Guest editorial: Special issue on brain inspired models of cognitive memory

Current memory technologies have experienced significant progress in terms of storage capacity, operation speed, integration capability, etc. However, their functions are highly constrained in storing and transferring data in space and time, prompting the need for improvement. In contrast to physical memories, the biological counterpart – cognitive memory – has versatile functions. For instance, human memory stores data associatively such that different modalities of data could be retrieved simultaneously; it can learn different concepts, categorize and store them in an organized manner; it can process and store data concurrently and in a distributed fashion; it can restore content even if some part is damaged; it can perceive the stimulus and predict the next event; it can adapt to the environment and perform selective storage. Functions such as adaptation, learning, perception, selforganization and prediction make human memory have distinct cognitive features. However, the mechanisms how human memory cognitively operates and the ways how to utilize the bioinspired mechanisms to practical applications are rarely known, yet urgently demanded. The scope of the questions on cognitive memory transcends several interdisciplinary boundaries and combines efforts in both hardware and software engineering. Increasing efforts towards cognitive memory have been made from researchers belonging to the various communities of computational intelligence, machine learning, cognitive modelling, as well as researchers in hardware (circuit level) implementation of cognitive systems and those working at materials level research, such as memristors and phase change materials. Hence, we felt that a special issue that discusses new ideas on the modelling of memory, latest results on the development of cognitively inspired memory devices, and approaches to create a bridge between hardware and system level research is very timely. This Special Issue presents six original articles covering braininspired learning rules for different cognitive tasks, and hardware implementations of the brain-inspired mechanisms. All the papers have went through a rigorous review process. The first paper, entitled A brain-inspired spiking neural network model with temporal encoding and learning, presents a biologically plausible architecture of spiking neurons for various recognition tasks. The whole system consistently operates in a temporal framework where precise spiking time is used for information processing. Both neural coding and learning are included in the system. A biologically plausible supervised synaptic learning rule is used for synaptic adaptation. This paper demonstrates a viable way of using spiking neurons to perform recognition tasks. The second paper, entitled Delay learning architectures for Memory and Classification, presents a new supervised learning algorithm for spiking neurons by modifying axonal delays rather than synaptic weights. Through tuning spike delays of presynaptic neurons, the postsynaptic neuron could have a desired response of firing or keeping salient, with synchronous/asynchronous postsynaptic currents causing large/low postsynaptic potential. Various properties of this delay learning algorithm are investigated through a binary classification task. In addition, this paper also presents that the delay learning algorithm could benefit VLSI implementation with requiring only one tunable parameter. The third paper, entitled Reinforcement Learning and Dopamine at the striatum: A Modeling Perspective, presents a review on various models of reinforcement learning with an emphasis on the cellular models of reinforcement learning. In particular, this paper emphasizes biochemical models of reinforcement learning, and some possible directions are also pointed out. The fourth paper, entitled SELP: A General-Purpose Framework for Learning the Norms from Saliencies in Spatiotemporal Data, presents a general-purpose data-driven biologically plausible deep (or hierarchical) learning framework that gives rise to a smart memory system. The framework can learn the norms or invariances as a hierarchy of features from spaceand time-varying data in an unsupervised and online manner from saliencies or surprises in the data. Given streaming data, this framework learns norms using four functions – detect salient event, explain the salient event, learn from its explanation and predict the future events. The fifth paper, entitled Learning to Predict Eye Fixations for Semantic