Latent Convergent Cross Mapping

Authors: Edward De Brouwer, Adam Arany, Jaak Simm, Yves Moreau

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

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate the performance of our approach on data sets from physical and neurophysiology models, namely a double pendulum and neurons activity data. We show that our method detects the right causal topology in all cases, outperforming multi-spatial CCM, as well as baselines designed to address the sporadicity of the time series. The code is available at https: //github.com/edebrouwer/latent CCM.
Researcher Affiliation Academia Edward De Brouwer ESAT-STADIUS KU LEUVEN Leuven, 3001, Belgium edward.debrouwer@esat.kuleuven.be Adam Arany ESAT-STADIUS KU LEUVEN Leuven, 3001, Belgium adam.arany@esat.kuleuven.be Jaak Simm ESAT-STADIUS KU LEUVEN Leuven, 3001, Belgium jaak.simm@esat.kuleuven.be Yves Moreau ESAT-STADIUS KU LEUVEN Leuven, 3001, Belgium moreau@esat.kuleuven.be
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code Yes The code is available at https: //github.com/edebrouwer/latent CCM.
Open Datasets No The paper describes generating its own datasets (double pendulum, neural activity data) for evaluation but does not provide concrete access information (link, DOI, repository) for these specific datasets to be publicly available.
Dataset Splits Yes We used 80% of available windows for training and used the remaining 20% for hyperparameter tuning with the MSE prediction on future samples used as model selection criterion.
Hardware Specification No The acknowledgements section states: 'We also thank Nvidia for supporting this research by donating GPUs.' However, this does not provide specific details about the GPU models, CPU models, or other hardware specifications used for running the experiments.
Software Dependencies Yes We generate time series of the average membrane potential of two populations of leaky integrate-and-fire neurons with alpha-function shaped synaptic currents (iaf psc alpha) simulated by NEST-2.20.0 (Fardet et al., 2020). [...] Implementation was done with GPflow (Matthews et al., 2017).
Experiment Setup Yes We used 80% of available windows for training and used the remaining 20% for hyperparameter tuning with the MSE prediction on future samples used as model selection criterion. [...] To account for the short length of time series usually encountered in the real world, we randomly split the trajectories in windows of 10 seconds. To simulate sporadicity, we sample observation uniformly at random with an average rate of 4 samples per second. Furthermore, for each of those samples, we apply an observation mask that keeps each individual dimension with probability 0.3.