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..
Spike-timing-dependent Hebbian learning as noisy gradient descent
Authors: Niklas Dexheimer, Sascha Gaudlitz, Johannes Schmidt-Hieber
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Theoretical | Our main contribution lies in connecting STDP to noisy gradient descent and providing a rigorous convergence analysis of the noisy learning scheme. To this end, we introduce a learning rule for the weights w1, . . . , wd, which captures the locality and spike-time dependence of Hebbian STDP. We rewrite the learning rule as a noisy gradient descent scheme with respect to a suitable loss function. The connection to noisy gradient descent and stochastic approximation [24, 34] paves the way for applying mathematical tools from stochastic process theory to analyse the STDP rule. Our analysis of STDP is inspired by the work on noisy gradient descent for non-convex loss functions of Mertikopoulos et al. [27]. By refining their arguments and carefully tracking the error terms, we show an exponentially fast alignment of the output neuron with the input neuron of the highest mean firing rate on an event of high probability. The results of the paper are of theoretical nature. Figures 2 and 3 serve only as illustrations. |
| Researcher Affiliation | Academia | 1University of Twente 2Humboldt-Universität zu Berlin EMAIL EMAIL |
| Pseudocode | Yes | Algorithm 1: Aligning multiple output neurons Algorithm 2: Sequential alignment of multiple output neurons |
| Open Source Code | Yes | The code used to generate the illustrations in both Figures is included in the submission and will be made publicly available. |
| Open Datasets | No | We considered input spike trains generated from Poisson point processes with fixed intensity. ... Figures 2 and 3 serve only as illustrations. The results of the paper are of theoretical nature. ... The paper does not include experiments. |
| Dataset Splits | No | The results of the paper are of theoretical nature. Figures 2 and 3 serve only as illustrations. The answer NA means that the paper does not include experiments. |
| Hardware Specification | No | The results of the paper are of theoretical nature. Figures 2 and 3 serve only as illustrations. The answer NA means that the paper does not include experiments. |
| Software Dependencies | No | The results of the paper are of theoretical nature. Figures 2 and 3 serve only as illustrations. The answer NA means that the paper does not include experiments. |
| Experiment Setup | Yes | All trajectories are simulated with 2000 iteration steps, learning rate α = 0.01 and Z(k) Unif([ -1, 1]d). Middle & right: The Frobenius error P(k) I 2/2 of 100 trajectories with learning rates 10 -3(1, 0.75, 0.5) and 10 -4(1, 0.75, 0.5) , respectively. |