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..
Contribution of task-irrelevant stimuli to drift of neural representations
Authors: Farhad Pashakhanloo
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Using both synthetic and real datasets (MNIST), we show that learning-induced drift leads to different predictions for the geometry and dimension-dependency of the drift than those caused by Gaussian synaptic noise. |
| Researcher Affiliation | Academia | Farhad Pashakhanloo Center for Brain Science Harvard University, Cambridge, MA EMAIL |
| Pseudocode | No | The paper describes algorithms and update rules using mathematical equations (e.g., Eq. 1, Eq. 8), but does not present them in a structured pseudocode or algorithm block format. |
| Open Source Code | Yes | The codes and core functions to reproduce the main results are publicly available at: https://github.com/fpashakhanloo/task-irrel-drift. |
| Open Datasets | Yes | Here, we apply the theory to MNIST data [32]. |
| Dataset Splits | No | To make the experiments more computationally manageable, we first project the MNIST data onto the top 20 principal components and use that as the input to the network (n = 20). The paper does not explicitly describe how the MNIST data itself was split into training, validation, or test sets; it implies online learning on the full dataset or a pre-processed version. |
| Hardware Specification | Yes | The computations in this paper were run on the FASRC Cannon cluster supported by the FAS Division of Science Research Computing Group at Harvard University. Experiments were executed on local institution s internal CPU cluster, with a total estimated compute time of approximately 1,000 CPU core-hours The maximum memory usage was 64GB. |
| Software Dependencies | No | The paper does not explicitly list specific software dependencies with version numbers (e.g., Python version, library versions like PyTorch, TensorFlow, or scikit-learn). |
| Experiment Setup | Yes | A one-layer network is continuously presented with input samples and updated using Oja s learning rule long after convergence (n = 50, m = 30, η = 0.025). Unless otherwise specified, for all simulations: n = 8, p = 10, m = 2, η = 0.2, γ = 0.05 and λ = 0.25. |