Theoretically Provable Spiking Neural Networks

Authors: Shao-Qun Zhang, Zhi-Hua Zhou

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

Reproducibility Variable Result LLM Response
Research Type Experimental The aforementioned results, time complexity bound in detail, can be verified by a simulated experiment. We here simulate a 4 × 10, 000 spike sequence from the Iris data sets with a timestamp of 0.001 using Poisson encoding. We employ the one-hidden-layer sc SNN [44] that contains self-connection structure as the conducted SNN model. The number of input channels and hidden neurons are 4 and 10, respectively. The self-connection matrix V is randomly sampled from [0, 1] with bias = 1/3. The above configurations meet the conditions of Theorem 3. Table 2 lists the hyper-parameter values in the conducted sc SNN. For any linear matrix G ∈ R4×10 with ||G||2 < ∞, we define an indicator tc = Ω(n ||G||2 /||V||2). Thus, by exploiting the relation among ϵ, t, and tc, we can verify the explicit polynomial bound, especially the order of magnitude function Ω( ) in Theorem 3. Figure 2 plots the experimental results.
Researcher Affiliation Academia Shao-Qun Zhang Zhi-Hua Zhou National Key Laboratory for Novel Software Technology Nanjing University, Nanjing 210093, China {zhangsq,zhouzh}@lamda.nju.edu.cn
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code No The simulation experiments employ the codes of Zhang and Zhou [44], and detail the data, settings, and configurations in Pages 7-8. So we believe it is easily to reproduce our experimental results. We will publicash our codes as soon as possible if this paper is accepted.
Open Datasets Yes We here simulate a 4 × 10, 000 spike sequence from the Iris data sets with a timestamp of 0.001 using Poisson encoding.
Dataset Splits No The paper mentions using the Iris data set but does not provide specific training, validation, or test dataset splits. It describes generating a
Hardware Specification No The paper does not provide any specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments. It only mentions general settings for the simulation.
Software Dependencies No The paper does not provide specific software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions) needed to replicate the experiment.
Experiment Setup Yes Table 2: Hyper-parameter Setting of sc SNNs. Parameters Value Parameters Value Time Step 0.001 Firing Threshold 1 Expect Spike Count (True) 100 Membrane Time τm 0.2 Expect Spike Count (False) 10 Time Constant of Synapse τs 0.008 Encoding Length T 10, 000 Maximum Firing 10 Refractory Period 0.016 Maximum Time 10