Abstract

Effective contrastive learning relies on choosing useful, informative "hard" negative samples. Current hard negative sampling methods designed for traditional artificial neural networks (ANNs) are, in general, computationally costly and are ill-suited to account for the temporal structure relevant to neuromorphic data. To tackle these issues, in this work, we formulate a similarity-guided semi-hard negative sampling strategy for neuromorphic computing based on spiking neural networks (SNNs). We propose SpikeRFF (Spiking Recurrent Forward-Forward), a gradient-free and unsupervised recurrent spiking framework that combines layer-local Forward-Forward learning with semi-hard negative selection and temporally structured negative construction. Positive samples are encouraged to produce high goodness, whereas negative samples are suppressed using local goodness signals without the need for end-to-end backpropagation. Recurrent spiking dynamics further capture temporal dependencies in spike-encoded static inputs and native event streams. We additionally investigate alternative local plasticity rules beyond conventional STDP-based updates to improve representation quality, training stability, and sparse neuronal firing. This method enables the learning of a discriminative and computationally efficient embedding space from spike-based sequences, as evaluated on standard benchmark datasets such as MNIST and CIFAR-10 and neuromorphic datasets such as Neuromorphic-MNIST (N-MNIST), DVS Gesture, N-Caltech101, and Spiking Heidelberg Digits (SHD). Our approach achieves competitive performance against backpropagation-free baselines in classification accuracy, representation quality, and computational cost, without requiring complex memory banks or costly offline mining procedures.

Publication Date

7-2026

Document Type

Thesis

Student Type

Graduate

Degree Name

Artificial Intelligence (MS)

College

Golisano College of Computing and Information Sciences

Advisor

Cory Merkel

Advisor/Committee Member

Alexander Ororbia

Advisor/Committee Member

Sathwika Bavikadi

Campus

RIT – Main Campus

Download

Share

COinS