http://hdl.handle.net/1903/7 Fri, 24 May 2013 19:44:21 GMT 2013-05-24T19:44:21Z http://drum.lib.umd.edu:80/retrieve/3205/umiacs.gif http://hdl.handle.net/1903/7 http://hdl.handle.net/1903/13688 Title: The compiler for the XMTC parallel language: Lessons for compiler developers and in-depth description Authors: Tzannes, Alexandros; Caragea, George C.; Vishkin, Uzi; Barua, Rajeev Abstract: In this technical report, we present information on the XMTC compiler and language. We start by presenting the XMTC Memory Model and the issues we encountered when using GCC, the popular GNU compiler for C and other sequential languages, as the basis for a compiler for XMTC, a parallel language. These topics, along with some information on XMT specific optimizations were presented in [10]. Then, we proceed to give some more details on how outer spawn statements (i.e., parallel loops) are compiled to take advantage of XMT’s unique hardware primitives for scheduling flat parallelism and how we incremented this basic compiler to support nested parallelism. Fri, 18 Feb 2011 00:00:00 GMT http://hdl.handle.net/1903/13688 2011-02-18T00:00:00Z http://hdl.handle.net/1903/13339 Title: Learning to Detect Carried Objects with Minimal Supervision Authors: Dondera, Radu; Morariu, Vlad I.; Davis, Larry S. Abstract: We propose a learning-based method for detecting carried objects that generates candidate image regions from protrusion, color contrast and occlusion boundary cues, and uses a classifier to filter out the regions unlikely to be carried objects. The method achieves higher accuracy than state of the art, which can only detect protrusions from the human shape, and the discriminative model it builds for the silhouette context-based region features generalizes well. To reduce annotation effort, we investigate training the model in a Multiple Instance Learning framework where the only available supervision is "walk" and "carry" labels associated with intervals of human tracks, i.e., the spatial extent of carried objects is not annotated. We present an extension to the miSVM algorithm that uses knowledge of the fraction of positive instances in positive bags and that scales to training sets of hundreds of thousands of instances. Fri, 21 Dec 2012 00:00:00 GMT http://hdl.handle.net/1903/13339 2012-12-21T00:00:00Z http://hdl.handle.net/1903/13333 Title: Clashes in the Infosphere, General Intelligence, and Metacognition: Final project report Authors: Perlis, Don; Cox, Michael. T. Abstract: Humans confront the unexpected every day, deal with it, and often learn from it. AI agents, on the other hand, are typically brittle—they tend to break down as soon as something happens for which their creators did not explicitly anticipate. The central focus of our research project is this problem of brittleness which may also be the single most important problem facing AI research. Our approach to brittleness is to model a common method that humans use to deal with the unexpected, namely to note occurrences of the unexpected (i.e., anomalies), to assess any problem signaled by the anomaly, and then to guide a response or solution that resolves it. The result is the Note-Assess-Guide procedure of what we call the Metacognitive Loop or MCL. To do this, we have implemented MCL-based systems that enable agents to help themselves; they must establish expectations and monitor them, note failed expectations, assess their causes, and then choose appropriate responses. Activities for this project have developed and refined a human-dialog agent and a robot navigation system to test the generality of this approach. Wed, 12 Dec 2012 00:00:00 GMT http://hdl.handle.net/1903/13333 2012-12-12T00:00:00Z http://hdl.handle.net/1903/12879 Title: An Effective Approach to Temporally Anchored Information Retrieval Authors: Wei, Zheng; JaJa, Joseph Abstract: We consider in this paper the information retrieval problem over a collection of time-evolving documents such that the search has to be carried out based on a query text and a temporal specification. A solution to this problem is critical for a number of emerging large scale applications involving archived collections of web contents, social network interactions, blog traffic, and information feeds. Given a collection of time-evolving documents, we develop an effective strategy to create inverted files and indexing structures such that a temporally anchored query can be processed fast using similar strategies as in the non-temporal case. The inverted files generated have exactly the same structure as those generated for the classical (non-temporal) case, and the size of the additional indexing structures is shown to be small. Well-known previous algorithms for constructing inverted files or for computing relevance can be extended to handle the temporal case. Moreover, we present high throughput, scalable parallel algorithms to build the inverted files with the additional indexing structures on multicore processors and clusters of multicore processors. We illustrate the effectiveness of our approach through experimental tests on a number of web archives, and include a comparison of space used by the indexing structures and postings lists and search time between our approach and the traditional approach that ignores the temporal information. Fri, 17 Aug 2012 00:00:00 GMT http://hdl.handle.net/1903/12879 2012-08-17T00:00:00Z