Machine Learning

Canonical Stratification for Non-Mathematicians

October 17, 2018

Our recent publication “Algorithmic Canonical Stratifications of Simplicial Complexes” proposes a new algorithm for data analysis that offers a topology-aware path towards explainable artificial intelligence. Despite (or, perhaps, due to) being mathematically rigorous, the text of the original work is virtually impenetrable for readers not familiar with the concepts, tools, and notation of topology. In order to convey our ideas to a wider audience, we present this supplemental introduction. Here, we summarize and explain in plain English the motivation, reasoning, and methods of our new topological data analysis algorithm that we term “canonical stratification”. Canonical-Stratification-for-Non-Mathematicians.pdf (38 KB) Table of Contents 1. Motivation 2. More on Canonical Stratification 3. Conclusion 1. Motivation Machine learning has advanced significantly in recent years and has proven itself to be a powerful and versatile tool in a variety of data-driven disciplines. Machine learning algorithms are now being used to make decisions in numerous areas [...]

Optimization of Real-Time Object Detection on Intel® Xeon® Scalable Processors

November 11, 2017

This publication demonstrates the process of optimizing an object detection inference workload on an Intel® Xeon® Scalable processor using TensorFlow. This project pursues two objectives: Achieve object detection with real-time throughput (frame rate) and low latency Minimize the required computational resources In this case study, a model described in the “You Only Look Once” (YOLO) project is used for object detection. The model consists of two components: a convolutional neural network and a post-processing pipeline. In this work, the original Darknet model is converted to a TensorFlow model. First, the convolutional neural network is optimized for inference. Then array programming with NumPy and TensorFlow is implemented for the post-processing pipeline. Finally, environment variables and other configuration parameters are tuned to maximize the computational performance. With these optimizations, real-time object detection is achieved while using a fraction of the available processing power of an Intel Xeon Scalable processor-based system. [...]

Optimization of Hamerly’s K-Means Clustering Algorithm: CFXKMeans Library

July 21, 2017

This publication describes the application of performance optimizations techniques to Hamerly’s K-means clustering algorithm. Starting with an unoptimized implementation of the algorithm, we discuss: Thread scheduling Reduction patterns SIMD reduction Unroll and jam Presented optimizations aggregate to 85.6x speedup compared to the original unoptimized implementation. Resulting implementation is packaged into a library named CFXKMeans with interfaces for C/C++ and Python. The Python interface is benchmarked using the MNIST 784 data set. The result for K=64 is compared to the performance of K-means clustering implementation in a popular machine learning framework, scikit-learn, from the Intel distribution for Python. CFXKMeans performed our benchmark tests faster than scikit-learn by a factor of 4.68x on an Intel Xeon processor E5-2699 v4 and 5.54x on an Intel Xeon Phi 7250 processor. The CFXKMeans library has C/C++ and Python API and is available under the MIT license at https://github.com/ColfaxResearch/CFXKMeans. Colfax-Kmeans-Clustering-Optimization.pdf (365 KB) [...]

FALCON Library: Fast Image Convolution in Neural Networks on Intel Architecture

November 9, 2016

We describe FALCON, an original open-source implementation of image convolution with a 3×3 filter based on Winograd’s minimal filtering algorithm. Compared to direct convolution, Winograd’s algorithm reduces the number of arithmetic operations at the cost of complicating the memory access pattern. This study is carried out in the context of image analysis in convolutional neural networks. Our implementation combines C language code with BLAS function calls for general matrix-matrix multiplication. The code is optimized for Intel Xeon Phi processors x200 (formerly Knights Landing) with Intel Math Kernel Library (MKL) used for BLAS call to the SGEMM function. To test the performance of FALCON in the context of machine learning, we benchmarked it for a set of image and filter sizes corresponding to the VGG Net architecture. In this test, FALCON achieves 10% greater overall performance than convolution from DNN primitives in Intel MKL. However, for some layers, FALCON is faster than MKL by 1.5x, but for other layers slower by as much as 4x. This indicates a possibility of a [...]

Machine Learning on 2nd Generation Intel® Xeon Phi™ Processors: Image Captioning with NeuralTalk2, Torch

June 20, 2016

In this case study, we describe a proof-of-concept implementation of a highly optimized machine learning application for Intel Architecture. Our results demonstrate the capabilities of Intel Architecture, particularly the 2nd generation Intel Xeon Phi processors (formerly codenamed Knights Landing), in the machine learning domain. Download as PDF: Colfax-NeuralTalk2-Summary.pdf (814 KB) or read online below. Code: see our branch of NeuralTalk2 for instructions on reproducing our results (in Readme.md). It uses our optimized branch of Torch to run efficiently on Intel architecture. See also: colfaxresearch.com/get-ready-for-intel-knights-landing-3-papers/ 1. Case Study It is common in the machine learning (ML) domain to see applications implemented with the use of frameworks and libraries such as Torch, Caffe, TensorFlow, and similar. This approach allows the computer scientist to focus on the learning algorithm, leaving the details of performance optimization to the framework. Similarly, the ML frameworks usually rely on a third-party library such as Atlas, CuBLAS, [...]

Introduction to Intel DAAL, Part 2: Distributed Variance-Covariance Matrix Computation

March 28, 2016

This is the part 2 of 3 of an introductory series of publications on the Intel® Data Analytics Acceleration Library (DAAL). DAAL is a data analytics library optimized for modern highly parallel computer architectures such as Intel Xeon and Intel Xeon Phi processors. The goal of this series is to provide developers a technical overview for developing applications using DAAL. In part 1 of the series we discussed how to implement batch mode computation on a single node. In the present publication, we discuss the distributed mode computation. Our discussion will focus both on how and when to implement distributed mode computation with Intel DAAL. As an example workload, we implement an application that uses DAAL to compute a covariance matrix of a set of vectors. We first demonstrate how to use distributed mode with this example. Then, using this example application, we scan the parameter space to determine what parameter ranges benefit from distributed computation. We also demonstrate how the output of this computation may be used in image processing to compute the eigenvectors of [...]

Introduction to Intel DAAL, Part 1: Polynomial Regression with Batch Mode Computation

October 28, 2015

This is the part 1 of 3 of an introductory series of publications on the Intel Data Analytics Acceleration Library (DAAL). DAAL is a data analytics library optimized for modern highly parallel computer architectures such as Intel Xeon and Intel Xeon Phi processors. The goal of this series is to provide developers a technical overview for developing applications using DAAL. In this paper we focus on two aspects of developing an application with Intel DAAL: data management and computation. As a practical example, we implement a simple machine learning application with polynomial regression using the library in the batch computation mode. We demonstrate using this application for data-based prediction of hydrodynamics properties of yachts. The source code and data for the sample application are available for free download. The second and third part of the series will discuss other aspects of data analysis with DAAL. In part 2, we discuss distributed data and computation in conjunction with MPI. In the third part, we discuss the case with multiple data sets and interfacing with a [...]