Choosing a first model
A bounded thalamocortical question, a larger visual-cortex reference, and the files we still need to inspect.
Update, 8 October 2026: the completed source audit changes the next execution target to the small LSV1M regression model. The entry below preserves our initial candidate selection and the questions we had before inspecting the files.
Our first model should answer a specific physiological question and offer a realistic route to reproduction. The choice is still provisional: a promising paper and a dataset listing are not enough to establish that a complete implementation can be run.
The smaller candidate
Madineh Sedigh-Sarvestani, Larry A. Palmer and Diego Contreras studied transmission between the lateral geniculate nucleus and primary visual cortex in cats. Their 2019 paper examines variable, unreliable synaptic transmission and uses biologically constrained models to investigate its computational consequences. Read the paper.
This is a useful candidate for Felisyn because it offers a bounded mechanism rather than an entire behavioural system. That assessment is our selection judgment; we have not yet tested the implementation.
The associated Dryad record lists a README and a 1.10 GB archive. We have checked the record, but have not inspected the archive. We still need to determine which files are recordings, analysis code, simulation code or reference outputs.
A larger reference
Antolík and colleagues’ 2024 paper, A comprehensive data-driven model of cat primary visual cortex, provides a larger reference. It models a cortical patch, including layers 4 and 2/3, under anatomical and functional constraints.
Its connections are model constructions constrained by evidence; they should not be presented as a complete set of individually traced connections from one cat. The scale and implementation make it a substantial reproduction task. We have not run it.
We will revisit this reference if the smaller candidate cannot support a sufficiently explicit comparison. Choosing a smaller target first is about making the first experiment interpretable, not assuming that either model is already suitable for a controller.
Questions for the package audit
The audit must establish:
- Whether the available code includes the simulation itself, the analysis, or both.
- Which software versions and numerical settings the original result used.
- Whether parameters, stimuli and initial conditions are specified completely enough to reproduce it.
- Which outputs are available as a reference, and how their units and experimental conditions are described.
- What reuse terms apply to the code and data separately.
If essential code is absent, we will record the gap. An implementation based on published equations would be labelled as our own implementation and evaluated against the available reference evidence.
Current position
The candidate sources are identified. Package inspection, environment setup, benchmark selection and execution remain pending. We have no measured runtime, memory requirement or reproduction accuracy for either candidate.
Next step: preserve the source files and their checksums, resolve the package contents and choose one quantitative result to reproduce.
References
- Sedigh-Sarvestani, M., Palmer, L. A. & Contreras, D. (2019). Thalamocortical synapses in the cat visual system in vivo are weak and unreliable. eLife. doi:10.7554/eLife.41925.
- Sedigh-Sarvestani, M., Palmer, L. A. & Contreras, D. (2019). Associated dataset. Dryad. doi:10.5061/dryad.57pv818.
- Antolík et al. (2024). A comprehensive data-driven model of cat primary visual cortex. Open-access article.