Cogan Lab

Come by tomorrow morning to hear about verbal working memory!

Stop by this afternoon to see some intracranial speech decoding in the hippocampus and to say hello!

Saturday Sept. 13 11am-12:30pm, Poster Session C

C54: Baishen Liang (Postdoctoral Associate) will be presenting his work on sensory-motor mechanisms for verbal working memory.

Hope to see you all there!

Friday Sept 12 4:30pm-6:00pm, Poster Session B

B70: Yuchao Wang (Rotation CNAP PhD Student) will be presenting his work on auditory pseudoword decoding in the hippocampus.

Coming to DC for SNL later this week?

Come check out our posters on speech decoding and verbal working memory using intracranial recordings!

@snlmtg.bsky.social
#SNL2025

❔3️⃣: In Figs. 4 and 5, do you obtain similar results if you operate directly on the spike trains instead of on the PCA-reduced spike trains? Why is PCA necessary first?

Thank you to the authors for your work!
cc: Alexis Arnaudon, Mauricio Barahona, Pierre Vandergheynst

If separate animals were treated as separate manifolds with an embedding-agnostic MARBLE, would you still expect an informative latent space to be learned without any need for post-hoc alignment?

❔2️⃣: It seems that a linear transformation between MARBLE representations of different animals was necessary because the same information is present in the latent space but not necessarily with the same ordering... (con't)

❔1️⃣: It is stated that non-neighbors (both within and across manifolds) are negative samples (mapped far) during the contrastive learning step. Does treating non-neighbors within and across manifolds as similarly “distant” lead to less interpretability of larger distances in latent space?

🤍3️⃣: The comparisons to state-of-the-art latent dynamical systems models are great for properly contextualizing the performance of MARBLE.

🤍1️⃣: The initial proximity graph is a clever way to define distance and neighborhoods between inputs that can be used for downstream training.
🤍2️⃣: The rotation invariance is important and likely useful for extracting shared latent representations from systems with minor differences.

MARBLE: interpretable representations of neural population dynamics using geometric deep learning - Nature Methods MARBLE uses geometric deep learning to map dynamics such as neural activity into a latent representation, which can then be used to decode the neural activity or compare it across systems.

They find that MARBLE successfully decomposes complex dynamical activity from spike trains into informative and easily decodable latent representations. This 🧵 explores our thoughts (🤍 & ❔). www.nature.com/articles/s41...

MARBLE: interpretable representations of neural population dynamics using geometric deep learning - Nature Methods MARBLE uses geometric deep learning to map dynamics such as neural activity into a latent representation, which can then be used to decode the neural activity or compare it across systems.

Last week, Zac Spalding (@zspald.bsky.social‬, 4th year PhD student, @dukeubme.bsky.social‬) presented Adam Gosztolai, Robert Peach, and colleagues’ 2025 paper on MARBLE, a method for finding interpretable latent representations of neural dynamics.

Exciting work by PhD student Zac Spalding @zspald.bsky.social and multiple labs across departments at Duke! A combination of human electrophysiology, modeling/decoding, and cognitive concepts to improve BCI for speech in patients. 👏 🧠📈

From Cogan Lab Journal Club with @zspald.bsky.social
these decomposition acronyms are getting out of hand!

Aligned neural space for speech production allows for improved decoding in micro-ECOG patients for BCI

We’re happy to present @zspald.bsky.social 's work on shared neural representations of speech production across individuals! We find that patient-specific data can be aligned to a shared space that preserves speech information, enabling cross-patient speech BCIs.
www.biorxiv.org/content/10.1...

❔3️⃣: Have the authors performed unique syllable-type decoding in other regions? Which regions might be responsible for more detailed/specific motor sequence planning?

Thank you to the authors @changlabucsf.bsky.social for your great work, and we look forward to following more of it!

❔2️⃣: Given that the specific content of a sequence from pre-speech activity can not be reliably decoded, how do you interpret the nature of the motor plan in the mPrCG?

❔1️⃣: Why do you think the sustained activity in the vPrCG is modulated by articulatory complexity, yet it is unable to predict reaction time (RT)?

🤍3️⃣: The paper does an excellent job of connecting a basic science finding to a real-world clinical disorder. The demonstration that mPrCG stimulation produces errors that are “consistent with those made in AOS, offering significant insights for both researchers and clinicians.

🤍2️⃣: I also like the clear narrative structure: identifying a network → defining its function → linking it to behavior → providing causal evidence.

🤍1️⃣: I appreciate that they were able to establish a causal role for the mPrCG through electrical stimulation.

Speech sequencing in the human precentral gyrus - Nature Human Behaviour Liu et al. examine the role of sustained neural activity in the planning and production of speech sequences, revealing a key role for the middle precentral gyrus.

Last week, Nanlin Shi (Postdoctoral Associate) presented Jessie Liu and colleague’s latest work on speech encoding in the middle Precentral Gyrus (mPrCG). This 🧵 explores our thoughts (🤍 & ❔)
www.nature.com/articles/s41...

Thank you to the authors at UC Berkeley Psychology and Neuroscience for your work, and we look forward to following more of it!

cc: @ucberkeleyofficial.bsky.social

❔3️⃣: In Fig. 2C, it looks like MFG has early low-frequency activity during the approach period that dissipates about 500ms before the decision to turn around. Could this be preparatory activity that reflects participants’ plans for the trial, or predictive of how many dots they decide to eat?

[cont.] I’m curious if there is a sharp rise in beta coherence between the STN and limbic regions right before the decision to turn around, and using Granger Causality, could it be shown that the STN drives activity in limbic regions during this brief window before the decision?

❔2️⃣: The study didn’t probe the role of subcortical beta during approach-avoidance conflict, perhaps due to limitations on recording sites. I wonder if STN beta activity may play a distinct role in this circuit as well, particularly as a signal to halt and turn around. [...]

[cont.] the high-frequency activity in ACC would reflect similar patterns as those in MFG? What exactly is the role of ACC during imminent threat? Might it serve a prediction role, as it does during other types of tasks?

❔1️⃣: The ACC (and MFC more broadly) is also construed as a conflict-control region. I’m surprised it didn’t play a larger role in regulating the other regions during ghost attack trials. Perhaps, if ghost attacks occurred more randomly, instead of based on distance...

🤍3️⃣: This experiment is well-suited for the recording locations that they had access to, and nicely contrasts the roles of limbic regions and MFG under anxiety and threat.