The NeuroBench framework provides a standardized benchmark suite for evaluating neuromorphic algorithms and systems. Published on 2025-02-11 in Nature Communications, it addresses the lack of common metrics by introducing an open-source harness with two tracks: algorithm (hardware-independent) and system (deployed). The v1.0 release includes four benchmarks covering few-shot continual learning, event camera object detection, motor cortical decoding, and chaotic forecasting. Metrics assess correctness, memory footprint, sparsity, and synaptic operations to enable fair comparison across diverse architectures like SNNs and ANNs.
Back
NeuroBench framework for neuromorphic computing benchmarks
NeuroBench is a community-driven benchmark framework for neuromorphic computing that standardizes metrics and tasks via an open-source harness released in February 2025.