Many architects believe that major improvements in cost-energy-performance must now come from domain-specific hardware. This paper evaluates a custom ASIC---called a Tensor Processing Unit (TPU) --- deployed in datacenters since 2015 that accelerates the inference phase of neural networks (NN). The heart of the TPU is a 65,536 8-bit MAC matrix multiply unit that offers a peak throughput of 92 TeraOps/second (TOPS) and a large (28 MiB) software-managed on-chip memory. The TPU's deterministic execution model is a better match to the 99th-percentile response-time requirement of our NN applications than are the time-varying optimizations of CPUs and GPUs that help average throughput more than guaranteed latency. The lack of such features helps explain why, despite having myriad MACs and a big memory, the TPU is relatively small and low power. We compare the TPU to a server-class Intel Haswell CPU and an Nvidia K80 GPU, which are contemporaries deployed in the same datacenters. Our workload, written in the high-level TensorFlow framework, uses production NN applications (MLPs, CNNs, and LSTMs) that represent 95% of our datacenters' NN inference demand. Despite low utilization for some applications, the TPU is on average about 15X -- 30X faster than its contemporary GPU or CPU, with TOPS/Watt about 30X -- 80X higher. Moreover, using the CPU's GDDR5 memory in the TPU would triple achieved TOPS and raise TOPS/Watt to nearly 70X the GPU and 200X the CPU.

% & ' ( ) * !+ , -) ./ 0 1 32 -4 5 6 1 & ) * 7 98 ": <; '= ?> <; ": A@ B DC &E )F HG I8 KJ L 9E )F NM O7 "8 QP RE 6S T> AU B <V W8 ":

Collaborative filtering systems help address information overload by usingthe opinions of users in a community to make personal recommendations fordocuments to each user. Many collaborative filtering systems have few useropinions relative to the large number of documents available. This sparsityproblem can reduce the utility of the filtering system by reducing thenumber of documents for which the system can make recommendations andadversely affecting the quality of recommendations. This paper defines andimplements a model for integrating content-based ratings into acollaborative filtering system. The filterbot model allows collaborativefiltering systems to address sparsity by tapping the strength of contentfiltering techniques. We identify and evaluate metrics for assessing theeffectiveness of filterbots specifically, and filtering system enhancementsin general. Finally, we experimentally validate the filterbot approach byshowing that even simple filterbots such as spell checking can increase theutility for users of sparsely populated collaborative filtering systems.Keywords Collaborative filtering, information filtering, content analysis,recommendation systems, social filtering, GroupLens Research, informationfiltering agents.

A Steiner minimum tree (SMT) is the shortest-length tree in a metric space interconnecting a set of points, called the regular points, possibly using additional vertices. A k-size Steiner minimum tree (kSMT) is one that can be split into components where all regular points are leaves and all components have at most k leaves. The k-Steiner ratio, $\rho_{k}$, is the infimum of the ratios SMT/kSMT over all finite sets of regular points in all possible metric spaces, where the distances are given by a complete graph. Previously, only $\rho_{2}$ and $\rho_{3}$ were known exactly in graphs, and some bounds were known for other values of k. In this paper, we determine $\rho_{k}$ exactly for all k. From this we prove a better approximation ratio for the Steiner tree problem in graphs.

Article The k-Steiner ratio in graphs Share on Authors: Al Borchers Department of Computer Science, University of Minnesota, Minneapolis, MN Department of Computer Science, University of Minnesota, Minneapolis, MNView Profile , Ding-Zhu Du Department of Computer Science, University of Minnesota, Minneapolis, MN Department of Computer Science, University of Minnesota, Minneapolis, MNView Profile Authors Info & Claims STOC '95: Proceedings of the twenty-seventh annual ACM symposium on Theory of computingMay 1995 Pages 641–649https://doi.org/10.1145/225058.225282Online:29 May 1995Publication History 8citation586DownloadsMetricsTotal Citations8Total Downloads586Last 12 Months13Last 6 weeks0 Get Citation AlertsNew Citation Alert added!This alert has been successfully added and will be sent to:You will be notified whenever a record that you have chosen has been cited.To manage your alert preferences, click on the button below.Manage my AlertsNew Citation Alert!Please log in to your account Save to BinderSave to BinderCreate a New BinderNameCancelCreateExport CitationPublisher SiteGet Access