Details
| Original language | English |
|---|---|
| Title of host publication | Volunteered Geographic Information |
| Subtitle of host publication | Interpretation, Visualization and Social Context |
| Editors | Dirk Burghardt, Elena Demidova, Daniel A. Keim |
| Place of Publication | Cham |
| Pages | 103-130 |
| Number of pages | 28 |
| ISBN (electronic) | 978-3-031-35374-1 |
| Publication status | Published - 9 Dec 2023 |
Abstract
Neural networks have demonstrated great success; however, large amounts of labeled data are usually required for training the networks. In this work, a framework for analyzing the road and traffic situations for cyclists and pedestrians is presented, which only requires very few labeled examples. We address this problem by combining convolutional neural networks and random forests, transforming the random forest into a neural network, and generating a fully convolutional network for detecting objects. Because existing methods for transforming random forests into neural networks propose a direct mapping and produce inefficient architectures, we present neural random forest imitation-an imitation learning approach by generating training data from a random forest and learning a neural network that imitates its behavior. This implicit transformation creates very efficient neural networks that learn the decision boundaries of a random forest. The generated model is differentiable, can be used as a warm start for fine-tuning, and enables end-to-end optimization. Experiments on several real- world benchmark datasets demonstrate superior performance, especially when training with very few training examples. Compared to state-of-the-art methods, we significantly reduce the number of network parameters while achieving the same or even improved accuracy due to better generalization.
Keywords
- Classification, Imitation learning, Localization, Neural networks, Object detection, Random forests
ASJC Scopus subject areas
- Computer Science(all)
- General Computer Science
- Earth and Planetary Sciences(all)
- General Earth and Planetary Sciences
- Social Sciences(all)
- General Social Sciences
Cite this
- Standard
- Harvard
- Apa
- Vancouver
- BibTeX
- RIS
Volunteered Geographic Information: Interpretation, Visualization and Social Context. ed. / Dirk Burghardt; Elena Demidova; Daniel A. Keim. 1. ed. Cham, 2023. p. 103-130.
Research output: Chapter in book/report/conference proceeding › Contribution to book/anthology › Research › peer review
}
TY - CHAP
T1 - Two Worlds in One Network
T2 - Fusing Deep Learning and Random Forests for Classification and Object Detection
AU - Reinders, Christoph
AU - Yang, Michael Ying
AU - Rosenhahn, Bodo
N1 - Publisher Copyright: © The Author(s) 2024. All rights reserved.
PY - 2023/12/9
Y1 - 2023/12/9
N2 - Neural networks have demonstrated great success; however, large amounts of labeled data are usually required for training the networks. In this work, a framework for analyzing the road and traffic situations for cyclists and pedestrians is presented, which only requires very few labeled examples. We address this problem by combining convolutional neural networks and random forests, transforming the random forest into a neural network, and generating a fully convolutional network for detecting objects. Because existing methods for transforming random forests into neural networks propose a direct mapping and produce inefficient architectures, we present neural random forest imitation-an imitation learning approach by generating training data from a random forest and learning a neural network that imitates its behavior. This implicit transformation creates very efficient neural networks that learn the decision boundaries of a random forest. The generated model is differentiable, can be used as a warm start for fine-tuning, and enables end-to-end optimization. Experiments on several real- world benchmark datasets demonstrate superior performance, especially when training with very few training examples. Compared to state-of-the-art methods, we significantly reduce the number of network parameters while achieving the same or even improved accuracy due to better generalization.
AB - Neural networks have demonstrated great success; however, large amounts of labeled data are usually required for training the networks. In this work, a framework for analyzing the road and traffic situations for cyclists and pedestrians is presented, which only requires very few labeled examples. We address this problem by combining convolutional neural networks and random forests, transforming the random forest into a neural network, and generating a fully convolutional network for detecting objects. Because existing methods for transforming random forests into neural networks propose a direct mapping and produce inefficient architectures, we present neural random forest imitation-an imitation learning approach by generating training data from a random forest and learning a neural network that imitates its behavior. This implicit transformation creates very efficient neural networks that learn the decision boundaries of a random forest. The generated model is differentiable, can be used as a warm start for fine-tuning, and enables end-to-end optimization. Experiments on several real- world benchmark datasets demonstrate superior performance, especially when training with very few training examples. Compared to state-of-the-art methods, we significantly reduce the number of network parameters while achieving the same or even improved accuracy due to better generalization.
KW - Classification
KW - Imitation learning
KW - Localization
KW - Neural networks
KW - Object detection
KW - Random forests
UR - http://www.scopus.com/inward/record.url?scp=85195113085&partnerID=8YFLogxK
U2 - 10.1007/978-3-031-35374-1_5
DO - 10.1007/978-3-031-35374-1_5
M3 - Contribution to book/anthology
AN - SCOPUS:85195113085
SN - 978-3-031-35373-4
SN - 978-3-031-35376-5
SP - 103
EP - 130
BT - Volunteered Geographic Information
A2 - Burghardt, Dirk
A2 - Demidova, Elena
A2 - Keim, Daniel A.
CY - Cham
ER -