Optimal sparse olfactory representations persist in a plastic network

Abstract

18The neural representation of a stimulus is repeatedly transformed as it moves from 19 the sensory periphery to deeper layers of the nervous system. Sparsening 20 transformations are thought to increase the separation between similar 21 representations, encode stimuli with great specificity, maximize storage capacity as 22 weight changes caused by spike timing dependent plasticity increase the distance 42 between the odor representations from the perspective of MBONs and do not lead to 43 a concomitant change in the optimal sparseness. 44 45 Author Summary 46 Kenyon cells (KCs) of the mushroom body represent odors as a sparse code. When 47 viewed from the perspective of follower neurons, mushroom body output neurons 48 (MBONs), an optimal level of KC sparseness maximally separates the 49 representations of odors. However, the KC-MBON synapse is highly plastic and may 50 be potentiated or depressed by odor-driven experience that could, in turn, perturb 51 the optimality formed by pre-synaptic circuits. Contrary to this expectation, we show 52 that synaptic plasticity based on spike timing of pre-and postsynaptic neurons 53 improves the ability of the system to distinguish between the representations of 54 similar odors while preserving the optimality determined by pre-synaptic circuits. 55 56 representations? Second, if an optimal sparseness exists, does plasticity at the KC-131 MBON synapse alter it, making post-learning odor representations sub-optimal? To 132 address these questions, we constructed a computational model of the locust 133 olfactory system consisting of the olfactory receptor neurons, the antennal lobe 134 network of PNs and local inhibitory interneurons, the KCs of the mushroom body, 135and a layer of MBONs (Figure 1a). The model antennal lobe network generates 136 many of the key responses previously documented in the locust (16,17). The output

A neuromorphic model of olfactory processing and sparse coding in the Drosophila larva brain

“…Our analysis in Fig. 3B showed that inhibitory feedback has the strongest effect, confirming experimental [109, 110] and modeling results [24, 26, 27, 111]. Cellular adaptation (SFA) showed a smaller but supporting effect in our network, which is partially in line with our previous models of the adult fly [27, 51, 109] in which we showed that SFA alone can suffice to generate high temporal sparseness.…”

“…Empirical evidence has been provided in particular in bees [93, 104] and adult flies [46, 81]. Several modeling studies have used feedback inhibition to support a sparse KC population code in larger adult KC populations [24, 26, 27, 105]. Indeed, our study shows that inhibitory feedback from the single APL neuron effectively implements a sparse code in the small population of 72 KCs (Fig.…”

“…To this end, the network topology includes three relevant motifs. First, the LN connectivity in the antennal lobe constitutes lateral inhibition as a motif that generally enhances neural contrast [33, 71] and that is implemented in the olfactory system of virtually all insects [35, 72–77], as well as in computational models thereof [24, 27, 51, 78]. Second, the random connectivity from PNs to a larger number of KCs is net divergent and sparse, expanding the dimensionality of the coding space [50, 79, 80].…”

“…At the same time, their compact nervous systems are optimized for energy efficient computation with limited numbers of neurons and synapses, making them ideally suited to meet current neuromorphic hardware limitations regarding network size and topology. Spiking neural networks modelled after the insect brain have been shown to support efficient sensory processing [24], learning [7, 25], foraging and navigation [26–28], and counting [27]. Model studies also include earlier neuromorphic implementations of insect-inspired computation [4, 5, 9, 29–32].…”