Carole Dulong

The IA-64 architecture at work

Two key architectural features: predication and control speculation, will enable IA-64 compilers to extract instruction level parallelism. To show how compilers will use IA-64 instructions, the author uses code fragments from the pointer chasing problem, an inherently serial code, and from a nested loop with difficult to predict branches.

Detecting phases in parallel applications on shared memory architectures

Most programs are repetitive, where similar behavior can be seen at different execution times. Algorithms have been proposed that automatically group similar portions of a programs execution into phases, where samples of execution in the same phase have homogeneous behavior and similar resource requirements. In this paper, we examine applying these phase analysis algorithms and how to adapt them to parallel applications running on shared memory processors. Our approach relies on a separate representation of each threads activity. We first focus on showing its ability to identify similar intervals of execution across threads for a single run. We then show that it is effective at identifying similar behavior of a program when the number of threads is varied between runs. This can be used by developers to examine how different phases scale across different number of threads. Finally, we examine using the phase analysis to pick simulation points to guide multithreaded simulation.

Semantic Event Detection using Conditional Random Fields

Semantic event detection is an active research field of video mining in recent years. One of the challenging problems is how to effectively model temporal and multi-modality characteristics of video. In this paper, we employ Conditional Random Fields (CRFs) to fuse temporal multi-modality cues for event detection. CRFs are undirected probabilistic models designed for segmenting and labeling sequence data. Compared with traditional SVM and Hidden Markov Models (HMMs), CRFs based event detection offers several particular advantages including the abilities to relax strong independence assumptions in the state transition and avoid a fundamental limitation of directed graphical models. To detect event, we use a three-level framework based on multi-modality fusion and mid-level keywords. The first level extracts audiovisual features, the mid-level detects semantic keywords, and the high-level infers semantic events from multiple keyword sequences. The experimental results from soccer highlights detection demonstrate that CRFs achieves better performance particularly in slice level measure.

Sampling Strategies for Active Learning in Personal Photo Retrieval

With the advent and proliferation of digital cameras and computers, the number of digital photos created and stored by consumers has grown extremely large. This created increasing demand for image retrieval systems to ease interaction between consumers and personal media content. Active learning is a widely used user interaction model for retrieval systems, which learns the query concept by asking users to label a number of images at each iteration. In this paper, we study sampling strategies for active learning in personal photo retrieval. In order to reduce human annotation efforts in a content-based image retrieval setting, we propose using multiple sampling criteria for active learning: informativeness, diversity and representativeness. Our experimental results show that by combining multiple sampling criteria in active learning, the performance of personal photo retrieval system can be significantly improved

Workload Characterization of a Parallel Video Mining Application on a 16-Way Shared-Memory Multiprocessor System

As video data become more and more pervasive, mining information from multimedia data sources becomes increasingly important, e.g., automatically extracting highlights from soccer game video content. However, the huge computation requirement of mining interested data limits its wide use in practice. Since the hardware imperative behind computer architecture is shifting from uniprocessors to multi-core processors, exploiting thread-level parallelism existing in multimedia mining applications is critical to utilizing the hardware resources and accelerating the complex processing of highlight events detection. In this paper we analyze the view type and playfield detection application, a widely used application in sports video mining systems, and we present several different schemes (task level, data-slicing-level, and a hybrid parallel scheme, as well as variations of the hybrid parallel scheme) for parallelizing this application. The hybrid parallel scheme, which exploits data-level and task-slicing-level parallelism, outperforms basic task-level and data-slicing-level schemes, delivering much better performance in terms of execution time and speedup. On a 16-way shared-memory multi-processing system with hardware prefetch enabled, the hybrid scheme achieves a speedup of 10.6x. Detailed performance analysis shows that because of the large working set, the workload often requires data from the off-chip memory. Therefore, the saturated bus bandwidth utilization is the likely cause of bottlenecks for achieving perfect scalability performance. With hardware prefetch enabled, the bus utilization rate on 16-processors system is about 76% for the hybrid scheme, and the projected bus bandwidth requirement for perfect scalability is about 3.1GB/s for 16 processors and 6.2 GB/s for 32 processors. In addition, our experiments also reveal that there are also no obvious scaling limiting factors, e.g., very low synchronization and load imbalance problems even with up to 16 processors

Requirements for benchmarking personal image retrieval systems

It is now common to have accumulated tens of thousands of personal ictures. Efficient access to that many pictures can only be done with a robust image retrieval system. This application is of high interest to Intel processor architects. It is highly compute intensive, and could motivate end users to upgrade their personal computers to the next generations of processors. A key question is how to assess the robustness of a personal image retrieval system. Personal image databases are very different from digital libraries that have been used by many Content Based Image Retrieval Systems.1 For example a personal image database has a lot of pictures of people, but a small set of different people typically family, relatives, and friends. Pictures are taken in a limited set of places like home, work, school, and vacation destination. The most frequent queries are searched for people, and for places. These attributes, and many others affect how a personal image retrieval system should be benchmarked, and benchmarks need to be different from existing ones based on art images, or medical images for examples. The attributes of the data set do not change the list of components needed for the benchmarking of such systems as specified in2: - data sets - query tasks - ground truth - evaluation measures - benchmarking events. This paper proposed a way to build these components to be representative of personal image databases, and of the corresponding usage models.

Performance analysis of Java concurrent programming: a case study of video mining system

As multi/many core processors become prevalent, programming language is important in constructing efficient parallel applications. In this work, we build a multithreaded video mining application with Java, examine the thread profiling information and micro-architecture metrics to identify the factors limiting the scalability, and employ a number of ways to improve performance. Besides, we conduct some thread scheduling experiments. According to the experiments and detailed analysis, we conclude that for this video mining application: (1) Java is a good parallel language candidate for many core processors in terms of performance, scalability, and ease of programming; (2) Thread affinity mechanism is effective in improving data locality, but brings little benefit to multithreaded Java application due to its conservative memory model in JVM

Dual diffusion model of spreading activation for content-based image retrieval

This paper introduces a content-based information retrieval method inspired by the ideas of spreading activation models. In response to a given query,the proposed approach computes document ranks as their final activation values obtained upon completion of a diffusion process. This diffusion process,in turn,is dual in the sense that it models the spreading of the query s initial activation simultaneously in two similarity domains: low-level feature-based and high-level semantic.The formulation of the diffusion process relies on an approximation that makes it possible to compute the final activation as a solution to a linear system of differential equations via a matrix exponential without the need to resort to an iterative simulation.The latter calculation is performed efficiently by adapting a sparse routine based on Krylov sub-space projection method.The empirical performance of the described dual diffusion model has been evaluated in terms of precision and recall on the task of content-based digital image retrieval in query-by-example scenario. The obtained experimental results demonstrate that the proposed method achieves better overall performance compared to traditional feature-based approaches. This performance improvement is attained not only when both similarity domains are used, but also when a diffusion model operates only on the feature-based similarities.

Towards the Parallelization of Shot Detection - a Typical Video Mining Application Study

As digital video data becomes more pervasive, mining information from multimedia data becomes increasingly important, e.g., extraction of goal events in soccer game automatically from the video content. Though all of these advances in multimedia mining have shown great potential in daily life, the huge computational requirement prohibits its wide use in practice. As computer architecture evolves from uniprocessor to the era of multi-core processors, accelerating the multimedia application by exploiting thread level parallelism would be more promising to boost performance and provide more functionality. This paper presents three different parallel approaches, i.e., task level, data slicing and hybrid parallel scheme, to parallelize shot detection, a widely used application in the video mining system. The hybrid scheme, with the exploration of data level and task level parallelism, delivers much better performance than the other two schemes. Besides, we also employ several software optimization techniques, e.g. data blocking and thread affinity, to improve the performance by more than 50%. Experimental results indicate that there are no obvious parallel limiting factors in the hybrid parallel scheme. It scales well the increasing number of processors, and exhibits 13.6x speedup on 16-way processor system

On Parallelization of a Video Mining System

As digital video data becomes more pervasive, mining information from multimedia data becomes increasingly important. Although researches in multimedia mining area have shown great potential in daily life, the huge computational requirement prohibits its wide use in practice. Since our personal computer is shifting from uniprocessors to multicore processors, exploiting thread level parallelism in multimedia mining applications is critical to utilize the hardware resources and accelerate the mining process. This paper presents three different parallel approaches (task level, data slicing and hybrid parallel) to parallelize one widely used application in video mining system. The hybrid scheme, with the exploration of data level and task level parallelism, delivers much better performance than other two schemes. We get 10x performance improvement on a 16-way multiprocessor system. Besides, we perform several efficient optimization techniques, such as subexpression optimization, SIMD, and data blocking, to improve the performance by more than 60%. Therefore, our parallelization and optimization of the application makes it 16x faster than it used to be. Our study shows that with proper parallelization and optimization, multimedia mining can be used widely in our daily life soon