Supervised Sadp Eval

supervised-sadp-eval

Supervised Spike Agreement Dependent Plasticity for Fast Local Learning in Spiking Neural Networks — Gouri Lakshmi S et al. (2026) (arXiv:2601.08526, 2026)

What this evaluates

Evaluates the ability of a gradient-free, locally-updated spiking neural network to classify images using population-level spike agreement metrics. It probes whether replacing backpropagation with supervised Spike Agreement-Dependent Plasticity (SADP) and Cohen’s κ can achieve competitive vision and biomedical classification performance while maintaining biological plausibility and hardware compatibility.

Datasets

• MNIST — total 70000; splits: train (60000), test (10000)
• Fashion-MNIST — total 70000; splits: train (-1), test (-1)
• CIFAR-10 — total 60000; splits: train (-1), test (-1)
• LC25000 — total 25000; splits: colon_binary (-1), lung_3class (-1)
• Brain MRI Tumor — total ?; splits: train (-1), test (-1)

Metrics

• accuracy (primary) — range: [0, 1]
  • Fraction of correctly classified samples in the test set.
• macro-averaged F1-score — range: [0, 1]
  • Mean of the F1-scores for each class, computed over the test set.
• Runtime per Epoch — range: other
  • Wall-clock time required to complete one full training epoch.

Input / output format

Input: Image pixels normalized to [0,1], converted to stochastic spike trains via Poisson encoding over T=25 or T=100 timesteps. Alternatively, 256-dimensional CNN-extracted features normalized and converted to Poisson spike trains.

Output: Class label prediction derived from the output layer's spike activity (1SADP or 2SADP architecture).

Scoring recipe

def compute_metrics(y_true, y_pred):
    accuracy = (y_true == y_pred).mean()
    f1_macro = f1_score(y_true, y_pred, average='macro')
    return accuracy, f1_macro

Common pitfalls

• Assuming standard backpropagation or surrogate gradients are used for weight updates instead of the local Hebbian and kappa-based SADP rules.
• Overlooking the impact of temporal resolution (T=25 vs T=100) on spike train length, which directly affects both runtime and learning dynamics.
• Confusing the two encoding schemes (Poisson-only vs CNN+Poisson) when comparing results across datasets of varying complexity.

Evidence (verbatim from paper)

Model evaluation metrics included accuracy, macro-averaged F1-score computed over the test set and Runtime per Epoch.

Citation

@misc{gourilakshmi2026supervised,
  title={Supervised Spike Agreement Dependent Plasticity for Fast Local Learning in Spiking Neural Networks},
  author={Gouri Lakshmi S et al. (2026)},
  year={2026},
  note={arXiv:2601.08526}
}

• arXiv: 2601.08526