Exploring Trade-Offs in Spiking Neural Networks Spiking neural networks (SNNs) have emerged as a promising alternative to traditional deep neural networks for low-power computing. However, the effectiveness of SNNs is not solely determined by their performance but also by their energy consumption, prediction speed, and robustness to noise. The recent method Fast & Deep, along with others, achieves fast and energy-efficient computation by constraining neurons to fire at most once. Known as time-to-first-spike (TTFS), this constraint, however, restricts the capabilities of SNNs in many aspects. In this work, we explore the relationships of performance, energy consumption, speed, and stability when using this constraint. More precisely, we highlight the existence of trade-offs where performance and robustness are gained at the cost of sparsity and prediction latency. To improve these trade-offs, we propose a relaxed version of Fast & Deep that allows for multiple spikes per neuron. Our experiments show that relaxing the spike constraint provides higher performance while also benefiting from faster convergence, similar sparsity, comparable prediction latency, and better robustness to noise compared to TTFS SNNs. By highlighting the limitations of TTFS and demonstrating the advantages of unconstrained SNNs, we provide valuable insight for the development of effective learning strategies for neuromorphic computing.
Attention and microsaccades: do attention shifts trigger new microsaccades or only bias ongoing microsaccades? Brain circuitry that controls where we look also contributes to attentional focusing of visual contents outside of current fixation or contents held within the spatial layout of working memory. A behavioural manifestation of the contribution of oculomotor brain circuitry to selective attention comes from modulations in microsaccade direction that accompany attention shifts. Here, we address whether such modulations come about because attention itself triggers new microsaccades or whether, instead, shifts in attention only bias the direction of ongoing microsaccades – i.e., naturally occurring microsaccades that would have been made whether or not attention was also shifted. We utilised an internal-selective-attention task that has recently been shown to yield clear spatial microsaccade modulations and compared microsaccade rates following colour retrocues that were matched for sensory input, but differed in whether they invited an attention shift or not. If shifts in attention trigger new microsaccades then we would expect more microsaccades following attention-directing cues than following neutral cues. In contrast, we found no evidence for an increase in overall microsaccade rate following attention-directing cues, despite observing robust modulations in microsaccade direction. This implies that shifting attention biases the direction of ongoing microsaccades without changing the probability that a microsaccade will occur. These findings provide relevant context for complementary and future work delineating the links between attention, microsaccades, and upstream oculomotor brain circuitry, such as by helping to explain why microsaccades and attention shifts are often correlated but not obligatorily linked.
Visuomotor memory is not bound to visual motion Sensory feedback plays a critical role in motor control and motor learning, with both processes adjusting outgoing motor commands relative to the error between actual sensory feedback and the predictions of a forward model. However, models of motor control rarely specify the exact nature of these predictions. We hypothesized that large differences in low-level perceptual feedback would delineate contextual boundaries for motor memory, as sufficiently different feedback would require a distinct mapping between motor commands and sensory states. We tested this hypothesis by measuring transfer of visuomotor adaptation across contexts where hand movements caused visual motion in opposite directions (180 ° away). Instead of observing that visual feedback is bound to distinct internal models, we found nearly complete transfer of learning across the contexts. Finally, we found evidence that the motor memory was bound to the planned displacement of the hand, rather than visual features of the task space.l transformations in VVS and PFC that give rise to distinct coding schemes of prioritized contents.
Examining Letter Detector Tolerance through Offset Letter Halves: Evidence from Lexical Decision Neurobiological models of reading assume that the specialized detectors at the letter level (e.g., the arrays of detectors for the letter ‘n’) possess a certain degree of tolerance (e.g., Local Combination Detectors model, Dehaene et al. 2005). In this study, we designed two lexical decision experiments that examined the limits of tolerance of letter detectors by introducing a novel manipulation involving shifting letter halves (e.g., animal in Experiment 1; animal in Experiment 2) relative to intact items. This manipulation alters the transition between upper and lower parts of the letters, adding junctions that do not exist in the intact letter forms. We included high- and low-frequency words in the stimulus list to investigate whether letter distortion affects processing beyond the letter level, reasoning that interactive effects would signal top-down lexical feedback. In Experiment 1, which employed a subtle letter shift, we observed a minimal cost of letter distortion that did not interact with word frequency. Experiment 2, employing a larger letter shift, revealed an overall greater reading cost that affected differentially high- and low-frequency words. Overall, these findings offer insights into the limits of resilience in letter detectors to distortion during word recognition and introduce a novel manipulation of letter distortion.
Stress relief as a natural resilience mechanism against depression-like behaviors Relief, the appetitive state after the termination of aversive stimuli, is evolutionarily conserved. Understanding the behavioral role of this well-conserved phenomenon and its underlying neurobiological mechanisms are open and important questions. Here, we discover that the magnitude of relief from physical stress strongly correlates with individual resilience to depression-like behaviors in chronic stressed mice. Notably, blocking stress relief causes vulnerability to depression-like behaviors, whereas natural rewards supplied shortly after stress promotes resilience. Stress relief is mediated by reward-related mesolimbic dopamine neurons, which show minute-long, persistent activation after stress termination. Circuitry-wise, activation or inhibition of circuits downstream of the ventral tegmental area during the transient relief period bi-directionally regulates depression resilience. These results reveal an evolutionary function of stress relief in depression resilience and identify the neural substrate mediating this effect. Importantly, our data suggest a behavioral strategy of augmenting positive valence of stress relief with natural rewards to prevent depression.
Optoα1AR activation in astrocytes modulates basal hippocampal synaptic excitation and inhibition in a stimulation-specific manner Astrocytes play active roles at synapses and can monitor, respond, and adapt to local synaptic activity. While there is abundant evidence that astrocytes modulate excitatory transmission in the hippocampus, evidence for astrocytic modulation of hippocampal synaptic inhibition remains more limited. Furthermore, to better investigate roles for astrocytes in modulating synaptic transmission, more tools that can selectively activate native G protein signaling pathways in astrocytes with both spatial and temporal precision are needed. Here, we utilized AAV8-GFAP-Optoα1AR-eYFP (Optoα1AR), a viral vector that enables activation of Gq signaling in astrocytes via light-sensitive α1-adrenergic receptors. To determine if stimulating astrocytic Optoα1AR modulates hippocampal synaptic transmission, recordings were made in CA1 pyramidal cells with surrounding astrocytes expressing Optoα1AR, channelrhodopsin (ChR2), or GFP. Both high-frequency (20 Hz, 45-ms light pulses, 5 mW, 5 min) and low-frequency (0.5 Hz, 1-s pulses at increasing 1, 5, and 10 mW intensities, 90 s per intensity) blue light stimulation were tested. 20 Hz Optoα1AR stimulation increased both inhibitory and excitatory postsynaptic current (IPSC and EPSC) frequency, and the effect on miniature IPSCs (mIPSCs) was largely reversible within 20 min. However, low-frequency stimulation of Optoα1AR did not modulate either IPSCs or EPSCs, suggesting that astrocytic Gq-dependent modulation of basal synaptic transmission in the hippocampus is stimulation-dependent. By contrast, low-frequency stimulation of astrocytic ChR2 was effective in increasing both synaptic excitation and inhibition. Together, these data demonstrate that Optoα1AR activation in astrocytes changes basal GABAergic and glutamatergic transmission, but only following high-frequency stimulation, highlighting the importance of temporal dynamics when using optical tools to manipulate astrocyte function.
Estrogen receptor beta in astrocytes modulates cognitive function in mid-age female mice Menopause is associated with cognitive deficits and brain atrophy, but the brain region and cell-specific mechanisms are not fully understood. Here, we identify a sex hormone by age interaction whereby loss of ovarian hormones in female mice at midlife, but not young age, induced hippocampal-dependent cognitive impairment, dorsal hippocampal atrophy, and astrocyte and microglia activation with synaptic loss. Selective deletion of estrogen receptor beta (ERβ) in astrocytes, but not neurons, in gonadally intact female mice induced the same brain effects. RNA sequencing and pathway analyses of gene expression in hippocampal astrocytes from midlife female astrocyte-ERβ conditional knock out (cKO) mice revealed Gluconeogenesis I and Glycolysis I as the most differentially expressed pathways. Enolase 1 gene expression was increased in hippocampi from both astrocyte-ERβ cKO female mice at midlife and from postmenopausal women. Gain of function studies showed that ERβ ligand treatment of midlife female mice reversed dorsal hippocampal neuropathology.
Nfia is Critical for AII Amacrine Cell Production: Selective Bipolar Cell Dependencies and Diminished ERG The NFI transcription factor genes Nfia, Nfib and Nfix are all enriched in late-stage retinal progenitor cells, and their loss has been shown to retain these progenitors at the expense of later-generated retinal cell types. Whether they play any role in the specification of those later generated fates is unknown, but the expression of one of these, Nfia, in a specific amacrine cell type may intimate such a role. Here, Nfia-CKO mice (both sexes) were assessed, finding a massive and largely selective absence of AII amacrine cells. There was, however, a partial reduction in Type 2 cone bipolar cells (CBCs), being richly interconnected to AII cells. Counts of dying cells showed a significant increase in Nfia-CKO retinas at P7, after AII cell numbers were already reduced but in advance of the loss of Type 2 CBCs detected by P10. Those results suggest a role for Nfia in the specification of the AII amacrine cell fate, and a dependency of the Type 2 CBCs upon them. Delaying the conditional loss of Nfia to the first postnatal week did not alter AII cell number nor differentiation, further suggesting that its role in AII cells is solely associated with their production. The physiological consequences of their loss were assessed using the ERG, finding the oscillatory potentials to be profoundly diminished. A slight reduction in the b-wave was also detected, attributed to an altered distribution of the terminals of rod bipolar cells, implicating a role of the AII amacrine cells in constraining their stratification.