Learning expectation in insects: A recurrent spiking neural model for spatio-temporal representation

Enhancing spiking neural networks with hybrid top-down attention

As the representatives of brain-inspired models at the neuronal level, spiking neural networks (SNNs) have shown great promise in processing spatiotemporal information with intrinsic temporal dynamics. SNNs are expected to further improve their robustness and computing efficiency by introducing top-down attention at the architectural level, which is crucial for the human brain to support advanced intelligence. However, this attempt encounters difficulties in optimizing the attention in SNNs largely due to the lack of annotations. Here, we develop a hybrid network model with a top-down attention mechanism (HTDA) by incorporating an artificial neural network (ANN) to generate attention maps based on the features extracted by a feedforward SNN. The attention map is then used to modulate the encoding layer of the SNN so that it focuses on the most informative sensory input. To facilitate direct learning of attention maps and avoid labor-intensive annotations, we propose a general principle and a corresponding weakly-supervised objective, which promotes the HTDA model to utilize an integral and small subset of the input to give accurate predictions. On this basis, the ANN and the SNN can be jointly optimized by surrogate gradient descent in an end-to-end manner. We comprehensively evaluated the HTDA model on object recognition tasks, which demonstrates strong robustness to adversarial noise, high computing efficiency, and good interpretability. On the widely-adopted CIFAR-10, CIFAR-100, and MNIST benchmarks, the HTDA model reduces firing rates by up to 50% and improves adversarial robustness by up to 10% with comparable or better accuracy compared with the state-of-the-art SNNs. The HTDA model is also verified on dynamic neuromorphic datasets and achieves consistent improvements. This study provides a new way to boost the performance of SNNs by employing a hybrid top-down attention mechanism.

Insect-Inspired Robots: Bridging Biological and Artificial Systems

This review article aims to address common research questions in hexapod robotics. How can we build intelligent autonomous hexapod robots that can exploit their biomechanics, morphology, and computational systems, to achieve autonomy, adaptability, and energy efficiency comparable to small living creatures, such as insects? Are insects good models for building such intelligent hexapod robots because they are the only animals with six legs? This review article is divided into three main sections to address these questions, as well as to assist roboticists in identifying relevant and future directions in the field of hexapod robotics over the next decade. After an introduction in section (1), the sections will respectively cover the following three key areas: (2) biomechanics focused on the design of smart legs; (3) locomotion control; and (4) high-level cognition control. These interconnected and interdependent areas are all crucial to improving the level of performance of hexapod robotics in terms of energy efficiency, terrain adaptability, autonomy, and operational range. We will also discuss how the next generation of bioroboticists will be able to transfer knowledge from biology to robotics and vice versa.

Motor-Skill Learning in an Insect Inspired Neuro-Computational Control System

In nature, insects show impressive adaptation and learning capabilities. The proposed computational model takes inspiration from specific structures of the insect brain: after proposing key hypotheses on the direct involvement of the mushroom bodies (MBs) and on their neural organization, we developed a new architecture for motor learning to be applied in insect-like walking robots. The proposed model is a nonlinear control system based on spiking neurons. MBs are modeled as a nonlinear recurrent spiking neural network (SNN) with novel characteristics, able to memorize time evolutions of key parameters of the neural motor controller, so that existing motor primitives can be improved. The adopted control scheme enables the structure to efficiently cope with goal-oriented behavioral motor tasks. Here, a six-legged structure, showing a steady-state exponentially stable locomotion pattern, is exposed to the need of learning new motor skills: moving through the environment, the structure is able to modulate motor commands and implements an obstacle climbing procedure. Experimental results on a simulated hexapod robot are reported; they are obtained in a dynamic simulation environment and the robot mimicks the structures of Drosophila melanogaster.

A Fly-Inspired Mushroom Bodies Model for Sensory-Motor Control Through Sequence and Subsequence Learning

Classification and sequence learning are relevant capabilities used by living beings to extract complex information from the environment for behavioral control. The insect world is full of examples where the presentation time of specific stimuli shapes the behavioral response. On the basis of previously developed neural models, inspired by Drosophila melanogaster, a new architecture for classification and sequence learning is here presented under the perspective of the Neural Reuse theory. Classification of relevant input stimuli is performed through resonant neurons, activated by the complex dynamics generated in a lattice of recurrent spiking neurons modeling the insect Mushroom Bodies neuropile. The network devoted to context formation is able to reconstruct the learned sequence and also to trace the subsequences present in the provided input. A sensitivity analysis to parameter variation and noise is reported. Experiments on a roving robot are reported to show the capabilities of the architecture used as a neural controller.

Modeling the insect mushroom bodies: application to a delayed match-to-sample task

Despite their small brains, insects show advanced capabilities in learning and task solving. Flies, honeybees and ants are becoming a reference point in neuroscience and a main source of inspiration for autonomous robot design issues and control algorithms. In particular, honeybees demonstrate to be able to autonomously abstract complex associations and apply them in tasks involving different sensory modalities within the insect brain. Mushroom Bodies (MBs) are worthy of primary attention for understanding memory and learning functions in insects. In fact, even if their main role regards olfactory conditioning, they are involved in many behavioral achievements and learning capabilities, as has been shown in honeybees and flies. Owing to the many neurogenetic tools, the fruit fly Drosophila became a source of information for the neuroarchitecture and biochemistry of the MBs, although the MBs of flies are by far simpler in organization than their honeybee orthologs. Electrophysiological studies, in turn, became available on the MBs of locusts and honeybees. In this paper a novel bio-inspired neural architecture is presented, which represents a generalized insect MB with the basic features taken from fruit fly neuroanatomy. By mimicking a number of different MB functions and architecture, we can replace and improve formerly used artificial neural networks. The model is a multi-layer spiking neural network where key elements of the insect brain, the antennal lobes, the lateral horn region, the MBs, and their mutual interactions are modeled. In particular, the model is based on the role of parts of the MBs named MB-lobes, where interesting processing mechanisms arise on the basis of spatio-temporal pattern formation. The introduced network is able to model learning mechanisms like olfactory conditioning seen in honeybees and flies and was found able also to perform more complex and abstract associations, like the delayed matching-to-sample tasks known only from honeybees. A biological basis of the proposed model is presented together with a detailed description of the architecture. Simulation results and remarks on the biological counterpart are also reported to demonstrate the possible applications of the designed computational model. Such neural architecture, able to autonomously learn complex associations is envisaged to be a suitable basis for an immediate implementation within an robot control architecture.