Yuan Xue

Accurate identification and localization of abnormalities from radiology images play an integral part in clinical diagnosis and treatment planning. Building a highly accurate prediction model for these tasks usually requires a large number of images manually annotated with labels and finding sites of abnormalities. In reality, however, such annotated data are expensive to acquire, especially the ones with location annotations. We need methods that can work well with only a small amount of location annotations. To address this challenge, we present a unified approach that simultaneously performs disease identification and localization through the same underlying model for all images. We demonstrate that our approach can effectively leverage both class information as well as limited location annotation, and significantly outperforms the comparative reference baseline in both classification and localization tasks.

The shared-medium multihop nature of wireless ad hoc networks poses fundamental challenges to the design of effective resource allocation algorithms that are optimal with respect to resource utilization and fair across different network flows. None of the existing resource allocation algorithms in wireless ad hoc networks have realistically considered end-to-end flows spanning multiple hops. Moreover, strategies proposed in wireline networks are not applicable in the context of wireless ad hoc networks, due to their unique characteristics of location-dependent contention. In this paper, we propose a new price-based resource allocation framework in wireless ad hoc networks to achieve optimal resource utilization and fairness among competing end-to-end flows. We build our pricing framework on the notion of maximal cliques in wireless ad hoc networks, as compared to individual links in traditional wide-area wireline networks. Based on such a price-based theoretical framework, we present a two-tier iterative algorithm. Distributed across wireless nodes, the algorithm converges to a global network optimum with respect to resource allocations. We further improve the algorithm toward asynchronous network settings and prove its convergence. Extensive simulations under a variety of network environments have been conducted to validate our theoretical claims.

Improving TCP performance has long been the focus of many research efforts in mobile ad hoc networks (MANET). In this paper, we address one aspect of this endeavor: how to properly set TCP's congestion window limit (CWL) to achieve optimal performance. Past research has shown that using a small CWL improves TCP performance in certain scenarios [M. Gerla et al., Feb. 1999], [Z. Fu et al., Apr. 2003], however, no comprehensive study has been given. To this end, we turn the problem of setting TCP's optimal CWL into identifying the bandwidth-delay product (BDP) of a path in MANET. We first show and prove that, independent of the MAC layer protocol being used, the BDP of a path in MANET cannot exceed the round-trip hop-count (RTHC) of the path. We further refine this upper bound based on the IEEE 802.11 MAC layer protocol, and show that in a chain topology, a tighter upper bound exists, which is approximately 1/5 of the RTHC of the path. Based on this tighter bound, we propose an adaptive CWL setting strategy to dynamically adjust TCP's CWL according to the current RTHC of its path. Using ns-2 simulations, we show that our simple strategy improves TCP performance by 8% to 16% in a dynamic MANET environment.