Thank you! We don't extrapolate our results to consider fMRI, but I've been curious about it and would love to hear if you have any thoughts about the implications.

This work wouldn’t have been possible without the support, expertise, and patience of @nicolecrust.bsky.social and @brettlfoster.bsky.social, the generosity and helpfulness of @simonbohn.bsky.social, and the support of countless others.

(10/10)

These results provide a framework for translating between spikes and LFPs, highlighting the scenarios likely to be fruitful for translation.

I call this “basic translational neuroscience” and I’m excited to continue with this approach in my research moving forward!

(9/10)

And this rule generalizes beyond visual memory!

Sorting previous studies by whether they examined magnitude or pattern-of-spikes codes demonstrates that magnitude codes have consistently been found to be aligned between spikes and LFPs, while heterogenous pattern-of-spikes codes have not.

(8/10)

We propose that it’s the neural coding scheme of the underlying spiking representation. HGA captures an average of local spikes. This increases signal for variables encoded as overall changes in local magnitude and “washes out” signals encoded as a pattern of heterogeneous responses.

(7/10)

But when we looked at the neural representations of object category, which are very strong in spiking activity, we found much weaker representations in HGA.

Why is alignment so striking for novelty, recency, and memorability, but not for category? 🤔

(6/10)

Not only were the signals well aligned, but we found that novelty signals were STRONGER in HGA than in spikes, requiring at least 4-fold less data to reached matched discriminability of novel from repeated images. In this case, you're better off with one channel of HGA than one neuron.

(5/10)

We started by examining a number of variables for which we’ve previously linked spiking neural representations to visual memory behavior: novelty, recency, and memorability.

For all three variables, we found a strong correspondence between the signals measured in spikes and HGA.

(4/10)

Others have suggested that high-gamma activity (HGA) captures a proxy of underlying spiking activity. We found that was true of our datasets as well, where HGA consistently captured spiking activity better than other frequency bands.

(3/10)

We leveraged datasets where we've previously reported on the spiking neural representations that support visual memory to ask a simple question: would we have made the same conclusions if we’d been limited to LFPs (similar to many human intracranial experiments)?

(2/10)