Sign in

Atanas Stankov

@neuronasko.bsky.social
346 followers 807 following 1.7K posts

CUNY neuro PhD. UMich[visiting student]. Prev at /Stanford/ETH Zurich & UZH/UIUC. Monist soft-determinism. Opinions are my own. Open science: democratize and fund science bottom-up, not just top-down. Neuro needs more brains doing the job à la Solar Punk.

PostsRepliesMedia
Atanas Stankov @neuronasko.bsky.social · 5h
And this team that did pull it off for the cerebellum had 23 authors. It worked across different probes because they used a single waveform and not a spatial template of activity. www.cell.com/cell/fulltex...
cell.com
A deep learning strategy to identify cell types across species from high-density extracellular recordings
We constructed a deep learning classifier that solves the problem of neuron-type identification from simultaneous extracellular recordings of many neurons. The classifier was trained on a ground-truth...
000
Atanas Stankov @neuronasko.bsky.social · 6h
Even with the availability of high resolution probe data like Neuropixels Ultra doing such biophysical model fits is at the moment extremely computationally expensive and tricky.
100
Atanas Stankov @neuronasko.bsky.social · 6h
One strategy towards developing a standard template is a biophysics layer to the NN that can extract biophysical parameters via neuron model fit like this pre-print: "Identifying multi-compartment Hodgkin-Huxley models with high-density extracellular voltage recordings" arxiv.org/abs/2506.202...
arxiv.org
Identifying multi-compartment Hodgkin-Huxley models with high-density extracellular voltage recordings
Multi-compartment Hodgkin-Huxley models are biophysical models of how electrical signals propagate throughout a neuron, and they form the basis of our knowledge of neural computation at the cellular l...
100
Atanas Stankov @neuronasko.bsky.social · 6h
Let's discuss cell typing classifiers. I want to use deep learning for cell typing and for tracking waveform evolution of single cells over the duration of recording. A first step would have be come up with a probe-neutral waveform template to deal with different large-scale recording datasets.
100
Atanas Stankov @neuronasko.bsky.social · 10h
Lula is 80 years old running for a 4th presidential term. What in the heck is happening when a political system cannot produce young progressive candidates?
000
Atanas Stankov @neuronasko.bsky.social · 16h
Hypothesis: Studying engrams with optogenetics is theoretically tricky because engram cells are intrinsically driven. If you perturb the cell using stimuli it is not accustomed to, like its network input, intracellular memory recall is reset.
110
Atanas Stankov @neuronasko.bsky.social · 16h
If one is a computational neuroscientist and AI took their job, they were never doing neuroscience to begin with.
010
Atanas Stankov @neuronasko.bsky.social · 05/10/2026
Got another one for you. Lighting sprites, super rare form of red lightening, look like cortical pyramidal cells. www.nasa.gov/general/spri...
nasa.gov
Spritacular: NASA’s New Citizen Science Project to Capture Elusive Upper Atmospheric Electrical Phenomena on Camera - NASA
A flash of lightning, and then – something else. High above the storm, a crimson figure blinks in and out of existence. If you saw it, you are a lucky witness
020
Atanas Stankov @neuronasko.bsky.social · 05/10/2026
Applied trolleyism?
010
Atanas Stankov @neuronasko.bsky.social · 04/10/2026
(4) Continued: ~7100 cortical units Kilosort sorted from anesthetized 20 rats. 30min in-vivo recordings.
100
Atanas Stankov @neuronasko.bsky.social · 04/10/2026
(4) They use a custom high spatial resolution probe in the spirit of Neuropixels Ultra albeit more course. ~1TB with raw data available. "Dataset of cortical activity recorded with high spatial resolution from anesthetized rats" www.nature.com/articles/s41... doi.gin.g-node.org/10.12751/g-n...
doi.gin.g-node.org
G-Node Open Data: Dataset of cortical activity recorded with high spatial resolution from anesthetized rats
100
Atanas Stankov @neuronasko.bsky.social · 04/10/2026
If I am going to embody backtracking via public discourse I will occasionally link more recent posts to older ones. David Lewis does touch upon shared memories their causal impact on communication. bsky.app/profile/neur...
000
Atanas Stankov @neuronasko.bsky.social · 04/10/2026
Also your post reminded of this paper I ran into a few weeks ago on the topic of ephys and emergence. Just sharing for the vibes. royalsocietypublishing.org/rstb/article...
royalsocietypublishing.org
From damage response to action potentials: early evolution of neural and contractile modules in stem eukaryotes
Abstract. Eukaryotic cells convert external stimuli into membrane depolarization, which in turn triggers effector responses such as secretion and contracti
130
Atanas Stankov @neuronasko.bsky.social · 04/10/2026
I once called @gabymichel.bsky.social a nerd for spotting said neurons/dendrite morphologies at the UZH Irchel campus icy ponds. Jokes on me I guess since now there is a putative connection.
120
Atanas Stankov @neuronasko.bsky.social · 04/10/2026
(3) Buzsaki Lab databases continued. This paper (www.science.org/doi/10.1126/...) in particular had up 500 neurons in one recording! These datasets used in that project (links from the supplementary material): dandiarchive.org/dandiset/000... buzsakilab.nyumc.org/datasets/Gro...
winnieyangwannan.github.io
Paper | 🧠🏷️ Selection of experience for memory by hippocampal sharp wave ripples
A simple, whitespace theme for academics. Based on [*folio](https://github.com/bogoli/-folio) design.
100
Atanas Stankov @neuronasko.bsky.social · 04/10/2026
(3) Buzsaki Lab databases (buzsakilab.com/wp/projects/ or buzsakilab.com/wp/datasets/). Some items on are repeats of the other datasets I mentioned previously.
100
Atanas Stankov @neuronasko.bsky.social · 04/10/2026
(2) Steinmetz Lab Neuropixels Ultra for all experiments (figshare.com/articles/dat...) and one experiment with the raw data available (github.com/jenniferColo...). Sidenote: For anyone that wants gorgeous ephys data for art's sake take a look.
figshare.com
Detailed neuronal waveforms across brain regions recorded with Neuropixels 'Ultra'
Spatiotemporal waveforms of ~6,000 neurons recorded across cortex, hippocampus, thalamus, striatum, and midbrain areas in awake mice, along with spike times.Associated with Ye, Shelton, et al. (2025) Neuron, in press. (preprint here)
100
Atanas Stankov @neuronasko.bsky.social · 04/10/2026
(1) Allen Institute Visual Coding dataset continued: the raw Neuropixels data is available via AWS but one has to pay for the data transfer. Plus it is over 80TB of data! github.com/AllenInstitu...
github.com
Use the Allen Brain Observatory – Visual Coding on AWS
code for reading and processing Allen Institute for Brain Science data - AllenInstitute/AllenSDK
100
Atanas Stankov @neuronasko.bsky.social · 04/10/2026
Public datasets to get cortical neuron waveforms to build a ephys waveform evolution tracker: (1) Allen Institute Visual Coding dataset (brain-map.org/our-research...) which can be accessed via Python (allensdk.readthedocs.io/en/latest/vi...) or MATLAB (www.mathworks.com/matlabcentra...) 🧵
brain-map.org
Ophys/Ephys Calibration Data
Optical physiology and electrophysiology calibration datasets. Validate calcium imaging with simultaneous electrical recordings for accurate analysis.
100
Atanas Stankov @neuronasko.bsky.social · 04/10/2026
And how was this connected to the idea that compatibilitsm being a poorly framed angle to operationalizing the study of free will? Well the counterfactual of “would have been” bothers me in the context of failure of critical collaborations bc fundamentally my choices are limited by other scientists.
000
Atanas Stankov @neuronasko.bsky.social · 04/10/2026
And now I find myself reading about modified Lewisian signaling systems which is something missed in neurosci comm bc the neurosci community doesn’t have models of intrinsic causality leading to one battle after between neurophysiologists and neurostatisticians. petergodfreysmith.com/wp-content/u...
petergodfreysmith.com
100
Atanas Stankov @neuronasko.bsky.social · 04/10/2026
Why is Pubpeer not succeeding as it should?
000
Atanas Stankov @neuronasko.bsky.social · 03/10/2026
The thing about such thought experiments is to create more thought experiments until you are satisfied with something empirically tractable. And then repeat the same process with the experimentation until it becomes second nature. Once such second nature is developed you might surprise yourself.
011
Atanas Stankov @neuronasko.bsky.social · 03/10/2026
Understanding of free will requires backtracking analysis. How do you even know your options of actual possibilities if you have no reference point of how your trajectory of life has evolved?
111
Reposted by Atanas Stankov
Kevin Mitchell @wiringthebrain.bsky.social · 18/09/2026
Like many people (though not enough), I regard compatibilism as a hopeless fudge, but it's also moot, because the universe sure doesn't seem to be deterministic 😊 link.springer.com/article/10.1...
386
Atanas Stankov @neuronasko.bsky.social · 03/10/2026
Both Kevin Mitchell, a neurogeneticist, and David Lewis, a 20th century analytic philosopher, have made arguments against soft-determinist free will for categorical (lack of a better word) free will. Both are monists and both come from different causality angles. I don’t understand the position yet.
100
Atanas Stankov @neuronasko.bsky.social · 03/10/2026
Serious question: Would an automated engram event detector of spike train data of spontaneous activity, including large recordings, be useful and/or meaningful? A detector that requires no direct intervention like stimulation or virus tagging. #neuroskyence
001
Atanas Stankov @neuronasko.bsky.social · 03/10/2026
I just need ONE neuronal intrinsic feature related to memory and the most common one that has come up is the action potential afterhyperpolarization (APH). What a paper! I think this is the 2nd post on it I've made. pmc.ncbi.nlm.nih.gov/articles/PMC...
pmc.ncbi.nlm.nih.gov
The role of intrinsic excitability in the evolution of memory: significance in memory allocation, consolidation, and updating
Memory is a dynamic process that is continuously regulated by both synaptic and intrinsic neural mechanisms. While numerous studies have shown that synaptic plasticity is important in various types an...
100
Atanas Stankov @neuronasko.bsky.social · 03/10/2026
Something like this but for intrinsic ephys dynamics like the AP afterpotential amongst other things. "Single cortical neurons as deep artificial neural networks" pubmed.ncbi.nlm.nih.gov/34380016/
pubmed.ncbi.nlm.nih.gov
Checking your browser - reCAPTCHA
100
Atanas Stankov @neuronasko.bsky.social · 03/10/2026
Maybe a RNN model of intrinsic neuronal stats/dynamics can be analytically sutured to the backtracking counterfactual causal model. What would that look like?
100
Atanas Stankov @neuronasko.bsky.social · 03/10/2026
The new issue revealed is an meaningful connection between intrinsic ephys data as a transnational deep learning NN to a structural causal model. It is not so much a data hunger problem.
100
Atanas Stankov @neuronasko.bsky.social · 03/10/2026
I am really hoping I do not need to train my own classifier of cortical extracell. recordings like this Maxime Beau paper for cerebellar extracell. recordings. "A deep learning strategy to identify cell types across species from high-density extracellular recordings" www.cell.com/cell/fulltex...
cell.com
A deep learning strategy to identify cell types across species from high-density extracellular recordings
We constructed a deep learning classifier that solves the problem of neuron-type identification from simultaneous extracellular recordings of many neurons. The classifier was trained on a ground-truth...
100
Atanas Stankov @neuronasko.bsky.social · 02/10/2026
Baby step. Took a look at the data. Found high temporal resolution spike times and waveforms. So far so good though I think I’ll just have to sit with the data and transfer it python somehow. One possible future issue came up and that is DL needs lots of data =/. The DeepBC code is hungry.
100
Atanas Stankov @neuronasko.bsky.social · 02/10/2026
This is useful (encoder trained on patch-seq data but tested in ephys data). “Using deep learning to quantify neuronal activation from single-cell and spatial transcriptomic data” www.nature.com/articles/s41...
nature.com
Using deep learning to quantify neuronal activation from single-cell and spatial transcriptomic data - Nature Communications
Neuronal activity is associated with transcriptional changes. Here, the authors present a deep learning model that integrates single cell transcriptomic signals to estimate neuronal activation.
100
Atanas Stankov @neuronasko.bsky.social · 02/10/2026
^^ Maybe that can be the explicit hypothesis. It changes from day to day so I just post as a note even if I repeat myself.
100
Atanas Stankov @neuronasko.bsky.social · 02/10/2026
A lot of this rests on the idea engram evolution is the neural marker of backtracking counterfactual reasoning. Basically a way to analyze the innate reasoning of closed intelligence systems by making an assumption how arrow of time changes the systems internal activity (Lewisian counterfactuals).
100
Atanas Stankov @neuronasko.bsky.social · 02/10/2026
Lastly, I realized I have been unconsciously taking a Marrian approach to link the algorithmic level (learning algo), the implementation level (ephys), and the computational level (counterfactual and behavior). Old schemas die hard. I have to give it the guy, he was a genius applicable definitions.
100
Atanas Stankov @neuronasko.bsky.social · 02/10/2026
Those two public datasets I had previously share have ephys of engram encoding (hypothetical factual) and recognition/recall (hypothetical backtracking counterfactual). An error signal can be constructed from intrinsic behavior during engram activity (ACG? firing rate? bursting activity?).
100
Atanas Stankov @neuronasko.bsky.social · 02/10/2026
I nice paper to try to operationalize engram learning error (O'Leary) to backtracking counterfactuals (Schölkopf) is (Kladny et al. 2024): "Deep Backtracking Counterfactuals for Causally Compliant Explanations" arxiv.org/abs/2310.07665 With code (in python ❤️): github.com/rudolfwillia...
arxiv.org
Deep Backtracking Counterfactuals for Causally Compliant Explanations
Counterfactuals answer questions of what would have been observed under altered circumstances and can therefore offer valuable insights. Whereas the classical interventional interpretation of counterf...
110
Atanas Stankov @neuronasko.bsky.social · 02/10/2026
More theory preamble. I wanted to further motivate my intuition about the backtracking counterfactuals (ideas of David Lewis, Bernhard Schölkopf) and learning (ideas of Timothy O'Leary) to the evolution of engram ephys (ideas of Denise Cai, various labs) and intrinsic activity (ideas of Denise Cai).
100
Atanas Stankov @neuronasko.bsky.social · 02/10/2026
This comment section and the down voting shows the disconnect in discourse. www.reddit.com/r/Professors...
reddit.com
From the Professors community on Reddit
Explore this post and more from the Professors community
000
Atanas Stankov @neuronasko.bsky.social · 02/10/2026
Data centers are proper fucked without infrastructure whose buildout has a natural rate limit that is slower than the current priced in rate of growth. m.youtube.com/watch?v=M8vA...
m.youtube.com
Building Out the Power Grid: Large Power Transformers || Peter Zeihan
YouTube video by Zeihan on Geopolitics
010
Atanas Stankov @neuronasko.bsky.social · 02/10/2026
The only thing that matters to understanding the world right now is non-stop counterfactual statements and questions. Addressing the outcomes is nice but the main goal is to make the ATTEMPT of counterfactual reasoning a more sustained conscious habit.
000
Atanas Stankov @neuronasko.bsky.social · 02/10/2026
Hypothesis: David Marr separating the algorithmic and implementation into separate entities is now an outdated understanding of brain function. #neuroskyence
130
Atanas Stankov @neuronasko.bsky.social · 02/10/2026
If David Marr was still around would denounce his three levels of analysis after seeing how AI has progressed? 🧪
000
Atanas Stankov @neuronasko.bsky.social · 02/10/2026
Edit: fixed link pubmed.ncbi.nlm.nih.gov/32512183/
pubmed.ncbi.nlm.nih.gov
Checking your browser - reCAPTCHA
000
Atanas Stankov @neuronasko.bsky.social · 02/10/2026
Engrams are distributed. Genetic expression study. No ephys. “Brain-wide mapping reveals that engrams for a single memory are distributed across multiple brain regions” www.nature.com/articles/s41...
nature.com
110
Atanas Stankov @neuronasko.bsky.social · 02/10/2026
"We have to move away from this image of the brain as a machine made of gears, with every gear having an exact purpose that we can attach a label to," No. Distributed systems are necessary but insufficient without locality. @stefanofusi.bsky.social zuckermaninstitute.columbia.edu/upending-dec...
zuckermaninstitute.columbia.edu
Upending Decades of Debate, Scientists Discover Most Neurons are Jacks-of-All-Trades
000
Atanas Stankov @neuronasko.bsky.social · 01/10/2026
And “Causes and consequences of representational drift” pmc.ncbi.nlm.nih.gov/articles/PMC...
pmc.ncbi.nlm.nih.gov
Causes and consequences of representational drift
The nervous system learns new associations while maintaining memories over long periods, exhibiting a balance between flexibility and stability. Recent experiments reveal that neuronal representations...
100
Atanas Stankov @neuronasko.bsky.social · 01/10/2026
And if representational drift is a link to backtracking counterfactual learning Timothy O’Leary’s papers are a good anchor here. “Representational drift as a window into neural and behavioural plasticity” www.sciencedirect.com/science/arti...
pmc.ncbi.nlm.nih.gov
Causes and consequences of representational drift
The nervous system learns new associations while maintaining memories over long periods, exhibiting a balance between flexibility and stability. Recent experiments reveal that neuronal representations...
101