Medical images may contain various types of artifacts with different patterns and mixtures, which depend on many factors such as scan setting, machine condition, patients’ characteristics, surrounding environment, etc. However, existing deep-learning-based artifact reduction methods are restricted by their training set with specific predetermined artifact types and patterns. As such, they have limited clinical adoption. In this paper, we introduce One-Shot medical image Artifact Reduction (OSAR), which exploits the power of deep learning but without using pre-trained general networks. Specifically, we train a light-weight image-specific artifact reduction network using data synthesized from the input image at test-time. Without requiring any prior large training data set, OSAR can work with almost any medical images that contain varying additive artifacts which are not in any existing data sets. In addition, Computed Tomography (CT) and Magnetic Resonance Imaging (MRI) are used as vehicles and show that the proposed method can reduce artifacts better than state-of-the-art both qualitatively and quantitatively using shorter test time.

Anomaly detection has been widely studied and used in diverse applications. Building an effective anomaly detection system requires researchers and developers to learn complex structure from noisy data, identify dynamic anomaly patterns, and detect anomalies with limited labels. Recent advancements in deep learning techniques have greatly improved anomaly detection performance, in comparison with classical approaches, and have extended anomaly detection to a wide variety of applications. This tutorial will help the audience gain a comprehensive understanding of deep learning based anomaly detection techniques in various application domains. First, we give an overview of the anomaly detection problem, introducing the approaches taken before the deep model era and listing out the challenges they faced. Then we survey the state-of-the-art deep learning models that range from building block neural network structures such as MLP, CNN, and LSTM, to more complex structures such as autoencoder, generative models (VAE, GAN, Flow-based models), to deep one-class detection models, etc. In addition, we illustrate how techniques such as transfer learning and reinforcement learning can help amend the label sparsity issue in anomaly detection problems and how to collect and make the best use of user labels in practice. Second to last, we discuss real world use cases coming from and outside LinkedIn. The tutorial concludes with a discussion of future trends.

The happened-before model (or the poset model) has been widely used for modeling the computations (execution traces) of parallel programs and detecting predicates (user-specified conditions). This model captures potential causality as well as locking constraints among the executed events of computations using Lamport's happened-before relation. The detection of a predicate in a computation is performed by checking if the predicate could become true in any reachable global state of the computation. In this paper, we argue that locking constraints are fundamentally different from potential causality. Hence, a poset is not an appropriate model for debugging purposes when the computations contain locking constraints. We present a model called Locking Poset, or a Loset, that generalizes the poset model for locking constraints. Just as a poset captures possibly an exponential number of total orders, a loset captures possibly an exponential number of posets. Therefore, detecting a predicate in a loset is equivalent to detecting the predicate in all corresponding posets. Since determining if a global state is reachable in a computation is a fundamental problem for detecting predicates, this paper first studies the reachability problem in the loset model. We show that the problem is NP-complete. Afterwards, we introduce a subset of reachable global states called lock-free feasible global states such that we can check whether a global state is lock-free feasible in polynomial time. Moreover, we show that lock-free feasible global states can act as "reset" points for reachability and be used to drastically reduce the time for determining the reachability of other global states. We also introduce strongly feasible global states that contain all reachable global states and show that the strong feasibility of a global state can be checked in polynomial time. We show that strong feasibility provides an effective approximation of reachability for many practical applications.

Verifying the correctness of executions of concurrent and distributed programs is difficult because they show nondeterministic behavior due to different process scheduling order. Predicate detection can alleviate this problem by predicting whether the user-specified condition (predicate) could have become true in any global state of the given concurrent or distributed computation. The method is predictive because it generates inferred global states from the observed execution path and then checks if those global states satisfy the predicate. An important part of the predicate detection method is global states enumeration, which generates the consistent global states, including the inferred ones, of the given computation. Cooper and Marzullo gave the first enumeration algorithm based on a breadth first strategy (BFS). Later, many algorithms have been proposed to improve the space and time complexity. Among the existing algorithms, the Tree algorithm due to Jegou et al. has the smallest time complexity and requires O(|P|) space, which is linear to the size of the computation P. In this paper, we present a fast algorithm, QuickLex, to enumerate global states in the lexical order. QuickLex requires much smaller space than O(|P|). From our experiments, the Tree algorithm requires 2-10 times more memory space than QuickLex. Moreover, QuickLex is 4 times faster than Tree even though the asymptotic time complexity of QuickLex is higher than that of Tree. The reason is that the worst case time complexity of QuickLex happens only in computations that are not common in practice. Moreover, Tree is built on linked-lists and QuickLex can be implemented using integer arrays. In comparison with the existing lexical algorithm (Lex), QuickLex is 7 times faster and uses almost the same amount of memory as Lex. Finally, we implement a parallel-and-online predicate detector for concurrent programs using QuickLex, which can detect data races and violation of invariants in the programs.

Verifying the correctness of the executions of a concurrent program is difficult because of its nondeterministic behavior. One of the verification methods is predicate detection, which predicts whether the user specified condition (predicate) could become true in any global states of the program. The method is predictive because it generates inferred execution paths from the observed execution path and then checks the predicate on the global states of inferred paths. One important part of predicate detection is global states enumeration, which generates the global states on inferred paths. Cooper and Marzullo gave the first enumeration algorithm based on a breadth first strategy (BFS). Later, many algorithms have been proposed to improve space and time complexity. None of them, however, takes parallelism into consideration. In this paper, we present the first parallel and online algorithm, named ParaMount, for global state enumeration. Our experimental results show that ParaMount speeds up the existing sequential algorithms by a factor of 6 with 8 threads. We have implemented an online predicate detector using ParaMount. For predicate detection, our detector based on ParaMount is 10 to 50 times faster than RV runtime (a verification tool that uses Cooper and Marzullo’s BFS enumeration algorithm).

In this paper we study a global routing problem that considers not only overflow and wirelength but also layer directives. A layer directive is often given to a timing-critical net for meeting target performance, and it specifies a range of consecutive layers on which the net should be routed. Unlike a previous work that focuses only on a restricted set of layer ranges, our problem allows arbitrary layer ranges to be specified. We present a global router which enhances an academic router by employing techniques to take general layer directives into account during two-dimensional (2D) routing and layer assignment. The experiment results show that our global router can produce encouraging solutions for all test cases.

Global routing is a very crucial stage in a design cycle, because it physically plans the routes of nets on a chip. In order to boost the research and development of global routing techniques, ISPD held contests and released benchmarks in 2007 and 2008, respectively. However, the contests may lead researchers away from facing other real problems in practice. In this paper we study a new global routing problem that not only considers traditional routing objectives such as overflow and wirelength but also focuses on honoring layer directives that are usually specified for timing-critical nets to alleviate performance degrading. Based on novel extensions of an academic router, we present a new global router called GLADE for the addressed problem. The experimental results show that GLADE can effectively generate a high-quality solution, which balances the metrics under consideration, for each test case from the set of recently released ICCAD 2009 benchmarks.

This paper presents a robust global router called NTHU-Route 2.0 that improves the solution quality and runtime of NTHU-Route by the following enhancements: 1) a new history based cost function; 2) new ordering methods for congested region identification and rip-up and reroute; and 3) two implementation techniques. We report convincing experimental results to show the effectiveness of each individual enhancement. With all these enhancements together, NTHU-Route 2.0 solves all ISPD98 benchmarks with very good quality. Moreover, NTHU-Route 2.0 routes 7 of 8 ISPD07 benchmarks and 12 of 16 ISPD08 benchmarks without any overflow. Compared with other state-of-the-art global routers, NTHU-Route 2.0 is able to produce better solution quality and/or run more efficiently.

We present in this paper a fast and stable global router called NTHU-Route 2.0 that improves the solution quality and runtime of a state-of-the-art router, NTHU-Route, by the following enhancements: (1) a new history based cost function, (2) new ordering methods for congested region identification and rip-up and reroute, and (3) two implementation techniques. The experimental results show that NTHU-Router 2.0 solves all ISPD98 benchmarks with very good quality. Moreover, it routes 7 of 8 ISPD07 benchmarks without any overflow. In particular, for one of the ISPD07 benchmarks which are thought to be difficult cases previously, NTHU-Route 2.0 can completely eliminate its total overflow. NTHU-Route 2.0 also successfully solves 12 of 16 ISPD08 benchmarks without causing any overflow.

Abstract The wireless sensor network (WSN) has attracted lots of attention recently. Broadcast is a fundamental operation for all kinds of networks and it has not been addressed seriously in the WSN. Therefore, we propose two types of power and time efficient broadcasting protocols, namely one‐to‐all and all‐to‐all broadcasting protocols, for five different WSN topologies. Our one‐to‐all broadcasting protocols conserve power and time by choosing as few relay nodes as possible to scatter packets to the whole network. Besides, collisions are carefully handled such that our one‐to‐all broadcasting protocols can achieve 100% reachability. By assigning each node a proper channel, our all‐to‐all broadcasting protocols are collision free and efficient. Numerical evaluation results compare the performance of the five topologies and show that our broadcasting protocols are power and time efficient. Copyright © 2005 John Wiley & Sons, Ltd.

The wireless sensor network (WSN) has attracted lots of attention recently. Since the sensor nodes usually have no plug-in power, we have to conserve power so that each sensor node can operate for a longer period of time. Here, we propose power and time efficient broadcasting protocols for four different WSN topologies. Our broadcasting protocols conserve power and time by choosing as few relay nodes as possible to scatter messages to the whole network. Besides, collisions are carefully handled such that our one-to-all broadcast protocols can achieve 100% reachability. Numerical evaluation results compare the performances of the four topologies and show that our broadcasting protocols are power and time efficient.