Source author record

Sungho Choi

Sungho Choi appears in the imported research catalog. Authorship, coauthor and topic links are available while profile ownership is still unclaimed.

ResearcherUnclaimed source record

Artificial Intelligence Biological Physics Machine Learning Populations and Evolution

Catalog footprint

What is connected

2works

4topics

4close collaborators

Actions

Connect this record

Open graph Browse works

Inspect adjacent papers, topics, institutions and collaborators without losing the researcher page.

Building this map preview

BZPEER is loading the nearby papers, people, topics and institutions for this page.

preprint2022arXiv

Robust Imitation Learning against Variations in Environment Dynamics

In this paper, we propose a robust imitation learning (IL) framework that improves the robustness of IL when environment dynamics are perturbed. The existing IL framework trained in a single environment can catastrophically fail with perturbations in environment dynamics because it does not capture the situation that underlying environment dynamics can be changed. Our framework effectively deals with environments with varying dynamics by imitating multiple experts in sampled environment dynamics to enhance the robustness in general variations in environment dynamics. In order to robustly imitate the multiple sample experts, we minimize the risk with respect to the Jensen-Shannon divergence between the agent's policy and each of the sample experts. Numerical results show that our algorithm significantly improves robustness against dynamics perturbations compared to conventional IL baselines.

preprint2015arXiv

Predicting Whole Forest Structure, Primary Productivity, and Biomass Density From Maximum Tree Size and Resource Limitations

In the face of uncertain biological response to climate change and the many critiques concerning model complexity it is increasingly important to develop predictive mechanistic frameworks that capture the dominant features of ecological communities and their dependencies on environmental factors. This is particularly important for critical global processes such as biomass changes, carbon export, and biogenic climate feedback. Past efforts have successfully understood a broad spectrum of plant and community traits across a range of biological diversity and body size, including tree size distributions and maximum tree height, from mechanical, hydrodynamic, and resource constraints. Recently it was shown that global scaling relationships for net primary productivity are correlated with local meteorology and the overall biomass density within a forest. Along with previous efforts, this highlights the connection between widely observed allometric relationships and predictive ecology. An emerging goal of ecological theory is to gain maximum predictive power with the least number of parameters. Here we show that the explicit dependence of such critical quantities can be systematically predicted knowing just the size of the largest tree. This is supported by data showing that forests converge to our predictions as they mature. Since maximum tree size can be calculated from local meteorology this provides a general framework for predicting the generic structure of forests from local environmental parameters thereby addressing a range of critical Earth-system questions.