Efficient coding, channel capacity, and the emergence of retinal mosaics

Authors: Na Young Jun, Greg Field, John Pearson

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Here, we use efficient coding theory to present a comprehensive account of mosaic organization in the case of natural videos as the retinal channel capacity the number of simulated RGCs available for encoding is varied. We analyzed the characteristics of the optimal spatiotemporal RFs obtained from the model (2, 3) trained on videos from the Chicago Motion Database [22].
Researcher Affiliation Academia Na Young Jun Department of Neurobiology Duke University Durham, NC 27710 nayoung.jun@duke.edu Greg D. Field Department of Neurobiology Duke University Durham, NC 27710 field@neuro.duke.edu John M. Pearson Department of Biostatistics & Bioinformatics Department of Neurobiology Department of Electrical and Computer Engineering Duke University Durham, NC 27710 john.pearson@duke.edu
Pseudocode No The paper contains mathematical formulations and descriptions of the model and experimental procedures, but it does not include any pseudocode blocks or sections explicitly labeled as an algorithm.
Open Source Code Yes All model code and reproducible examples are available at https://github.com/pearsonlab/efficientcoding.
Open Datasets Yes We analyzed the characteristics of the optimal spatiotemporal RFs obtained from the model (2, 3) trained on videos from the Chicago Motion Database [22].
Dataset Splits No The paper mentions training on video data and dividing the dataset into subsets, but it does not explicitly provide details about train/validation/test splits (e.g., percentages or sample counts). The checklist also states: "our work did not include data splits".
Hardware Specification No The provided text discusses
Software Dependencies No The paper mentions using "Adam [23]" for optimization, but it does not specify software versions for any key libraries, frameworks, or programming languages (e.g., Python, PyTorch, TensorFlow versions).
Experiment Setup Yes Model parameters for spatial kernels, temporal kernels, and the nonlinearities were jointly optimized using Adam [23] to maximize (2) subject to the mean firing rate constraint (3) using the augmented Lagrangian method with the quadratic penalty = 1 [24]. We used RF size D = 82 pixels for Slow and D = 122 for Fast A and Fast B cell types to allow the size of spatial kernels to be similar to those of the previous experiments, and we imposed the additional constraint that the shape parameters aj, bj, and cj in (8) be shared across RGCs.