Graph neural networks (GNNs) have emerged as the mainstream paradigm for graph representation learning due to their effective message aggregation. However, this advantage also amplifies biases inherent in graph topology, raising fairness concerns. Existing fairness-aware GNNs provide satisfactory performance on fairness metrics such as Statistical Parity and Equal Opportunity while maintaining acceptable accuracy trade-offs. Unfortunately, we observe that this pursuit of fairness metrics neglects the GNN's ability to predict negative labels, which renders their predications with extremely high False Positive Rates (FPRs), resulting in negative effects in high-risk scenarios. To this end, we advocate that classification performance should be carefully calibrated while improving fairness, rather than simply constraining accuracy loss. Furthermore, we propose Fair GNN via Structural Entropy (FairGSE), a novel framework that maximizes two-dimensional structural entropy (2D-SE) to improve fairness without neglecting false positives. Experiments on several real-world datasets show FairGSE reduces FPR by 39% vs. state-of-the-art fairness-aware GNNs, with comparable fairness improvement.

Lightweight protocols that are both bandwidth and power thrifty are desirable for sensor networks. In addition, for many sensor network applications, timeliness of data delivery at a sink that collects and interprets raw sensor data is of great importance. In this work, we propose a lightweight framework for source-to-sink data transfer in a wireless sensor network that is geared towards achieving the above two objectives. Our integrated framework consists of three elements: 1) simple labels that eliminate complex addressing requirements, 2) implicit routing that provides an inherent robustness during sleep/wake schedules, and 3) MAC layer anycast to support routing. Our framework, in addition, facilitates the self-organization of sensor nodes into a network that efficiently relays information from the sources to the sink. The key idea of our framework is to associate each sensor node with a hierarchical level with respect to a sink and using MAC layer anycast to simply further packets to higher levels towards the sink. There are no explicit route tables created or maintained; this eliminates the overhead due to route queries or updates, the need for complex processing and the memory requirements for caching routing information. Furthermore, with our framework, the energy costs of data transmission are evenly distributed across the nodes, thereby improving the longevity of the network. Our MAC layer anycast mechanism not only facilitates routing, but also reduces the number of MAC layer back-offs incurred and, consequently, the waiting times for data transmission. This in turn, improves the timeliness of data delivery at the sink. To summarize, our framework is a) energy efficient, b) inherently robust, and c) conceptually simple. We qualitatively assess our scheme to show its efficiency in terms of power consumption, robustness to failure, ease of setup. The results from our simulations and assessments demonstrate the aforementioned benefits and the viability and potential of using our framework.

In mobile ad hoc networks, one might expect multipath routing to provide some robustness to link failures and facilitate the transmission of packets along paths that avoid regions of congestion. Consequently, one would expect an improvement in network performance in terms of the achieved throughput. We consider TCP goodput as the metric of performance. We find that, contrary to expectations, not all TCP connections enjoy the benefits of multipath routing. Specifically, we find that while long (in terms of hop-count) TCP connections seem to benefit, short connections in fact suffer a slight degradation in goodput as compared to TCP using the single shortest path. Furthermore, we find that alternate path routing, wherein packets are routed on a secondary alternate path only upon the failure of the primary path, helps achieve almost the same goodput as when the multiple paths are used simultaneously. The main benefits of currently proposed multipath routing schemes seem to be limited to improving the efficiency of route discoveries that are initiated either due to real route failures (due to mobility) or due to false failures (due to interference effects) for long TCP connections.

We focus on self-contention: contention between packets of the same transport layer connection along the path from source to destination. We observe that self-contention plays an important role in degrading TCP performance in multi-hop wireless networks and that the use of the popular IEEE 802.11 MAC protocol exacerbates self-contention. We propose and study two MAC-layer approaches to alleviate self-contention. The first approach, called quick-exchange (QE), is designed with the intent of reducing the effects of inter-flow self-contention (e.g. between packets of the same connection traveling in opposite directions). The design of our second mechanism, called fast-forward (FF), is geared towards decreasing intra-flow self-contention (e.g. between packets of the same connection traveling in the same direction). We simulate and study our proposed schemes and observe that quick-exchange consistently improves net-work aggregate goodput (by as much as 20% in string topologies, 15% in random static scenarios, and 10% in random mobile scenarios). In contrast to our expectations, fast-forward causes sporadic and often negative effects on goodput for TCP connections. Upon investigation we find that while the MAC is, in some respect, operating more efficiently, as demonstrated by improved UDP throughput; interactions with TCPs congestion control mechanism cause the goodput to degrade. We analyze various effects that cause the respective behaviors with QE and FF in detail.

Spatially separating TCP sessions such that they inflict much lower interference effects on each other may provide gains in performance. We first investigate the possibilities of achieving such gains by considering a centralized, ideal, and unrealistic congestion aware routing approach. We then consider the implementation of a distributed routing protocol to achieve the aforementioned spatial separation benefits. We find that due to practicalities such as the need for the exchange of congestion state, the existence of stale congestion information and the creation of sub-optimal paths, the benefits due to spatial separation are considerably undermined. We perform both macroscopic simulations and microscopic studies of specific constructed examples to understand the reasons and quantify the various effects with both the centralized and the distributed approaches. Our studies suggest that achieving noteworthy performance gains by spatially separating TCP sessions may be extremely difficult if not impossible in ad hoc networks.

Mobile ad hoc networks consist of nodes that are often vulnerable to failure. As such, it is important to provide redundancy in terms of providing multiple node-disjoint paths from a source to a destination. We first propose a modified version of the popular AODV protocol that allows us to discover multiple node-disjoint paths from a source to a destination. We find that very few of such paths can be found. Furthermore, as distances between sources and destinations increase, bottlenecks inevitably occur and thus, the possibility of finding multiple paths is considerably reduced. We conclude that it is necessary to place what we call reliable nodes (in terms of both being robust to failure and being secure) in the network for efficient operations. We propose a deployment strategy that determines the positions and the trajectories of these reliable nodes such that we can achieve a framework for reliably routing information. We define a notion of a reliable path which is made up of multiple segments, each of which either entirely consists of reliable nodes, or contains a preset number of multiple paths between the end points of the segment. We show that the probability of establishing a reliable path between a random source and destination pair increases considerably even with a low percentage of reliable nodes when we control their positions and trajectories in accordance with our algorithm.