Policy compression for aircraft collision avoidance systems

By Kyle D. Julian, Jessica Lopez, J. Brush, Michael P. Owen, Mykel J. Kochenderfer. 2016

272 citations recorded by the source corpus.

Subfields: Aerospace Engineering, Artificial Intelligence, Automotive Engineering, Efficiency & Compression, Robotics & Control

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Aerospace Engineering

In Artificial Intelligence

In Automotive Engineering

In Efficiency & Compression

In Robotics & Control

Citations in the corpus

Efficient Formal Safety Analysis of Neural Networks (2018, methodology): “Specifically, we evaluate on fully connected ACAS Xu models [12], three fully connected Drebin models [5], three fully connected MNIST models [17], one convolutional MNIST model [32], and one convolutional self-driving car model [2].”
Formal Security Analysis of Neural Networks using Symbolic Intervals (2018, background): “DNNs are becoming increasingly popular in such systems due to better accuracy and less performance overhead than traditional rule-based systems [19].”
DeepXplore: Automated Whitebox Testing of Deep Learning Systems (2017, background): “These advances have led to widespread adoption and deployment of DL in security- and safety-critical systems such as self-driving cars [10], malware detection [88], and aircraft collision avoidance systems [35].”
Reluplex: An Efficient SMT Solver for Verifying Deep Neural Networks (2017, methodology): “To overcome this challenge, a DNN representation was explored as a potential replacement for the table [13].”
Ground-Truth Adversarial Examples (2017, background): “The intended use of deep neural networks as controllers in safety-critical systems [1, 10] has sparked an interest in developing techniques for verifying that they satisfy various properties [4, 9, 12, 17, 18].”
Towards Proving the Adversarial Robustness of Deep Neural Networks (2017, methodology): “We thank Neal Suchy from the FAA and Lindsey Kuper from Intel for their valuable comments and support.”
DeepSafe: A Data-driven Approach for Checking Adversarial Robustness in Neural Networks (2017, methodology): “The FAA is currently exploring an implementation of ACAS Xu that uses an array of 45 deep neural networks.”
MixTrain: Scalable Training of Formally Robust Neural Networks (2018, background): “They prevent the models being reliably applied in the security-critical systems such as autonomous vehicles [2], [3], [4] and unmanned aircrafts [5], [6], [7], [8].”
Provably Minimally-Distorted Adversarial Examples (2017, methodology): “The intended use of deep neural networks as controllers in safety-critical systems (Julian et al., 2016; Bojarski et al., 2016) has sparked an interest in developing techniques for verifying that they satisfy various properties (Pulina and Tacchella, 2010; 2012; Huang”
Better the Devil you Know: An Analysis of Evasion Attacks using Out-of-Distribution Adversarial Examples (2019, background): “Machine learning (ML), spurred by the advent of deep neural networks, has become ubiquitous due to its impressive performance in domains as varied as image recognition [72, 108], natural language and speech processing [29, 33, 55], game-playing [16, 89, 107] and aircraft collision avoidance [63].”
CRADLE: Cross-Backend Validation to Detect and Localize Bugs in Deep Learning Libraries (2019, methodology): “Keras has been used to implement neural networks in critical domains, including aircraft collision avoidance systems [1], inflammatory bowel disease diagnosis [13], chemical reaction predictions [14], medical imaging [15], [16], air quality control [17] and computer network security [18].”
Toward Scalable Verification for Safety-Critical Deep Networks (2018, methodology): “The ACAS Xu network [5] used this approach due to considerations that did not include verification — rather, due to hardware constraints, the developers found that many smaller networks were preferable to one large one.”