Local goodness-of-fit measures for neural decoding

Local goodness-of-fit measures for neural decoding identify specific spikes inconsistent with a state-space model’s predictions, outperforming standard Kullback-Leibler divergence in simulations with known model errors.

• Why it matters: Cross-validation can show that a model is imperfect but cannot pinpoint when during a recording it fits poorly, limiting targeted refinement of neural state-space decoders.
• What they did: The authors score each spike by checking whether it is consistent with the model’s prediction from all preceding spikes, using two consistency measures based on high-probability region overlap and predictive checks.
• The result: In simulations, these measures flag inconsistent spikes, while Kullback-Leibler divergence can flag spikes that only narrow broad predictions; in a rat hippocampal spatial foraging recording, the overlap measure flags more spikes during immobility and allowing jumps removes most continuous-model flags.

AI summary of the abstract · may contain errors · not medical advice · read the paper.