In causal inference, interference refers to the phenomenon in which the actions of peers in a network can influence an individual's outcome. Peer effect refers to the difference in counterfactual outcomes of an individual for different levels of peer exposure, the extent to which an individual is exposed to the treatments, actions, or behaviors of peers. Estimating peer effects requires deciding how to represent peer exposure. Typically, researchers define an exposure mapping function that aggregates peer treatments and outputs peer exposure. Most existing approaches for defining exposure mapping functions assume peer exposure based on the number or fraction of treated peers. Recent studies have investigated more complex functions of peer exposure which capture that different peers can exert different degrees of influence. However, none of these works have explicitly considered the problem of automatically learning the exposure mapping function. In this work, we focus on learning this function for the purpose of estimating heterogeneous peer effects, where heterogeneity refers to the variation in counterfactual outcomes for the same peer exposure but different individual's contexts. We develop EgoNetGNN, a graph neural network (GNN)-based method, to automatically learn the appropriate exposure mapping function allowing for complex peer influence mechanisms that, in addition to peer treatments, can involve the local neighborhood structure and edge attributes. We show that GNN models that use peer exposure based on the number or fraction of treated peers or learn peer exposure naively face difficulty accounting for such influence mechanisms. Our comprehensive evaluation on synthetic and semi-synthetic network data shows that our method is more robust to different unknown underlying influence mechanisms when estimating heterogeneous peer effects when compared to state-of-the-art baselines.

Traffic signal control is important in intelligent transportation system, of which cooperative control is difficult to realize but yet vital. Many methods model multi-intersection traffic networks as grids and address the problem using multi-agent reinforcement learning (RL). Despite these existing studies, there is an opportunity to further enhance our understanding of the connectivity and globality of the traffic networks by capturing the spatiotemporal traffic information with efficient neural networks in deep RL. In this paper, we propose a novel multi-agent actor-critic framework based on an interpretable influence mechanism with a centralized learning and decentralized execution method. Specifically, we first construct an actor-critic framework, for which the piecewise linear neural network (PWLNN), named biased ReLU (BReLU), is used as the function approximator to obtain a more accurate and theoretically grounded approximation. Finally, our proposed framework is validated on two synthetic traffic networks to coordinate signal control between intersections, achieving lower traffic delays across the entire traffic network compared to state-of-the-art (SOTA) performance.

As the potential for societal integration of multi-agent robotic systems increases [1], the need to manage the collective behaviors of such systems also increases [2, 3, 4]. There has been significant research effort directed towards the examination of how humans can assist in controlling such collective behaviors, such as in human-swarm interactions [5, 6, 7]. Agent-agent interactions in a swarm of small unmanned aerial systems (sUAS) lead to the emergence of collective behaviors that enable effective coverage and exploration across large spatial extents. However, the same inherent collective behaviors can occasionally limit the ability of the sUAS swarm to focus on specific objects of interest during coverage or exploration missions [8]. In these scenarios, the human operator or supervisor should have the opportunity to fractionally revoke or limit emergent swarm behaviors, and guide the swarm to achieve mission objectives. For most applications, including in industry-and defense-related contexts, such human-swarm interaction (HSI) will likely require intuitive and predictable mechanisms of control to quickly translate the input of the human (such as a gesture) to an influence or effect on the sUAS swarm. The goal of our work is to create an intuitive interface for a human supervisor to influence or guide an sUAS swarm without excessive incursions on decentralized control afforded by these systems, while attempting to create more predictable behaviors. This is a potentially valuable approach that can enable the fully utilization of swarm capabilities, while also retaining an ongoing macroscopic-level of swarm control in scenarios where focus on specific regions of interest is required (e.g., search and rescue, surveillance operations) [9]. The influence mechanism has been implemented and tested using 16 drones in a photo-realistic virtual reality (VR) environment (as shown in Figure 1).

Deep learning models for graphs, especially Graph Convolutional Networks (GCNs), have achieved remarkable performance in the task of semi-supervised node classification. However, recent studies show that GCNs suffer from adversarial perturbations. Such vulnerability to adversarial attacks significantly decreases the stability of GCNs when being applied to security-critical applications. Defense methods such as preprocessing, attention mechanism and adversarial training have been discussed by various studies. While being able to achieve desirable performance when the perturbation rates are low, such methods are still vulnerable to high perturbation rates. Meanwhile, some defending algorithms perform poorly when the node features are not visible. Therefore, in this paper, we propose a novel mechanism called influence mechanism, which is able to enhance the robustness of the GCNs significantly. The influence mechanism divides the effect of each node into two parts: introverted influence which tries to maintain its own features and extroverted influence which exerts influences on other nodes. Utilizing the influence mechanism, we propose the Influence GCN (I-GCN) model. Extensive experiments show that our proposed model is able to achieve higher accuracy rates than state-of-the-art methods when defending against non-targeted attacks.