Design principles of the sparse coding network and the role of "sister cells" in the olfactory system of Drosophila

Multielectrode recordings of cockroach antennal lobe neurons in response to temporal dynamics of odor concentrations

The initial representation of the instantaneous temporal information about food odor concentration in the primary olfactory center, the antennal lobe, was examined by simultaneously recording the activity of antagonistic ON and OFF neurons with 4-channel tetrodes. During presentation of pulse-like concentration changes, ON neurons encode the rapid concentration increase at pulse onset and the pulse duration, and OFF neurons the rapid concentration decrease at pulse offset and the duration of the pulse interval. A group of ON neurons establish a concentration-invariant representation of odor pulses. The responses of ON and OFF neurons to oscillating changes in odor concentration are determined by the rate of change in dependence on the duration of the oscillation period. By adjusting sensitivity for fluctuating concentrations, these neurons improve the representation of the rate of the changing concentration. In other ON and OFF neurons, the response to changing concentrations is invariant to large variations in the rate of change due to variations in the oscillation period, facilitating odor identification in the antennal-lobe. The independent processing of odor identity and the temporal dynamics of odor concentration may speed up processing time and improve behavioral performance associated with plume tracking, especially when the air is not moving.

Long-range connections are crucial for synchronization transition in a computational model of Drosophila brain dynamics

The synchronization transition type has been the focus of attention in recent years because it is associated with many functional characteristics of the brain. In this paper, the synchronization transition in neural networks with sleep-related biological drives in Drosophila is investigated. An electrical synaptic neural network is established to research the difference between the synchronization transition of the network during sleep and wake, in which neurons regularly spike during sleep and chaotically spike during wake. The synchronization transition curves are calculated mainly using the global instantaneous order parameters S. The underlying mechanisms and types of synchronization transition during sleep are different from those during wake. During sleep, regardless of the network structure, a frustrated (discontinuous) transition can be observed. Moreover, the phenomenon of quasi periodic partial synchronization is observed in ring-shaped regular network with and without random long-range connections. As the network becomes dense, the synchronization of the network only needs to slightly increase the coupling strength g. While during wake, the synchronization transition of the neural network is very dependent on the network structure, and three mechanisms of synchronization transition have emerged: discontinuous synchronization (explosive synchronization and frustrated synchronization), and continuous synchronization. The random long-range connections is the main topological factor that plays an important role in the resulting synchronization transition. Furthermore, similarities and differences are found by comparing synchronization transition research for the Hodgkin-Huxley neural network in the beta-band and gammma-band, which can further improve the synchronization phase transition research of biologically motivated neural networks. A complete research framework can also be used to study coupled nervous systems, which can be extended to general coupled dynamic systems.

Combining STDP and binary networks for reinforcement learning from images and sparse rewards

Spiking neural networks (SNNs) aim to replicate energy efficiency, learning speed and temporal processing of biological brains. However, accuracy and learning speed of such networks is still behind reinforcement learning (RL) models based on traditional neural models. This work combines a pre-trained binary convolutional neural network with an SNN trained online through reward-modulated STDP in order to leverage advantages of both models. The spiking network is an extension of its previous version, with improvements in architecture and dynamics to address a more challenging task. We focus on extensive experimental evaluation of the proposed model with optimized state-of-the-art baselines, namely proximal policy optimization (PPO) and deep Q network (DQN). The models are compared on a grid-world environment with high dimensional observations, consisting of RGB images with up to 256 × 256 pixels. The experimental results show that the proposed architecture can be a competitive alternative to deep reinforcement learning (DRL) in the evaluated environment and provide a foundation for more complex future applications of spiking networks.

Gap junction networks in mushroom bodies participate in visual learning and memory in Drosophila

Gap junctions are widely distributed in the brains across species and play essential roles in neural information processing. However, the role of gap junctions in insect cognition remains poorly understood. Using a flight simulator paradigm and genetic tools, we found that gap junctions are present in Drosophila Kenyon cells (KCs), the major neurons of the mushroom bodies (MBs), and showed that they play an important role in visual learning and memory. Using a dye coupling approach, we determined the distribution of gap junctions in KCs. Furthermore, we identified a single pair of MB output neurons (MBONs) that possess a gap junction connection to KCs, and provide strong evidence that this connection is also required for visual learning and memory. Together, our results reveal gap junction networks in KCs and the KC-MBON circuit, and bring new insight into the synaptic network underlying fly's visual learning and memory.