Alper Yilmaz

[introductory/intermediate] Deep Learning and Deep Reinforcement Learning for Geospatial Localization

Summary

Geospatial information has become part of our life with availability of map and navigation services on smart devices. As the demand for autonomy of ground, aerial and maritime platforms increase, the dependability on geospatial information has increased in manifolds requiring solutions that use and update the models utilizing the geospatial information. The recent autonomous solutions in the industry use deep learning for detecting objects of interest and utilize an additional control mechanism for platform motion.

This lecture series will present a unified control mechanism for geolocalization and navigation of platforms by utilizing deep learning and reinforcement learning. We will start with a short intro to deep learning, reinforcement learning and their combination. This introduction will be followed by discussions on their application on the geolocalization and navigation problems. The series is to introduce the newcomers to these concepts and while it will benefit researchers with previous knowledge on deep learning and reinforcement learning by expanding their application to geospatial applications.

Syllabus

• Perceptron and multilayer perceptron
• Basics of deep learning
• Reinforcement learning
• Deep reinforcement learning and its variants
• Geospatial data organization for geolocalization and navigation
• Deep learning for geolocalization
• Deep reinforcement learning for navigation

References

Zhang et al. Dive into Deep Learning. Online book with example codes implemented with NumPy/MXNet, PyTorch, and TensorFlow.

Richard S. Sutton and Andrew G. Barto, Reinforcement Learning: An Introduction. MIT Press.

M. T. Koroglu, M. Korkmaz, A. Yilmaz and A. Durdu. Multiple hypothesis testing approach to pedestrian inertial navigation. IEEE Indoor Positioning Indoor Navigation Conference (IPIN), Pisa, Italy, 2019, pp. 1-8.

Y. Han, A. Yilmaz. August 2022. Learning to Drive Using Sparse Imitation Reinforcement Learning. International Conference on Pattern Recognition. Montreal, Canada. (to appear)

Y. Han, A. Yilmaz. July 2021. Dynamic routing for navigation in changing unknown maps using deep reinforcement learning. ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci., V-1-2021, 145–150, https://doi.org/10.5194/isprs-annals-V-1-2021-145-2021, 2021.

Pre-requisites

Basic knowledge of machine learning and artificial intelligence.

Short bio

Alper Yilmaz is a Professor at The Ohio State University. He was inducted into the U.S. National Academy of Inventors in 2020 and is a Fellow of the American Society for Photogrammetry and Remote Sensing (ASPRS). Dr. Yilmaz’s research focuses on biomimetic navigation systems for unmanned systems and mining anomalies in multi-physics and multi-dimensional data and has published over 200 publications and patents. His research has received national and international recognition that resulted in a number of awards. A recent study from Elsevier lists him among the top 2% most cited researchers in “Artificial Intelligence & Image Processing” and “Geological & Geomatics Engineering.” Dr. Yilmaz has advised 23 Ph.D. and 15 M.Sc. students to completion.