Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
High-dimensional neuronal activity from low-dimensional latent dynamics: a solvable model
Authors: Valentin Schmutz, Ali Haydaroğlu, Shuqi Wang, Yixiao Feng, Matteo Carandini, Kenneth D Harris
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | To uncover the pre-activation dimension of high-dimensional activity in visual cortex, we perform two-photon calcium recordings of tens of thousands of neurons from mouse visual cortex, and infer the pre-activation dimension using the Neural Cross-Encoder (NCE), an interpretable, nonlinear latent variable modeling method which models the activity of each neuron as a simple linear-nonlinear readout of low-dimensional latents. |
| Researcher Affiliation | Academia | Valentin Schmutz University College London WC1E 6BT London, UK EMAIL, Ali Haydaroglu University College London WC1E 6BT London, UK EMAIL, Shuqi Wang École Polytechnique Fédérale de Lausanne 1015 Lausanne, Switzerland EMAIL, Yixiao Feng Shanghai Jiao Tong University 200240 Shanghai, China EMAIL, Matteo Carandini University College London WC1E 6BT London, UK EMAIL, Kenneth D. Harris University College London WC1E 6BT London, UK EMAIL |
| Pseudocode | No | The paper describes methods and derivations in narrative text and mathematical equations throughout the main body and appendices, but it does not include any explicitly labeled 'Pseudocode' or 'Algorithm' blocks or figures. |
| Open Source Code | Yes | All data and code used to train the NCE and generate the results, as well as code used for simulations in Fig. 2, are shared via Figshare: https://figshare.com/s/aac82eac1829b7ec406e and Github: https://github.com/alihaydaroglu/NCE. |
| Open Datasets | Yes | All data and code used to train the NCE and generate the results, as well as code used for simulations in Fig. 2, are shared via Figshare: https://figshare.com/s/aac82eac1829b7ec406e and Github: https://github.com/alihaydaroglu/NCE. Preprocessed neural datasets, comprised of deconvolved spontaneous activity and averaged, deconvolved stimulus responses, is also anonymized and submitted with the supplementary materials. |
| Dataset Splits | Yes | In all three conditions, the paired source-target datasets were split into train-validation-test sets with a 50%-20%-30% split. |
| Hardware Specification | Yes | Volumetric calcium movies were processed using a custom workstation running Ubuntu 20.04.6 LTS with an Intel(R) Xeon(R) w9-3475X CPU (36 cores, 2.20 GHz), NVIDIA RTX A4500 GPU (20GB VRAM), and 512 GB DDR5 RAM. ... All NCE experiments were performed on a workstation running Windows 10 Pro, with an Intel (R) Core(TM) i7-11700k CPU (8 cores, 3.6 GHz), NVIDIA RTX 3060 GPU (12GB VRAM), and 128 GB DDR4 RAM. |
| Software Dependencies | No | This work makes use of several publicly available software resources including: Python (PSF License), Matplotlib (PSF License), NumPy (BSD License), SciPy (BSD License), CuPy (MIT License), PyTorch (BSD-3 License), Suite3D (AGPL-3 License). |
| Experiment Setup | Yes | parameters were optimized through stochastic gradient descent using Adam [96] and a batch size of 2,048, with a learning rate of 10 3, momentum parameters β1 = 0.85, β2 = 0.95, and an L2 penalty of 10 5. Model parameters were randomly initialized with Kaiming initialization [97]. |