Learning Interaction Kernels for Agent Systems on Riemannian Manifolds

Authors: Mauro Maggioni, Jason J Miller, Hongda Qiu, Ming Zhong

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

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate the performance of our estimator on two classical first order interacting systems: Opinion Dynamics and a Predator-Swarm system, with each system constrained on two prototypical manifolds, the 2-dimensional sphere and the Poincar e disk model of hyperbolic space. (...) We conduct extensive experiments on these four scenarios to demonstrate the performance of the estimators both in terms of the estimation errors (approximating φ s) and trajectory estimator errors (estimating the observed dynamics) over [0, T].
Researcher Affiliation Academia 1Department of Applied Mathematics & Statistics, Johns Hopkins University 2Department of Mathematics, Department of Applied Mathematics & Statistics, Mathematical Institute for Data Science, Johns Hopkins University.
Pseudocode Yes Algorithm1 1 shows the detailed steps on how to construct the estimator to φ given the observation data.
Open Source Code Yes Implementation of the algorithm can be found on https:// github.com/Ming Zhong Codes/Learning Dynamics, which also includes code to reproduce the results presented here.
Open Datasets No The paper states: "For each system of N = 20 agents, we take M = 500 and L = 500 to generate the training data." However, it does not provide concrete access information (e.g., specific link, DOI, repository name, or formal citation) for this generated training data to be considered publicly available or open.
Dataset Splits No The paper mentions generating "training data" and evaluating on "new random initial conditions," but it does not explicitly specify training/validation/test dataset splits with percentages, absolute sample counts, or references to predefined splits.
Hardware Specification No The main paper does not explicitly describe the hardware used for experiments. It mentions in the "Addressing Reviewers Comments" section that "We have added a section, namely Sec. D.1., in the Supplementary Information (SI) to discuss the computing platform used to run the simulations," implying it's not in the main text.
Software Dependencies No The paper mentions using "a geometric numerical integrator (Hairer, 2001) (4th order Backward Differentiation Formula with a projection scheme)" but does not provide specific version numbers for this or any other software dependencies (e.g., Python, PyTorch, specific libraries).
Experiment Setup Yes For each system of N = 20 agents, we take M = 500 and L = 500 to generate the training data. For each HM, we use first-degree clamped B-splines as the basis functions with dim(HM) = O(n ) = O(( ML log(ML)) 1 3 ). We use a geometric numerical integrator (Hairer, 2001) (4th order Backward Differentiation Formula with a projection scheme) for the evolution of the dynamics.