# Paper authors introduce Neural-DISCO for counterfactual editing of neural population activity [Canonical reading](https://escapemeridian.com/readings/paper-authors-introduce-neural-disco-for-counterfactual-editing-4631cf6a) The authors introduce Neural-DISCO, a framework for modeling how neural activity might change when one measured behavioral or stimulus label is edited, while retaining other observed labels and source-specific variability. They argue that this goes beyond predicting neural activity from labels: the model starts from a recorded response and edits one label to generate a counterfactual response. Neural-DISCO divides its latent space into components associated with individual observed labels and a residual latent for variability not supplied by those labels. The authors report that, across simulated and real neural recording datasets, this structure enables selective editing while preserving other content in the original activity. They also pair the framework with feature attribution methods to identify neurons important for each label, and report improved recovery of known functional cell roles in synthetic and real data. The approach could give researchers a way to probe how neural populations encode behavior and sensory information, but the abstract does not establish how well its counterfactual responses match changes in actual neural activity. What evidence in the full paper tests whether these edited responses remain accurate when labels or recording conditions differ from those used to build the model? Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims. ## Original source arXiv q-bio.NC · Article Original source date: Oct 23, 2026 (2026-10-23). Date basis: Date basis not provided. Added to Escape Meridian: 2026-11-05T07:17:28.793Z. First observed: 2026-10-27T04:15:06.426Z. Retrieved: 2026-10-29T20:31:49.277Z. - [Neural-DISCO: Source-Conditioned Counterfactual Editing of Neural Population Activity](https://arxiv.org/abs/2610.33145) ## Limits The supplied evidence is the paper's abstract; it does not provide quantitative results, evaluation details, or evidence about performance outside the simulated and real datasets described there. The abstract reports generated counterfactual responses but does not establish whether they match neural activity under experimentally observed changes in labels. ## Open question What evidence in the full paper tests whether edited responses remain accurate when labels or recording conditions differ from those used to build the model? ## Source-stated The authors present Neural-DISCO as a framework for counterfactual modeling of neural activity that edits a measured label in a recorded response while preserving other observed labels and source-specific variability. > given a recorded source response, we should be able to edit one measured behavioral or stimulus label and predict how neural activity would change while preserving other observed labels and source-specific variability not explained by the labels. - [Supporting source](https://arxiv.org/abs/2610.33145) ## Source-stated Neural-DISCO partitions its latent space into label-specific components and a residual latent for variability not supplied through observed labels. > Neural-DISCO partitions the latent space into label-specific components, each associated with an observed behavioral or stimulus label, together with a residual latent that captures variability not supplied through observed labels. - [Supporting source](https://arxiv.org/abs/2610.33145) ## Source-stated The authors report selective editing with preserved remaining activity content across simulated and real neural recording datasets. > Across simulated and real neural recording datasets, we show that this structure enables selective editing of measured labels to generate counterfactual neural responses while preserving the remaining content of the original activity. - [Supporting source](https://arxiv.org/abs/2610.33145) ## Source-stated The authors report that disentanglement enhances recovery of known functional cell roles in synthetic and real data. > We find that disentanglement enhances the recovery of known functional cell roles in both synthetic and real data. - [Supporting source](https://arxiv.org/abs/2610.33145) ## Source and contributor context Policzer, Nico, Ho, James, Soo, Joseph, Tang, Xin · Article author These are credits stated by the source; they do not establish a profile match or authorship of every passage. ## Source media No suitable media was found by the supported page extraction. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## Counterfactual An argument or scenario that considers what would have resulted if events had gone differently from how they actually did, such as if a policy had never been adopted; as an adjective, relating to this kind of thinking about what did not happen but could have. - [Definition reference](https://dictionary.cambridge.org/dictionary/english/counterfactual) ## arXiv arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed. - [Explore this context](/entities/arxiv) - [Reference](https://arxiv.org/) - [Reference](https://info.arxiv.org/help/api/index.html) ## Editorial provenance Version 1. Publication ID: 6b44cf0a-0bd0-4b20-8471-6e104ab2a543. Prepared by gpt-6-luna; grounding review by gpt-6-luna. Source check: passed by anthropic/claude-opus-5.5 on 2026-11-05T07:17:28.793Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: material\_development. Observation: aea67f21890e933cd6d74c4851cd01bba47fc3b2617ab2826442f50d62015439. Source text hash: bfb04a7e0c3553bc4c4a659a14ce6debba0b3f4c615e1572a2ef3f8d2e7beea6. ## Revision and source-check history ### Version 1 — 2026-11-05T07:17:28.793Z The authors introduce Neural-DISCO, a framework for modeling how neural activity might change when one measured behavioral or stimulus label is edited, while retaining other observed labels and source-specific variability. They argue that this goes beyond predicting neural activity from labels: the model starts from a recorded response and edits one label to generate a counterfactual response. Neural-DISCO divides its latent space into components associated with individual observed labels and a residual latent for variability not supplied by those labels. The authors report that, across simulated and real neural recording datasets, this structure enables selective editing while preserving other content in the original activity. They also pair the framework with feature attribution methods to identify neurons important for each label, and report improved recovery of known functional cell roles in synthetic and real data. The approach could give researchers a way to probe how neural populations encode behavior and sensory information, but the abstract does not establish how well its counterfactual responses match changes in actual neural activity. What evidence in the full paper tests whether these edited responses remain accurate when labels or recording conditions differ from those used to build the model? ## Original source arXiv q-bio.NC · Article Original source date: Oct 23, 2026 (2026-10-23). Date basis: Date basis not provided. Added to Escape Meridian: 2026-11-05T07:17:28.793Z. First observed: 2026-10-27T04:15:06.426Z. Retrieved: 2026-10-29T20:31:49.277Z. - [Neural-DISCO: Source-Conditioned Counterfactual Editing of Neural Population Activity](https://arxiv.org/abs/2610.33145) ## Limits The supplied evidence is the paper's abstract; it does not provide quantitative results, evaluation details, or evidence about performance outside the simulated and real datasets described there. The abstract reports generated counterfactual responses but does not establish whether they match neural activity under experimentally observed changes in labels. ## Open question What evidence in the full paper tests whether edited responses remain accurate when labels or recording conditions differ from those used to build the model? ## Source-stated The authors present Neural-DISCO as a framework for counterfactual modeling of neural activity that edits a measured label in a recorded response while preserving other observed labels and source-specific variability. > given a recorded source response, we should be able to edit one measured behavioral or stimulus label and predict how neural activity would change while preserving other observed labels and source-specific variability not explained by the labels. - [Supporting source](https://arxiv.org/abs/2610.33145) ## Source-stated Neural-DISCO partitions its latent space into label-specific components and a residual latent for variability not supplied through observed labels. > Neural-DISCO partitions the latent space into label-specific components, each associated with an observed behavioral or stimulus label, together with a residual latent that captures variability not supplied through observed labels. - [Supporting source](https://arxiv.org/abs/2610.33145) ## Source-stated The authors report selective editing with preserved remaining activity content across simulated and real neural recording datasets. > Across simulated and real neural recording datasets, we show that this structure enables selective editing of measured labels to generate counterfactual neural responses while preserving the remaining content of the original activity. - [Supporting source](https://arxiv.org/abs/2610.33145) ## Source-stated The authors report that disentanglement enhances recovery of known functional cell roles in synthetic and real data. > We find that disentanglement enhances the recovery of known functional cell roles in both synthetic and real data. - [Supporting source](https://arxiv.org/abs/2610.33145) ## Source and contributor context Policzer, Nico, Ho, James, Soo, Joseph, Tang, Xin · Article author These are credits stated by the source; they do not establish a profile match or authorship of every passage. ## Source media No suitable media was found by the supported page extraction. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## Counterfactual An argument or scenario that considers what would have resulted if events had gone differently from how they actually did, such as if a policy had never been adopted; as an adjective, relating to this kind of thinking about what did not happen but could have. - [Definition reference](https://dictionary.cambridge.org/dictionary/english/counterfactual) ## arXiv arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed. - [Explore this context](/entities/arxiv) - [Reference](https://arxiv.org/) - [Reference](https://info.arxiv.org/help/api/index.html) ## Editorial provenance Version 1. Publication ID: 6b44cf0a-0bd0-4b20-8471-6e104ab2a543. Prepared by gpt-6-luna; grounding review by gpt-6-luna. Source check: passed by anthropic/claude-opus-5.5 on 2026-11-05T07:17:28.793Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: material\_development. Observation: aea67f21890e933cd6d74c4851cd01bba47fc3b2617ab2826442f50d62015439. Source text hash: bfb04a7e0c3553bc4c4a659a14ce6debba0b3f4c615e1572a2ef3f8d2e7beea6. ### passed — 2026-11-05T07:17:28.793Z — anthropic/claude-opus-5.5 Publication: 6b44cf0a-0bd0-4b20-8471-6e104ab2a543 attribution — pass: The reading consistently attributes Neural-DISCO and its reported results to the paper's authors ('The authors introduce', 'The authors report'). This matches the arXiv abstract by Policzer, Ho, Soo and Tang. Source claims are not presented as established facts. > Authors:Nico Policzer, James Ho, Joseph Soo, Xin Tang summary — pass: The title, summary and question faithfully convey the abstract. It covers counterfactual editing from a recorded response, the latent partition with a residual, results on simulated and real data, and feature attribution for cell roles. The editing idea is explained in plain words. Technical terms such as latent space, disentanglement and feature attribution are kept exactly, consistent with tooltip handling. The question follows from the abstract's gap and does not lead the reader. > Neural-DISCO partitions the latent space into label-specific components, each associated with an observed behavioral or stimulus label date — pass: The originalPublished date of 2026-10-23 matches the arXiv v1 submission date. It is not the later feed date of 2026-10-27. > \[Submitted on 23 Oct 2026\] support — pass: Each claim is backed by a verbatim abstract excerpt and labeled as source-stated. The conditional significance, that it 'could give researchers a way to probe' encoding, mirrors the authors' own framing. The stated limitations accurately note that the abstract gives no quantitative validation against observed neural changes. No benefits or costs are invented. > provide a practical interface for probing how neural populations encode behavior and sensory information media — pass: The selected media is the verified direct arXiv abstract page for this paper. It is the original source and the most direct access point. ## Recorded references Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter. No public references recorded here yet. [Reference data (JSON)](/readings/4631cf6a6f8b07d9f2a5a72ba500887385bed976767c046a300878273b1cb6d5/references.json) ## Corrections and challenges [Send a correction or an opposing source](https://escapemeridian.com/corrections/) [JSON record](https://escapemeridian.com/readings/paper-authors-introduce-neural-disco-for-counterfactual-editing-4631cf6a.json)