Andrew L. Clark

Andrew Clark, Stuart Phinn, Peter Scarth

In this paper, we trialled different methods of data preparation for Convolutional Neural Network (CNN) training and semantic segmentation of land use land cover (LULC) features within aerial photography over the Wet Tropics and Atherton Tablelands, Queensland, Australia. This was conducted through trialling and ranking various training patch selection sampling strategies, patch and batch sizes and data augmentations and scaling. We also compared model accuracy through producing the LULC classification using a single pass of a grid of patches and averaging multiple grid passes and three rotated version of each patch. Our results showed: a stratified random sampling approach for producing training patches improved the accuracy of classes with a smaller area while having minimal effect on larger classes; a smaller number of larger patches compared to a larger number of smaller patches improves model accuracy; applying data augmentations and scaling are imperative in creating a generalised model able to accurately classify LULC features in imagery from a different date and sensor; and producing the output classification by averaging multiple grids of patches and three rotated versions of each patch produced and more accurate and aesthetic result. Combining the findings from the trials, we fully trained five models on the 2018 training image and applied the model to the 2015 test image with the output LULC classifications achieving an average kappa of 0.84 user accuracy of 0.81 and producer accuracy of 0.87. This study has demonstrated the importance of data pre-processing for developing a generalised deep-learning model for LULC classification which can be applied to a different date and sensor. Future research using CNN and earth observation data should implement the findings of this study to increase LULC model accuracy and transferability.

Abdullah Al Maruf, Luyao Niu, Bhaskar Ramasubramanian et al.

Autonomous cyber and cyber-physical systems need to perform decision-making, learning, and control in unknown environments. Such decision-making can be sensitive to multiple factors, including modeling errors, changes in costs, and impacts of events in the tails of probability distributions. Although multi-agent reinforcement learning (MARL) provides a framework for learning behaviors through repeated interactions with the environment by minimizing an average cost, it will not be adequate to overcome the above challenges. In this paper, we develop a distributed MARL approach to solve decision-making problems in unknown environments by learning risk-aware actions. We use the conditional value-at-risk (CVaR) to characterize the cost function that is being minimized, and define a Bellman operator to characterize the value function associated to a given state-action pair. We prove that this operator satisfies a contraction property, and that it converges to the optimal value function. We then propose a distributed MARL algorithm called the CVaR QD-Learning algorithm, and establish that value functions of individual agents reaches consensus. We identify several challenges that arise in the implementation of the CVaR QD-Learning algorithm, and present solutions to overcome these. We evaluate the CVaR QD-Learning algorithm through simulations, and demonstrate the effect of a risk parameter on value functions at consensus.

Junlin Wu, Andrew Clark, Yiannis Kantaros et al.

While ensuring stability for linear systems is well understood, it remains a major challenge for nonlinear systems. A general approach in such cases is to compute a combination of a Lyapunov function and an associated control policy. However, finding Lyapunov functions for general nonlinear systems is a challenging task. To address this challenge, several methods have been proposed that represent Lyapunov functions using neural networks. However, such approaches either focus on continuous-time systems, or highly restricted classes of nonlinear dynamics. We propose the first approach for learning neural Lyapunov control in a broad class of discrete-time systems. Three key ingredients enable us to effectively learn provably stable control policies. The first is a novel mixed-integer linear programming approach for verifying the discrete-time Lyapunov stability conditions, leveraging the particular structure of these conditions. The second is a novel approach for computing verified sublevel sets. The third is a heuristic gradient-based method for quickly finding counterexamples to significantly speed up Lyapunov function learning. Our experiments on four standard benchmarks demonstrate that our approach significantly outperforms state-of-the-art baselines. For example, on the path tracking benchmark, we outperform recent neural Lyapunov control baselines by an order of magnitude in both running time and the size of the region of attraction, and on two of the four benchmarks (cartpole and PVTOL), ours is the first automated approach to return a provably stable controller. Our code is available at: https://github.com/jlwu002/nlc_discrete.

Bhaskar Ramasubramanian, Luyao Niu, Andrew Clark et al.

This paper studies the synthesis of control policies for an agent that has to satisfy a temporal logic specification in a partially observable environment, in the presence of an adversary. The interaction of the agent (defender) with the adversary is modeled as a partially observable stochastic game. The goal is to generate a defender policy to maximize satisfaction of a given temporal logic specification under any adversary policy. The search for policies is limited to the space of finite state controllers, which leads to a tractable approach to determine policies. We relate the satisfaction of the specification to reaching (a subset of) recurrent states of a Markov chain. We present an algorithm to determine a set of defender and adversary finite state controllers of fixed sizes that will satisfy the temporal logic specification, and prove that it is sound. We then propose a value-iteration algorithm to maximize the probability of satisfying the temporal logic specification under finite state controllers of fixed sizes. Lastly, we extend this setting to the scenario where the size of the finite state controller of the defender can be increased to improve the satisfaction probability. We illustrate our approach with an example.

Bhaskar Ramasubramanian, Andrew Clark, Linda Bushnell et al.

This paper studies the synthesis of control policies for an agent that has to satisfy a temporal logic specification in a partially observable environment, in the presence of an adversary. The interaction of the agent (defender) with the adversary is modeled as a partially observable stochastic game. The search for policies is limited to over the space of finite state controllers, which leads to a tractable approach to determine policies. The goal is to generate a defender policy to maximize satisfaction of a given temporal logic specification under any adversary policy. We relate the satisfaction of the specification in terms of reaching (a subset of) recurrent states of a Markov chain. We then present a procedure to determine a set of defender and adversary finite state controllers of given sizes that will satisfy the temporal logic specification. We illustrate our approach with an example.

Andrew Clark, Qiqiang Hou, Linda Bushnell et al.

This paper studies the problem of selecting a submatrix of a positive definite matrix in order to achieve a desired bound on the smallest eigenvalue of the submatrix. Maximizing this smallest eigenvalue has applications to selecting input nodes in order to guarantee consensus of networks with negative edges as well as maximizing the convergence rate of distributed systems. We develop a submodular optimization approach to maximizing the smallest eigenvalue by first proving that positivity of the eigenvalues of a submatrix can be characterized using the probability distribution of the quadratic form induced by the submatrix. We then exploit that connection to prove that positive-definiteness of a submatrix can be expressed as a constraint on a submodular function. We prove that our approach results in polynomial-time algorithms with provable bounds on the size of the submatrix. We also present generalizations to non-symmetric matrices, alternative sufficient conditions for the smallest eigenvalue to exceed a desired bound that are valid for Laplacian matrices, and a numerical evaluation.

Phillip Lee, Andrew Clark, Linda Bushnell et al.

Networked control systems consist of distributed sensors and actuators that communicate via a wireless network. The use of an open wireless medium and unattended deployment leaves these systems vulnerable to intelligent adversaries whose goal is to disrupt the system performance. In this paper, we study the wormhole attack on a networked control system, in which an adversary establishes a link between two distant regions of the network by using either high-gain antennas, as in the out-of-band wormhole, or colluding network nodes as in the in-band wormhole. Wormholes allow the adversary to violate the timing constraints of real-time control systems by delaying or dropping packets, and cannot be detected using cryptographic mechanisms alone. We study the impact of the wormhole attack on the network flows and delays and introduce a passivity-based control-theoretic framework for modeling the wormhole attack. We develop this framework for both the in-band and out-of-band wormhole attacks as well as complex, hereto-unreported wormhole attacks consisting of arbitrary combinations of in-and out-of band wormholes. We integrate existing mitigation strategies into our framework, and analyze the throughput, delay, and stability properties of the overall system. Through simulation study, we show that, by selectively dropping control packets, the wormhole attack can cause disturbances in the physical plant of a networked control system, and demonstrate that appropriate selection of detection parameters mitigates the disturbances due to the wormhole while satisfying the delay constraints of the physical system.