Neural Noise

Temporal coding and stable cortical visual representations for neural decoding

2 min read

  • Cortical visual representations are more stable over time when using temporal coding than when using firing rate (Nature, electrophysiology).1
  • Decoder design and choice of neural feature (temporal vs rate) matter for long-term BCI stability.1 1

Gardner updates

  • Cortical visual representations are more stable over time when using temporal coding than firing rate; decoder design and feature choice matter for long-term BCI stability. 1

Weekly enrichment (2026-07-20)

  • The underlying study, “Temporal coding carries more stable cortical visual representations than firing rate over time,” appeared in Nature Communications (2025) from Zhu, He, Tolias, Luan and colleagues, using custom large-scale ultraflexible nanoelectronic-thread (NET) electrode arrays about 1 μm thick that integrate without glial scarring.2
  • Researchers tracked the same visual-cortex neuronal populations in male mice across 15 consecutive days, presenting drifting gratings, static gratings, receptive-field Gabors, and natural scenes to compare rate- versus temporal-code stability.2 3
  • Firing-rate (slow, hundreds of ms to 1–5 s) tuning drifted day to day, with stability correlated to each neuron’s tuning reliability; adding temporal codes (fast, tens-of-ms spiking structure) increased single-neuron tuning stability, most of all for the least reliable neurons.2
  • Temporal coding improved population-level discriminability and raised stimulus decoding accuracy on future days by 0.0178 ± 0.0002 (paired t-test P < 1.3 × 10⁻⁷⁷).2
  • Temporal-code stability tracked network functional connectivity more closely than rate-code stability, suggesting relative spike-timing relations rather than mean firing rate carry the durable representation.2 3
  • A related Cell Reports study using SpikeShip (an optimal-transport measure of relative spike timing) across six visual areas found spike sequences encode natural videos more precisely and remain stable while firing-rate vectors drift within and between sessions.4
  • For BCI decoding, these results imply timing-based features can reduce representational drift and cut recalibration burden versus rate-only decoders, though the evidence is from mouse visual cortex rather than human motor/BCI recordings.2 4
  • The dataset supporting the paper is openly archived on figshare, enabling independent reanalysis of the multi-day recordings.5

Footnotes

  1. https://news.google.com/rss/articles/CBMiX0FVX3lxTE9ZOF9Rb0FjUWdpamo5YkdBY3lhUUlrMnZ4ZU8zR3N3bm94M096d3Q3RlNPcVZJSUZWUHpaTy1VOWxFY2JUNkZPR21mT0ZTaWxRc1RmQXF6T2FTZTdpcHRJ?oc=5 ↩ ↩2 ↩3 ↩4

  2. https://www.nature.com/articles/s41467-025-62069-2 ↩ ↩2 ↩3 ↩4 ↩5 ↩6

  3. https://doi.org/10.1101/2025.05.13.652528 ↩ ↩2

  4. https://www.sciencedirect.com/science/article/pii/S2211124725003183 ↩ ↩2

  5. https://doi.org/10.6084/m9.figshare.28877813 ↩

Graph View