A significant amount of society's infrastructure can be modeled using graph structures, from electric and communication grids, to traffic networks, to social networks. Each of these domains are also susceptible to the cascading spread of negative impacts, whether this be overloaded devices in the power grid or the reach of a social media post containing misinformation. The potential harm of a cascade is compounded when considering a malicious attack by an adversary that is intended to maximize the cascading impact. However, by exploiting knowledge of the cascading dynamics, targets with the largest cascading impact can be preemptively prioritized for defense, and the damage an adversary can inflict can be mitigated. While game theory provides tools for finding an optimal preemptive defense strategy, existing methods struggle to scale to the context of large graph environments because of the combinatorial explosion of possible actions that occurs when the attacker and defender can each choose multiple targets in the graph simultaneously. The proposed method enables a data-driven deep learning approach that uses multi-node representation learning and counterfactual data augmentation to generalize to the full combinatorial action space by training on a variety of small restricted subsets of the action space. We demonstrate through experiments that the proposed method is capable of identifying defense strategies that are less exploitable than SOTA methods for large graphs, while still being able to produce strategies near the Nash equilibrium for small-scale scenarios for which it can be computed. Moreover, the proposed method demonstrates superior prediction accuracy on a validation set of unseen cascades compared to other deep learning approaches.

In this work, we propose an approach to predict multiple design parameters of products using 2D images and supervised learning techniques. Fully parametric 2D or 3D vector representations have a high degree of flexibility since they are generally capable of morphing between all designs in the design space. A reverse design method is applied to parametrically designed products to extract design knowledge from existing designs and apply it to future designs. The goal of this research is to develop an accurate and efficient predictive model for multiple design parameters. The paper introduces several novel aspects, including learning parametric patterns from a 2D model dataset and using them to predict design parameters for new engineering systems. Additionally, the paper proposes a finetuning multilabel approach for predicting multiple design parameters of different types simultaneously, including binary and continuous parameters. The results demonstrate that the proposed approach achieves high prediction accuracy and suggest that the proposed approach can be used as a reliable tool for predicting multiple design parameters of products in practical engineering design scenarios from only 2D images, potentially allowing designer to iterate through more existing engineering designs.

Power grids that are interdependent with communication networks create more possible modes of failure (e.g., cyberattacks) as well as more complex propagation of failure through the coupled networks. To ensure robust defense of smart grids, it is important to model both attacker and defender as intelligent, a scenario that the framework of game theory provides methods to analyze. However, prior works in applying game theoretic models to smart grid security limit the problem space to a small number targets under threat due to the inability of state-of-the-art methods to scale to large networks. Our method scales to large networks by combining neural networks that use featurized action representations with an approximation of large combinatorial actions to generalize knowledge about the best targets to attack/defend across graphs of various topologies and sizes. Our model’s invariance to the size of the input graph allows us to transfer knowledge from games played on small graphs during training to large graphs during evaluation. Our experiments show that our method can learn Nash equilibrium strategies on small networks, and demonstrate low exploitability when generalized to large networks, especially compared to the common heuristics currently used to simulate attacks on large graphs.

We present a form-aware reinforcement learning (RL) method to extend control knowledge from one design form to another without losing the ability to control the original design. A major challenge in developing control knowledge is the creation of generalized control policies across designs of varying form. Our presented RL policy is form-aware because in addition to receiving dynamic state information about the environment, it also receives states that encode information about the form of the design that is being controlled. In this paper, we investigate the impact of this mixed state space on transfer learning. We present a transfer learning method for extending a control policy to a different design form, while continuing to expose the agent to the original design during the training of the new design. To demonstrate this concept, we present a case study of a multi-rotor aircraft simulation, wherein the designated task is to achieve a stable hover. We show that by introducing form states, an RL agent is able to learn a control policy to achieve the hovering task with both a four rotor and three rotor design at once, whereas without the form states it can only hover with the four rotor design. We also benchmark our method against a test case that removes the transfer learning component, as well as a test case that removes the continued exposure to the original design to show the value of each of these components. We find that form states, transfer learning, and parallel learning all contribute to a more robust control policy for the new design, and that parallel learning is especially important for maintaining control knowledge of the original design.

Generative neural networks (GNNs) have successfully used human-created designs to generate novel 3D models that combine concepts from disparate known solutions, which is an important aspect of design exploration. GNNs automatically learn a parameterization (orlatent space) of a design space, as opposed to alternative methods that manually define a parameterization. However, GNNs are typically not evaluated using an explicit notion of physical performance, which is a critical capability needed for design. This work bridges this gap by proposing a method to extract a set of functional designs from the latent space of a point cloud generating GNN, without sacrificing the aforementioned aspects of a GNN that are appealing for design exploration. We introduce a sparsity preserving cost function and initialization strategy for a genetic algorithm (GA) to optimize over the latent space of a point cloud generating autoencoder GNN. We examine two test cases, an example of generating ellipsoid point clouds subject to a simple performance criterion and a more complex example of extracting 3D designs with a low coefficient of drag. Our experiments show that the modified GA results in a diverse set of functionally superior designs while maintaining similarity to human-generated designs in the training data set.

This work investigates surrogate modeling techniques for learning to approximate a computationally expensive function evaluation of 3D models. While in the past, 3D point clouds have been a data format that is too high dimensional for surrogate modeling, by leveraging advances in 3D object autoencoding neural networks, these point clouds can be mapped to a one-dimensional latent space. This leads to the fundamental research question: what surrogate modeling technique is most suitable for learning relationships between the 3D geometric features of the objects captured in the encoded latent vector and the physical phenomena captured in the evaluation software? Radial basis functions (RBFs), Kriging, and shallow 1D analogs of popular deep 2D image classification neural networks are investigated in this work. We find the nonintuitive result that departing from neural networks to decode latent representations of 3D objects into performance predictions is far more efficient than using a neural network decoder. In test cases using datasets of aircraft and watercraft 3D models, the non-neural network surrogate models achieve comparable accuracy to the neural network models. We find that an RBF surrogate model is able to approximate the lift and drag coefficients of 234 aircraft models with a mean absolute error of 1.97 × 10−3 and trains in only 3 seconds. Furthermore, the RBF surrogate model is able to rank a set of designs with an average percentile error of less than 8%. In comparison, a 1D ResNet achieves an average absolute error of 1.35 × 103 in 38 min for the same test case. We validate the comparable accuracy of the four techniques through a test case involving 214 3D watercraft models, but we also find that the distribution of the performance values of the data, in particular the presence of many outliers, has a significant negative impact on accuracy. These results contradict a common perception of neural networks as an efficient “one-size-fits-all” solution for learning black-box functions and suggests that even within systems that utilize multiple neural networks, potentially more efficient alternatives should be considered for each network in the system. Depending on the required accuracy of the application, this surrogate modeling approach could be used to approximate an expensive simulation software, or if the tolerance for error is low, it serves as a first pass which can narrow down the number of candidate designs to be analyzed more thoroughly.