In this demonstration, we present a system for mining causal knowledge from large corpuses of text documents, such as millions of news articles. Our system provides a collection of APIs for causal analysis and retrieval. These APIs enable searching for the effects of a given cause and the causes of a given effect, as well as the analysis of existence of causal relation given a pair of phrases. The analysis includes a score that indicates the likelihood of the existence of a causal relation. It also provides evidence from an input corpus supporting the existence of a causal relation between input phrases. Our system uses generic unsupervised and weakly supervised methods of causal relation extraction that do not impose semantic constraints on causes and effects. We show example use cases developed for a commercial application in enterprise risk management.

In this paper, we study the problem of answering questions of type "Could X cause Y?" where X and Y are general phrases without any constraints. Answering such questions will assist with various decision analysis tasks such as verifying and extending presumed causal associations used for decision making. Our goal is to analyze the ability of an AI agent built using state-of-the-art unsupervised methods in answering causal questions derived from collections of cause-effect pairs from human experts. We focus only on unsupervised and weakly supervised methods due to the difficulty of creating a large enough training set with a reasonable quality and coverage. The methods we examine rely on a large corpus of text derived from news articles, and include methods ranging from large-scale application of classic NLP techniques and statistical analysis to the use of neural network based phrase embeddings and state-of-the-art neural language models.

A technique to speed up stencil computation is introduced. Computation and data reuse schemes are developed for its application to 1- and 3-dimensional stencils. The approach traverses the data domain fewer times than a state-of-the-art, straightforward iterative stencil implementation would. Performance results are shown for a variety of platforms, exemplifying how it can be straightforwardly applied with existing techniques and frameworks. The technique, named Aggregate Stencil-Loop Iteration (ASLI), works by applying a stencil obtained by the original stencil operator convolved with itself one or more times. This more complex operator creates new opportunities for in-register data reuse and increases the FLOPs-to-load ratio. The total number of FLOPs decreases for 1D but increases for 2D and 3D star-shaped stencils. In both scenarios, speed-up relative to the state-of-the-art is achieved. ASLI is relatively easy to implement and works synergistically with existing methods to optimize stencil computations.

We present an optimized Blue Gene/P implementation of Reverse Time Migration, a seismic imaging algorithm widely used in the petroleum industry today. Our implementation is novel in that it uses large communication bandwidth and low latency to convert an embarrassingly parallel problem into one that can be efficiently solved using massive domain partitioning. The success of this seemingly counterintuitive approach is the result of several key aspects of the imaging problem, including very regular and local communication patterns, balanced compute and communication requirements, scratch data handling, multiple-pass approaches, and most importantly, the fact that partitioning the problem allows each sub-problem to fit in cache, dramatically increasing locality and bandwidth and reducing latency. This approach can be easily extended to next-generation imaging algorithms currently being developed. In this paper we present details of our implementation, including application-scaling results on Blue Gene/P.

In this paper we present our recent research and development work for multicore computing of Line of Sight (LoS) on the Cell Broadband Engine (CBE) processors. LoS can be found in many applications where real-time high performance computing is required. We will describe an efficient LoS multi-core parallel computing algorithm, including the data partition and computation load allocation strategies to fully utilize the CBE's computational resources for efficient LoS viewshed parallel computing. In addition, we will also illustrate a successive fast transpose algorithm to prepare the input data for efficient Single-Instruction-Multiple-Data (SIMD) operations. Furthermore, we describe the data input and output (I/O) management scheme to reduce the (I/O) latency in Direct-Memory-Access (DMA) data fetching and storing operations. The performance evaluation of our LoS viewshed computing scheme over an area of interest (AOI) with more than 4.19 million points has shown that our parallel computing algorithm on CBE takes less than 25.5 ms, which is several orders of magnitude faster than the available commercial systems.

Video surveillance has become one of the most important tools for public safety and security. In this paper, we present our new work on developing efficient parallel computing algorithms and schemes for implementing H.264 video encoder engine on Cell Broadband Engine (CBE) blade for high performance large scale video surveillance applications. Extending our previous work on H.264 video encoding on CBE, we have developed new parallel computing schemes in computational load partition, dynamic task scheduling, motion estimation, and mode selection. We partition the intensive H.264 video encoding computation load into four major functional modules, namely, the pre-processing module, the motion estimation module, the mode selection and transform/quantization module, and the Context Adaptive Binary Arithmetic Coding (CABAC) module. The task scheduler dynamically assigns a waiting computing task to a SPE as soon as it becomes available. Our new implementation has achieved more than 5X performance improvement to encode 32 standard-definition (SD 720x480 pixel resolution) H.264 video streams simultaneously at 30 frames per second with one Cell Blade that consists of 16 Synergistic Processor Elements (SPEs) and two control Power Processor Elements (PPEs) or 448 SD channels of H.264 video streams on a single chassis Cell Blade Center with 14 Cell Blades.

This paper presents HMMer-Cell, an implementation of the computationally intensive bioinformatics application HMMER on the Cell/B.E. multiprocessor architecture. The core of the HMMER workload is the Plan7 Viterbi algorithm, which has a memory requirement of O(N*M) where N is the length of the HMM and M is the length of the input sequence length. The main challenge in implementing the Plan7 Viterbi algorithm on the novel Cell/B.E. multiprocessor is the limited local storage space of the Cell/B.E. SPEs (synergistic processing element). We describe our approach to modifying the Viterbi algorithm to reduce the space complexity from 0(M*N) to O(N). We then proceed to discuss design considerations such as task parallelization and code partitioning, in addition to other optimizations we used to implement HMMer-Cell. We demonstrate near-linear speedup when processing relatively larger HMM profiles; and compare our results to those obtained on commodity x86 processors.

The Cell Broadband Engine (BE) processor provides the potential to achieve an impressive level of performance for scientific applications. This level of performance can be reached by exploiting several dimensions of parallelism, such as thread-level parallelism using several synergistic processing elements, data streaming parallelism, vector parallelism in the form of 128-bit SIMD operations, and pipeline parallelism by issuing multiple instructions in the same clock cycle. In our exploration to achieve the optimum level of performance for Sweep3D, we have enjoyed many pleasant surprises, such as a very high floating point performance, reaching 64% of the theoretical peak in double precision, and an over all performance speedup ranging from 4.5 times when compared with "heavy iron" processors, up to over 20 times with conventional processors.

Does the Cell BE Processor have a place in HPC? If so, what is that place? What are the strengths and weaknesses of the Cell relative to HPC? What changes would make Cell better suited to HPC? What kinds of performance can we expect from Cell in HPC? Where will the bottlenecks be and how can we avoid them? What kind of performance might we expect from a Cell-based supercomputer? These and related questions will be discussed in a moderated forum.Additional topics for open discussion will include:The Cell BE SDK and its components The Cell BE system simulator Systemsim Installing the SDK Building the SDK Programming to Cell BE Compilers & Tools Samples codeVarious application-oriented source code including Fast Fourier Transform, image, audio resample, math, game math, intrinsics, matrix operation, multi-precision math, noise generation, oscillator, surface, synchronization, and vectorSample port of scalar code to the Cell BE

In order to improve the readability and the automatic recognition of handwritten document images, preprocessing steps are imperative. These steps in addition to conventional steps of noise removal and filtering include text normalization such as baseline correction, slant normalization and skew correction. These steps make the feature extraction process more reliable and effective. Recently Arabic handwriting recognition has received some attention from the research community. Due to the unique nature of the script, the conventional methods do not prove to be effective. In our work, we describe an orientation independent technique for baseline detection of Arabic words. In addition to that we describe, in the rest of the paper, our techniques for slant normalization, slope correction, line and word separation in handwritten Arabic documents. We show how the baseline can be exploited for slope and skew correction before proceeding with the steps of line and word separation.

This paper presents a new probability table compression method based on mixture models applied to N-tuple recognizers. Joint probability tables are modeled by lower dimensional probability mixtures and their mixture coefficients. The maximum likelihood parameters of the mixture models are trained by the expectation-maximization (EM) algorithm and quantized to one byte integers. The probability elements which mixture models do not estimate reliably are kept separately. Experimental results with on-line handwritten UNIPEN digits show that the total memory size of an N-tuple recognizer is reduced from 11.8 Mbytes to 0.55 Mbytes, while the recognition rate drops from 97.7% to 97.5%.

This paper presents a new probability table memory compression method based on mixture models and its application to N-tuple recognizers and N-gram character language models. Joint probability tables are decomposed into lower dimensional probability components and their mixtures. The maximum likelihood parameters of the mixture models are trained by the expectation maximization (EM) algorithm and quantized to one byte integers. Probability elements that mixture models do not estimate reliably are kept separately. Experimental results with on-line handwritten UNIPEN uppercase and lowercase characters show that the total memory size of an on-line scanning N-tuple recognizer is reduced from 12.3 MB to 0.66 MB bytes, while the recognition rate drops from 91.64% to 91.13% for uppercase characters and from 88.44% to 87.31% for lowercase characters. The N-gram character language model was compressed from 73.6 MB to 0.58 MB with minimal reduction in performance.

This paper presents a new probabilistic approachto document retrieval based on the assumption thata Markov process can explain the process by whichhumans rank the relevance of documents to queries.The model ranks documents for retrieval based on theirprobability of relevance. Two training methods are presented. The model is compared with Latent Semantic Analysis (LSA) on two publicly available databases.The results show that the new algorithm achieves Precision/Recall performance equivalent to or better than LSA.

We apply confidence-scoring techniques to verify the output of an off-line handwritten-character recognizer. We evaluate a variety of scoring functions, including likelihood ratios and estimated posterior probabilities of correctness, in a post-processing mode, to generate confidence scores. Using the post-processor in conjunction with a neural-net-based recognizer, on mixed-case letters, receiver-operating-characteristic (ROC) curves reveal that our post-processor is able to reject correctly 90% of recognizer errors while only falsely rejecting 18.6% of correctly-recognized letters. For isolated-digit recognition, we achieve a correct rejection rate of 95% while keeping false rejection down to 8.7%.

This work describes an online handwriting recognition system working in combination with an offline recognizer. The online input data is first transcribed by theonline recognizer, then converted into an offline bitmapand recognized by the offline system. The outputs ofthe two recognizers are then combined probabilisticallyresulting in a classifier out-performing both individualsystems. Experiments were performed over a databaseof single digits. The error rate of the online recognizeris reduced by 43% when the combination with the offlinesystem is applied.

We investigate a novel method for adaptively improving the machine recognition of handwritten words by applying a k-nearest neighbor (k-NN) classifier to the N-best word-hypothesis lists generated by a writer-independent hidden Markov model (HMM). Each new N-best list from the HMM is compared to the N-best lists in the k-NN classifier. A decision module is used to select between the output of the HMM and the matches found by the k-NN classifier. The N-best list chosen by the decision module can be automatically added to the k-NN classifier if it is not already in the k-NN classifier. This dynamic update of the k-NN classifier enables the system to adapt to new data without retraining. On a writer-independent set of 1158 handwritten words, this method reduces the error rate by approximately 30%. This method is fast and memory-efficient, and lends itself to many interesting generalizations.

In this paper, we present a framework for adapting a writer independent system to a user from samples of the user's writing. The writer independent system is modeled using hidden Markov models. Training for a writer involves recomputing the topology and parameters of the hidden Markov models using the writer's data. The framework uses the writer independent system to get an initial alignment of the writer's data. The system described reduces the error rate by an average of 65%. For the results presented, no language model was used.

We describe experiments on a writer dependent large vocabulary (20000) handwriting recognition system. The goal was to investigate the effects of different degrees of tying (and hence, of the number) of parameters on the error rate. The system recognizes cursive, printed or any combination thereof of script in real time on small PC platforms. Baseline results for a writer independent system are also included. Since we are only interested in the shape models no language models were used.

Article Regression NSS: an alternative to cross validation Share on Authors: Michael P. Perrone IBM T.J. Watson Research Center, Yorktown Heights, NY and Institute for Brain and Neural Systems, Brown University IBM T.J. Watson Research Center, Yorktown Heights, NY and Institute for Brain and Neural Systems, Brown UniversityView Profile , Brian S. Blais Institute for Brain and Neural Systems, Brown University, Providence, RI Institute for Brain and Neural Systems, Brown University, Providence, RIView Profile Authors Info & Claims COLT '95: Proceedings of the eighth annual conference on Computational learning theoryJuly 1995 Pages 385–391https://doi.org/10.1145/225298.225345Online:05 July 1995Publication History 0citation154DownloadsMetricsTotal Citations0Total Downloads154Last 12 Months0Last 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

In this paper we present a new version of the standard multilayer perceptron (MLP) algorithm for the state-of-the-art in neural net(cid:173) work VLSI implementations: the Intel Ni1000. This new version of the MLP uses a fundamental property of high dimensional spaces which allows the 12-norm to be accurately approximated by the It -norm. This approach enables the standard MLP to utilize the parallel architecture of the Ni1000 to achieve on the order of 40000, 256-dimensional classifications per second. 1 The Intel NilOOO VLSI Chip The Nestor/Intel radial basis function neural chip (Ni1000) contains the equivalent of 1024 256-dimensional artificial digital neurons and can perform at least 40000 classifications per second [Sullivan, 1993]. To attain this great speed, the Ni1000 was designed to calculate "city block" distances (Le. the II-norm) and thus to avoid the large number of multiplication units that would be required to calculate Euclidean dot products in parallel. Each neuron calculates the city block distance between its stored weights and the current input: neuron activity = L IWi - :eil (1) where w, is the neuron's stored weight for the ith input and :ei is the ith input. Thus the Nil000 is ideally suited to perform both the RCE [Reillyet al., 1982] and 748 Michael P. Perrone. Leon N. Cooper PRCE [Scofield et al., 1987] algorithms or any of the other commonly used radial basis function (RBF) algorithms. However, dot products are central in the calcula(cid:173) tions performed by most neural network algorithms (e.g. MLP, Cascade Correlation, etc.). Furthermore, for high dimensional data, the dot product becomes the compu(cid:173) tation bottleneck (i.e. most ofthe network's time is spent calculating dot products). If the dot product can not be performed in parallel there will be little advantage using the NilOOO for such algorithms. In this paper, we address this problem by showing that we can extend the NilOOO to many of the standard neural network algorithms by representing the Euclidean dot product as a function of Euclidean norms and by then using a city block norm approximation to the Euclidean norm. Section 2, introduces the approximate dot productj Section 3 describes the City Block MLP which uses the approximate dot productj and Section 4 presents ex(cid:173) periments which demonstrate that the City Block MLP performs well on the NIST OCR data and on human face recognition data. 2 Approximate Dot Product Consider the following approximation [Perrone, 1993]: 1