The Applications Research Lab (ARL) conducts research on next-generation computer architectures designed to handle applications for the future computing environment. We explore emerging computing methods under the Recognition-Mining-Synthesis framework, which includes an interactive visual system (CPU-centric photo-realism rendering, physical-realism motions, and real time computer vision recognition technology), as well as mass market data mining (i.e., data analysis using statistical computing methods). We conduct research on emerging workload-driven scalable system architecture, and characterize parallel workloads and study their impact on future computer systems to help users interactively extract useful knowledge from the massive amount of multi-modal digital data in the digital home, digital enterprise, etc.

ARL works closely with architecture teams on the design of next generation microprocessors, is active in professional conferences, and collaborates extensively with academia. One example is our work with Ohio State University on the paper, Cache-Conscious Frequent Pattern Mining on a Modern Processor, which won the best paper award in the VLDB 2005 conference. Our labs are located in Santa Clara, California and Dupont, Washington, USA; Beijing, China; and Nizhny Novgorod, Russia.

Some of our work is described in the May 19, 2005, issue of the Intel Technology Journal, Volume 3, Issue 2.

Scalable Photo-realistic computer graphics rendering and architecture
We investigate system issues in an effort to achieve interactive photo-realistic rendering through parallel processing such as Ray Tracing. (Also see the Siggraph 2005 paper, Multi-Level Ray Tracing Algorithm [PDF 411KB].) Previously, we worked on image-based rendering methods such as Light Field Mapping.

Machine Learning and Machine Vision
We advance the state of the art in machine learning by focusing on problem areas that directly impact Intel, such as increasing yield in manufacturing and decreasing design cycle time in computer architecture. Externally, we have worked with the academic and industrial research community to enable the use of machine learning and machine vision techniques by releasing the Open Source Computer Vision Library (OpenCV). An introduction to the Open CV Library is in Intel Technology Journal, Volume 3, Issue 2. Our project on Probabilistic Graphical Models/Bayesian Networks is being implemented as the Probabilistic Network Library (PNL).

Computational NanoVision Research
We are developing physics-based Computer Vision techniques for quantitative nanofeature analysis. The goal is to use combinations of (nonstandard) image sensors, and seamlessly integrate reliable physical models of complex phenomena into statistical data analysis. This will bridge the gap between the image formation and image analysis, and establish physics-based Computer Vision as a scientific quantitative instrument.

Workload Analysis
We analyze current and next-generation workloads to provide input to the designing of innovative features for future processors and platforms. In particular, we are researching multithreading techniques, and better ways to partition applications between the different processing units available in future systems. A few examples of our work: the scalability analysis on the large scale optimization problems, the scalability analysis of data mining work in bioinformatics, and workload analysis of a case study in computer vision. We have also applied statistical methods to the phase analysis of the workloads, which are being extended to cover the multi-threaded workloads.

Parallel Computer Architecture for emerging workloads:
We look beyond the current architecture whose primary function is to process character-based textual data and explore features for parallel computer architectures which will simultaneously process various media, graphics, data mining, and pattern recognition/matching workloads in the framework of RMS workloads. We also explore parallel algorithms and how they map onto different workloads, different programming environments, and different architectures. An example of this is our collaborative work with the academic community on Design Pattern Language for Parallel programming.¹

¹ Patterns for Parallel Programming", by Tim Mattson, Beverly Sanders, and Berna Massingill, Software Patterns Series, Addison Wesley, 2004