Deep Neural Network Compression for Aircraft Collision Avoidance Systems

Julian, Kyle D.; Kochenderfer, Mykel J.; Owen, Michael P.

doi:10.2514/1.g003724

Cited by 145 publications

(112 citation statements)

References 21 publications

Supporting

Mentioning

111

Contrasting

Order By: Relevance

“…Extensive progress in machine learning has enabled computers to model expected behavior with minimal human guidance and has led to its integration into many safety-critical systems [5,24]. Since all software is prone to unanticipated and undesirable defects, creating test suites and assessing their quality is an important part of building confidence during the software lifecycle.…”

Section: Introductionmentioning

confidence: 99%

Is neuron coverage a meaningful measure for testing deep neural networks?

Harel-Canada

Wang

Gulzar³

et al. 2020

Proceedings of the 28th ACM Joint Meeting on European Software Engineering Conference and Symposium on the Foundations of Softw

141

View full text Add to dashboard Cite

Recent effort to test deep learning systems has produced an intuitive and compelling test criterion called neuron coverage (NC), which resembles the notion of traditional code coverage. NC measures the proportion of neurons activated in a neural network and it is implicitly assumed that increasing NC improves the quality of a test suite. In an attempt to automatically generate a test suite that increases NC, we design a novel diversity promoting regularizer that can be plugged into existing adversarial attack algorithms. We then assess whether such attempts to increase NC could generate a test suite that (1) detects adversarial attacks successfully, (2) produces natural inputs, and (3) is unbiased to particular class predictions. Contrary to expectation, our extensive evaluation finds that increasing NC actually makes it harder to generate an effective test suite: higher neuron coverage leads to fewer defects detected, less natural inputs, and more biased prediction preferences. Our results invoke skepticism that increasing neuron coverage may not be a meaningful objective for generating tests for deep neural networks and call for a new test generation technique that considers defect detection, naturalness, and output impartiality in tandem. CCS CONCEPTS • Software and its engineering → Software testing and debugging; Software reliability; • Computing methodologies → Neural networks.

show abstract

Section: Introductionmentioning

confidence: 99%

Is neuron coverage a meaningful measure for testing deep neural networks?

Harel-Canada

Wang

Gulzar³

et al. 2020

Proceedings of the 28th ACM Joint Meeting on European Software Engineering Conference and Symposium on the Foundations of Softw

141

View full text Add to dashboard Cite

show abstract

“…Whereas IG helps understand why a DNN made a particular prediction about a single input point, another major task is visualizing the decision boundaries of a DNN on infinitely-many input points. Figure 2 shows a visualization of an ACAS Xu DNN [31] which takes as input the position of an airplane and an approaching attacker, then produces as output one of five advisories instructing the plane, such as "clear of conflict" or to move "weak left." Every point in the diagram represents the relative position of the approaching plane, while the color indicates the advisory.…”

Section: Visualization Of Dnn Decision Boundariesmentioning

confidence: 99%

“…Deep Neural Networks (DNNs) [18] have become the state-of-the-art in a variety of applications including image recognition [53,33] and natural language processing [12]. Moreover, they are increasingly used in safety-and security-critical applications such as autonomous vehicles [31] and medical diagnosis [10,38,28,37]. These advances have been accelerated by improved hardware and algorithms.…”

Section: Introductionmentioning

confidence: 99%

“…Owing to the variety of application domains and deployment constraints, DNNs come in many different sizes. For instance, large image-recognition and natural-language processing models are trained and deployed using cloud resources [33,12], medium-size models could be trained in the cloud but deployed on hardware with limited resources [31], and finally small models could be trained and deployed directly on edge devices [47,9,22,34,35]. There has also been a recent push to compress trained models to reduce their size [24].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

SyReNN: A Tool for Analyzing Deep Neural Networks

Sotoudeh

Thakur

2021

Tools and Algorithms for the Construction and Analysis of Systems

View full text Add to dashboard Cite

Deep Neural Networks (DNNs) are rapidly gaining popularity in a variety of important domains. Formally, DNNs are complicated vector-valued functions which come in a variety of sizes and applications. Unfortunately, modern DNNs have been shown to be vulnerable to a variety of attacks and buggy behavior. This has motivated recent work in formally analyzing the properties of such DNNs. This paper introduces SyReNN, a tool for understanding and analyzing a DNN by computing its symbolic representation. The key insight is to decompose the DNN into linear functions. Our tool is designed for analyses using low-dimensional subsets of the input space, a unique design point in the space of DNN analysis tools. We describe the tool and the underlying theory, then evaluate its use and performance on three case studies: computing Integrated Gradients, visualizing a DNN’s decision boundaries, and patching a DNN.

show abstract

“…When an intruder arrives in the vicinity of the owner's neighborhood, after intruder detection [22], the ACAS resolution advisories commands should be transferred to the fixed-wing aircraft control system, which is supposed to deflect control surfaces with the aim to modify future trajectory. The control system of the owner aircraft (that is flying in a specific route) updates its subsequent trajectory with respect to the built-in regression model, while sensory radar information is being provided simultaneously [23]. Finally, the safety control system takes proper actions in order to avoid a possible collision [24].…”

Section: Related Work On Trajectory-based Operationsmentioning

confidence: 99%

New Reliability Studies of Data-Driven Aircraft Trajectory Prediction

Hashemi¹,

Botez²,

Grigorie

2020

Aerospace

View full text Add to dashboard Cite

Two main factors, including regression accuracy and adversarial attack robustness, of six trajectory prediction models are measured in this paper using the traffic flow management system (TFMS) public dataset of fixed-wing aircraft trajectories in a specific route provided by the Federal Aviation Administration. Six data-driven regressors with their desired architectures, from basic conventional to advanced deep learning, are explored in terms of the accuracy and reliability of their predicted trajectories. The main contribution of the paper is that the existence of adversarial samples was characterized for an aircraft trajectory problem, which is recast as a regression task in this paper. In other words, although data-driven algorithms are currently the best regressors, it is shown that they can be attacked by adversarial samples. Adversarial samples are similar to training samples; however, they can cause finely trained regressors to make incorrect predictions, which poses a security concern for learning-based trajectory prediction algorithms. It is shown that although deep-learning-based algorithms (e.g., long short-term memory (LSTM)) have higher regression accuracy with respect to conventional classifiers (e.g., support vector regression (SVR)), they are more sensitive to crafted states, which can be carefully manipulated even to redirect their predicted states towards incorrect states. This fact poses a real security issue for aircraft as adversarial attacks can result in intentional and purposely designed collisions of built-in systems that can include any type of learning-based trajectory predictor.

show abstract

Deep Neural Network Compression for Aircraft Collision Avoidance Systems

Cited by 145 publications

References 21 publications

Is neuron coverage a meaningful measure for testing deep neural networks?

Is neuron coverage a meaningful measure for testing deep neural networks?

SyReNN: A Tool for Analyzing Deep Neural Networks

New Reliability Studies of Data-Driven Aircraft Trajectory Prediction

Contact Info

Product

Resources

About