Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces a hybrid architecture that integrates pretrained artificial neural networks (ANNs) with spiking neural networks (SNNs) to enhance performance on image classification tasks. By employing a pretrained EfficientNet encoder to generate spike trains through rate-coding and training a CoLaNET spiking classifier with local plasticity rules, the authors achieve a remarkable 99.09% accuracy on a 64-class ImageNet benchmark. This method not only matches the performance of traditional deep networks but also offers a biologically inspired approach to leveraging ANN embeddings in SNNs, highlighting the potential for more efficient neural computation.
Achieving 99.09% accuracy on ImageNet with a hybrid ANN-SNN model that uses local plasticity could redefine the boundaries of spiking neural network capabilities.
This work proposes a hybrid ANN-SNN pipeline that effectively leverages the rich embeddings of pretrained artificial neural networks (ANNs) to enable high-performance spiking neural networks (SNNs). The architecture couples a pretrained EfficientNet encoder with a CoLaNET spiking classifier. We convert the encoder's activations into spike trains via rate-coding and train the subsequent SNN classifier using local, biologically inspired learning rules, bypassing end-to-end gradient propagation. This approach achieves 99.09% accuracy on a 64-class ImageNet benchmark, demonstrating performance on par with conventional deep networks. The work presents a biologically plausible and efficient framework for adapting powerful pretrained encoders to downstream spiking neural network tasks.