TL;DR: This work adopts a biologically plausible saliency-informed dropout technique as an explainable alternative to the unstructured standard random dropout approach and presents empirical results on regimes where such structured dynamic and static sparsities interact optimally to prune a ResNet model on the Imagenette dataset.
Biological vision has evolved to make efficient use of the limited information processing capability and tight energy budget of the brain by preferentially processing the most salient features of visual scenes. In contrast, modern deep vision models rely on expansive, high-dimensional representations. This may offer potential recognition gains but increases computing costs. As a consequence, the computer vision community has been exploring sparsity-enforcing techniques such as activation dropout and weight pruning. Beyond ameliorating the burden of computation, sparsity techniques such as random dropout have been shown to regularize model training, thus allowing for better generalization. Here, we pursue both dropout and weight pruning in tandem and adopt a biologically plausible saliency-informed dropout technique as an explainable alternative to the unstructured standard random dropout approach. Our approach involves a hierarchical “retinotopic" gating of convolutional feature maps, which promotes efficient deletion of “redundant" weights by iterative magnitude pruning. We explore the effectiveness of saliency-informed dropout based on different approaches to dropping based on the saliency map, dropout through all layers or only early layers, and by additionally applying dropout at inference. We compare throughout with standard random dropout. We present empirical results on regimes where such structured dynamic and static (weight) sparsities interact optimally to prune a ResNet model on the Imagenette dataset.
TL;DR: The research rigorously investigates how neuromorphic architectures encode and integrate temporal information by conducting a comprehensive ablation study using a hybrid network, and demonstrates that shallow neuromorphic integration effectively maximizes the gains from temporal integration while mitigating the information loss inherent in binary spike quantization.
While neuromorphic systems offer a promising path for processing dynamic, event-based data, current benchmarks often fail to isolate the specific impact of temporal integration on model performance. To address this, our research rigorously investigates how neuromorphic architectu…
TL;DR: A compact 2D convolutional neural network framework based on wavelet synchrosqueezing transform (WSST) time–frequency (TF) representations for EEG-based dementia classification demonstrates that the WSST-based representation provides an effective and robust input for CNN-based dementia classification under strict subject-disjoint validation.
Electroencephalography (EEG)-based discrimination of Alzheimer’s disease (AD), frontotemporal dementia (FTD), and cognitively normal controls (CN) remains challenging under clinically realistic subject-disjoint evaluation. This study presents a compact 2D convolutional neural net…
TL;DR: A sleep-inspired, replay-driven memory consolidation-based temporal learning framework that reconstructs context from replayed memory during brief sleep periods for temporal context learning in resource-constrained edge systems is proposed.
In edge computing applications, storing long temporal sequences is memory and energy intensive, while sensory data arrives sequentially, requiring online learning. Although online updates reduce storage requirements, they lack access to broader temporal context needed for accurat…
TL;DR: This work proposes heterogeneous neural networks that combine spiking neural networks (SNNs) and artificial neural networks (ANNs) at bandwidth-limited regions, such as chip boundaries, where spike-based communication reduces data transfer overhead.
Efficient communication is central to both biological and artificial intelligence (AI) systems. In biological brains, the challenge of long-range communication across regions is addressed through sparse, spike-based signaling, minimizing energy and latency. Conversely, modern AI…
TL;DR: This work presents a flexible, end-to-end workflow for the training, optimization, and deployment of SNNs across multiple neuromorphic hardware systems, with an emphasis on extensibility to future neuromorphic platforms.
Surrogate gradient methods have emerged as the dominant approach for enabling backpropagation for spiking neural networks (SNNs). However, many implementations are tightly coupled to target hardware platforms, constraining flexibility. In this work, we present a flexible, end-to-…
TL;DR: CRISP (Compiled RISP) is a neuromorphic simulator based on RISP, which compiles a given network into a minimal sequence of machine code for the desired platform entirely at runtime.
Simulators for spiking neural networks lie on the critical path between a researcher with an idea and a working network. Many of the current simulators have various implementation inefficiencies that hamper their performance across a wide variety of tasks. CRISP (Compiled RISP) i…