spiking neural networks (snn)

**Spiking Neural Networks (SNNs)** are **third-generation neural networks that mimic biological neurons more closely than standard formulations** — communicating via discrete binary spikes in time rather than continuous numerical values, enabling extreme energy efficiency. **What Is an SNN?** - **Neuron Model**: Leaky Integrate-and-Fire (LIF). Membrane potential accumulates charge; when it hits threshold, it "spikes" and resets. - **Signal**: Binary ($0$ or $1$) but carries information in the *timing* (rate coding or temporal coding). - **Hardware**: Ideally suited for Neuromorphic chips (Loihi) which are event-driven. **Why They Matter** - **Energy**: Sparse binary spikes mean expensive multiplications are replaced by cheap additions (or no op if 0). - **Efficiency**: Can be 100-1000x more energy efficient than ANNs for certain temporal tasks. - **Training**: Traditionally hard to train (non-differentiable spike), but Surrogate Gradient methods (SuperSpike) have solved this recently. **Spiking Neural Networks** are **silicon brains** — bringing the temporal dynamics and sparsity of biology into artificial intelligence algorithms.