Multiple Instance learning (MIL) models have been extensively used in pathology to predict biomarkers and risk-stratify patients from gigapixel-sized images. Machine learning problems in medical imaging often deal with rare diseases, making it important for these models to work in a label-imbalanced setting. In pathology images, there is another level of imbalance, where given a positively labeled Whole Slide Image (WSI), only a fraction of pixels within it contribute to the positive label. This compounds the severity of imbalance and makes imbalanced classification in pathology challenging. Furthermore, these imbalances can occur in out-of-distribution (OOD) datasets when the models are deployed in the real-world. We leverage the idea that decoupling feature and classifier learning can lead to improved decision boundaries for label imbalanced datasets. To this end, we investigate the integration of supervised contrastive learning with multiple instance learning (SC-MIL). Specifically, we propose a joint-training MIL framework in the presence of label imbalance that progressively transitions from learning bag-level representations to optimal classifier learning. We perform experiments with different imbalance settings for two well-studied problems in cancer pathology: subtyping of non-small cell lung cancer and subtyping of renal cell carcinoma. SC-MIL provides large and consistent improvements over other techniques on both in-distribution (ID) and OOD held-out sets across multiple imbalanced settings.

Alignment of many 3D point clouds, possibly captured by multiple devices at different times, is a critical step for increasingly popular applications such as 3D model construction and augmented reality. For very large data sets, traditional methods such as ICP can become computationally intractable, or produce poor results. We present an efficient method for accurately aligning very large numbers of dense 3D point clouds, and apply it to a city-scale data set. The method relies on the novel combination of 1) partitioning the point clouds based on loop structures detected across a combined network of all device capture paths, and 2) making use of the loop closure property to accurately align point clouds within each sub-problem. Final global alignment of the loop-based results is formulated as a least squares optimization with closed form solution. Experimental results are shown for aligning 3D points across the entire city of San Francisco with centimeter-scale accuracy, via an efficient parallelized architecture.

We present a fast construction algorithm for the hierarchical tree decompositions that lie at the heart of oblivious routing strategies and that form the basis for approximation and online algorithms for various cut problems in graphs. Given an undirected graph G = (V, E, c) with edge capacities, we compute a single tree T = (VT,ET,cT), where the leaf nodes of T correspond to nodes in G, such that the tree approximates the cut-structure of G up to a factor of (log4 n). The best existing construction by Harrelson, Hildrum, and Rao [12] just guarantees a polynomial running time but offers a better approximation guarantee of (log2 n log log n). Phrasing our results in terms of vertex sparsifiers, we obtain the following: For a graph G = (V, E) with a subset S of terminals, we compute a tree T with at most 2|S| vertices (and the leafs of T correspond to nodes in S) such that T is a flow-sparsifier for S in G with quality (log2 nlog2 k), where |V| = n and |S| = k. The running time is (polylogn · T(m, 1/log3n)) where T(m, ∊) is the time for computing an approximate maxflow in a graph with m edges. The latter is almost linear due to the recent results of Sherman [23] and Kelner et al. [13].

Recently, the largest concern for corporate owners of Web servers is how to minimize the response time. In order to minimize the response time, a new efficient file transmission technique for World-Wide-Web, called Web file transmission by object packaging, is proposed. The prototype design and preliminary performance evaluation of object packaging have been presented in the 27th conference on local computer networks. In this paper, the performance of object packaging is compared to HTTP 1.0 and HTTP 1.1 persistent connection in terms of the server and client response time and CPU workload by experiment method. It is found that object packaging reduces the response time and CPU workload at the server by up to 92.0 and 91.1% from HTTP 1.0 and 64.5 and 82.4% from HTTP 1.1 persistent connection, respectively. From the results of the experiments, it is found that object packaging will be an efficient technique to minimize the response time in transferring web files. The results of the experiments also suggest the significance of disk I/O overhead for Web servers in high speed networks.