory computing and accelerator concepts
Biologically inspired learning algorithms
Information-theoretic studies of spiking neural networks
Applications enabled neuromorphic algorithms and systems
Topic Editor Angeliki Pantazi is currently the manager of a research group at IBM Research-Europe. The rest of Topic Editors declare no competing interests with regards to the Research Topic.
Today’s advances in Artificial Intelligence (AI) have been primarily driven by deep learning and have led to astounding progress in several tasks such as image classification, multiple object detection, language translation, speech recognition and even in the ability to play strategic games. However, the AI systems of today have several limitations. Specifically, the hardware infrastructure is limited to high-power and large-scale processing systems that are based on the von Neumann computing paradigm. Moreover, there is a growing demand for applications with cognitive functionality that will be able to operate in real time and in an autonomous manner in the field. The limitations of contemporary AI systems are in stark contrast to the capabilities of the brain which is able to learn and adapt very quickly consuming just about 20W of power.
Neuromorphic computing, inspired by neuroscience, is a promising path towards the next-generation AI systems. The research focuses on different levels of the design stack, i.e. the computing model, the architecture and the learning algorithms. The computing model is based on the so-called Spiking Neural Networks (SNNs), which possess more biologically-realistic neuronal dynamics compared to those of Artificial Neural Networks (ANNs). Recent advances have demonstrated the SNN/ANN duality, but so far SNN’s applicability to large scale tasks is rather limited compared to the spectacular success of deep ANNs. At the architectural level, mimicking the neuronal structure by combining memory and processing leads to the so-called in-memory computing. The concept of in-memory computing, which recently demonstrated substantial acceleration for ANNs, is well suited for SNN hardware realizations. Finally, at the algorithmic level, neuro-inspired learning paradigms are based on the insight that the brain continuously processes incoming information and is able to adapt to changing conditions. Thus, novel biologically-inspired algorithms such as online learning, learning-to-learn and unsupervised learning have become essential for low-power, accurate and reliable operation of neuromorphic computing systems.
This Research Topic aims to provide a comprehensive overview of the recent advances in the different aspects of neuromorphic computing: the computing model, the architecture and the learning algorithms. Specific themes of interest for this Research Topic include but are not limited to:
Neuronal models inspired by neuroscience and their impact on AI applications
Neuromorphic hardware realizations, including analog in-memory computing and accelerator concepts
Biologically inspired learning algorithms
Information-theoretic studies of spiking neural networks
Applications enabled neuromorphic algorithms and systems
Topic Editor Angeliki Pantazi is currently the manager of a research group at IBM Research-Europe. The rest of Topic Editors declare no competing interests with regards to the Research Topic.