A Differentiable Point Process with Its Application to Spiking Neural Networks

Authors: Hiroshi Kajino

ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We investigate the effectiveness of our gradient estimator through numerical simulation.
Researcher Affiliation Industry IBM Research Tokyo, Tokyo, Japan. Correspondence to: Hiroshi Kajino <kajino@jp.ibm.com>.
Pseudocode Yes Algorithm 1 Thinning algorithm for MPP Algorithm 2 Thinning algorithm for PP Algorithm 3 Generic learning algorithm
Open Source Code Yes All the experiments are conducted on IBM Cloud4, and the code is publicly available (Kajino, 2021). ... Kajino, H. diffsnn, 2021. URL https://github.com/ ibm-research-tokyo/diffsnn.
Open Datasets No Data set. We use a synthetic data set generated by the vanilla SNN (Equation (7)). ... We generate training/test sets consisting of Ntrain/100 examples of length 50 respectively.
Dataset Splits Yes We generate training/test sets consisting of Ntrain/100 examples of length 50 respectively.
Hardware Specification Yes All the experiments are conducted on IBM Cloud4, and the code is publicly available (Kajino, 2021). [Footnote 4]: Intel Xeon Gold 6248 2.50GHz 48 cores and 192GB memory.
Software Dependencies No The paper mentions using "Ada Grad (Duchi et al., 2011)" as an optimizer but does not specify version numbers for any key software components or libraries (e.g., Python, PyTorch, etc.).
Experiment Setup Yes Network size D = 6, |O| = 2, |H| = 4 Activation/filter functions a = 5, L = 2, s1 = 0, s2 = 10 PP τ = 0.3, λ = 20 # of samplings 100 (Eq. (5)), 1 (Eq. (9)). ... with initial learning rate 0.05 for 10 epochs.