Naveen Gv Engineering Manager, Intel Software Products Intel Corporation

Transcription

1 Naveen Gv Engineering Manager, Intel Software Products Intel Corporation

2 Artificial Intelligence In our smart and connected world, machines are increasing learning to sense, learn, reason, act and adapt to the real world. This is Artificial Intelligence. Machine Learning (ML) is a computational method that allows machines to act or think without being explicitly directed to perform specific functions. Deep learning is a branch of ML that uses neural network models to understand large amounts of data. It can accelerate processes like image and speech recognition, and natural language recognition. 2

3 Types of Machine Learning Supervised Learning Unsupervised Learning Reinforcement Learning Teach desired behavior with labeled data Make inferences without labeled data Act in an environment to maximize reward Today s focus 3

4 Machine Learning: Example Use Cases Cloud Service Providers Financial Services Healthcare Automotive Image classification and detection Image recognition/tagging Natural language recognition Big data pattern detection Targeted ads Fraud / face detection Gaming, check processing Computer server monitoring Financial forecasting and prediction Network intrusion detection 4

5 What Is Deep Learning? Step 1: Training (In Data Center Over Hours/Days/Weeks) Step 2: Scoring (End point or Data Center - Instantaneous) Lots of labeled input data Person New input from camera and sensors Create Deep neural net math model Trained Model Trained neural network model Output Classification 90% person 8% traffic light 97% person Output Classification 5

6 6

7 Exponential Growth Of Unstructured Data Industry Realities 80% of world s data is unstructured Example: Medical knowledge & data 1950: 50 years to double knowledge 1980: 7 years to double knowledge 2015: <3 years to double knowledge 1M GB of health data generated per person * From IBM Computing, Cognition, & the Future of Knowing by Dr. John E Kelly III Consequences Rise of connected things & machines Tremendous data that needs to be automatically analyzed & interpreted Definition: Unstructured Data refers to information that either does not have a pre-defined data model or is not organized in a pre-defined manner. Source: IDC Survey How do we analyze & interpret all that data? 7

8 Machine Learning Software Challenges Machine learning open source frameworks and libraries are often not well-optimized for newer Intel-based systems Frameworks can be difficult to configure and use Need to target heterogeneous hardware from model training in the datacenter to deployment on Datacenter Endpoint endpoint systems 8

9 Essential Ingredients can be Hard to Get (Massive datasets, specialized talent, optimized data tools,etc.) Pictures of big images dataset/ voices etc How much data is enough? Can I use my current datasets? How do I know my analytics is accurate? Data scientist are really expensive I need someone who understands my vertical and create solution for my business Do I need to hire new solution developers? Can I use my existing infrastructure? Do I need to invest in accelerators for ML? How do I know I am getting the best TCO? Cost reduction scenarios described are intended as examples of how a given Intel- based product, in the specified circumstances and configurations, may affect future costs and provide cost savings. Circumstances will vary. Intel does not guarantee any costs or cost reduction. Results have been estimated or simulated using internal Intel analysis or architecture simulation or modeling, and provided to you for informational purposes. Any differences in your system hardware, software or configuration may affect your actual performance. 9

10 Evolving Software Ecosystem Top Frameworks Caffe Programming languages C/C++ *Other names and brands may be claimed as property of others. 10

11 Machine Learning: Your Path to Deeper Insight Driving increasing innovation and competitive advantage across industries Solutions for reference across industries Tools/Platforms to accelerate deployment Trusted Analytics Platform Intel Deep Learning SDK for Training & Deployment strategy provides the foundation for success using AI Optimized Frameworks to simplify development Libraries/Languages featuring optimized building blocks Hardware Technology portfolio that is broad and crosscompatible Intel Math Kernel Library (Intel MKL & MKL-DNN) Datacenter Intel Data Analytics Acceleration Library (Intel DAAL) Endpoint Intel Distribution for Python* +Network +Memory +Storage 11

12 Machine Learning SW Investments & Positioning Trusted Analytics Platform Open Source Frameworks Intel Math Kernel Library (Intel MKL) Intel Data Analytics Acceleration Library (Intel DAAL) High Level Overvie w Single platform from Data Science to Application Development Toolkits driven by Academia and Industry for scripting and training ML algorithms High Performance Math Primitives granting low level of control Broad Data Analytics Acceleration object oriented library supporting distributed ML at the algorithm level Target audience Application Developers and Data Scientists Machine Learning Researchers and Data Scientists Consumed by developers of higher level libraries and Applications Wider Data Analytics and ML audience, Algorithm level development, needing predesigned algorithms for all stages of data analytics Example Usage Application creation from the Big Data infrastructure, Data Science tools all the way to app development Script and Train a Convolution Neural Network for Image Recognition in Python* Call Matrix Multiplication, Convolution Functions Call K-Means, Linear Regression, ALS Algorithms 12

13 Intel Math Kernel Library (Intel MKL) Introduction Highly optimized threaded math routines Performance, Performance, Performance! Industry s leading math library Widely used in science, engineering, data processing Tuned for Intel processors current and next generation EDC North America Development Survey 2016, Volume I More math library users depend on MKL than any other library Intel Engineering Algorithm Experts Optimized code path dispatch auto Be multiprocessor aware Cross-Platform Support Intel Compatible Processors Past Intel Processors Current Intel Processors Future Intel Processors Be vectorised, threaded, and distributed multiprocessor aware 13

14 Performance (GFlops) Performance (GFlops) Intel MKL unleashes the performance benefits of Intel architectures DGEMM Performance Boost by using Intel MKL vs. ATLAS* Intel Core Processor i7-4770k Intel Xeon Processor E v Matrix size (M = 10000, N = 6000, K = 64,80,96,, 384) Intel MKL - 1 thread Intel MKL - 2 threads Intel MKL - 4 threads ATLAS - 1 thread ATLAS - 2 threads ATLAS - 4 threads Matrix size (M = N) Intel MKL - 1 thread Intel MKL - 18 threads Intel MKL - 36 threads ATLAS - 1 thread ATLAS - 18 threads ATLAS - 36 threads Configuration Info - Versions: Intel Math Kernel Library (Intel MKL) 11.3, ATLAS* ; Hardware: Intel Xeon Processor E5-2699v3, 2 Eighteen-core CPUs (45MB LLC, 2.3GHz), 64GB of RAM; Intel Core Processor i7-4770k, Quad-core CPU (8MB LLC, 3.5GHz), 8GB of RAM; Operating System: RHEL 6.4 GA x86_64; Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such as SYSmark and MobileMark, are measured using specific computer systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. * Other brands and names are the property of their respective owners. Benchmark Source: Intel Corporation : Intel s compilers may or may not optimize to the same degree for non-intel microprocessors for optimizations that are not unique to Intel microprocessors. These optimizations include SSE2, SSE3, and SSSE3 instruction sets and other optimizations. Intel does not guarantee the availability, functionality, or effectiveness of any optimization on microprocessors not manufactured by Intel. Microprocessor-dependent optimizations in this product are intended for use with Intel microprocessors. Certain optimizations not specific to Intel microarchitecture are reserved for Intel microprocessors. Please refer to the applicable product User and Reference Guides for more information regarding the specific instruction sets covered by this notice. Notice revision #

15 Performance (GFlops) Intel MKL 2017 Performance Performance (GFlops) DGEMM Performance On Intel Xeon Processor E v4 DGEMM Performance On Intel Xeon Phi Processor Matrix size (M = N) Matrix size (M = N) Intel MKL - 1 thread Intel MKL - 44 threads Intel MKL - 22 threads Intel MKL - 16 threads Intel MKL - 68 threads Intel MKL - 34 threads Configuration Info - Versions: Intel Math Kernel Library (Intel MKL) 2017; Hardware:Hardware: Intel Xeon Processor E v4, 2 Twenty-two-core CPU (55MB smart cache, 2.2GHz), 64GB of RAM; Intel Xeon Phi Processor 7250, 68 cores (34MB L2 cache, 1.4GHz), 96 GB of DDR4 RAM and 16 GB MCDRAM; Operating System: RHEL 7.2 GA x86_64; Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such as SYSmark and MobileMark, are measured using specific computer systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. * Other brands and names are the property of their respective owners. Benchmark Source: Intel Corporation : Intel s compilers may or may not optimize to the same degree for non-intel microprocessors for optimizations that are not unique to Intel microprocessors. These optimizations include SSE2, SSE3, and SSSE3 instruction sets and other optimizations. Intel does not guarantee the availability, functionality, or effectiveness of any optimization on microprocessors not manufactured by Intel. Microprocessor-dependent optimizations in this product are intended for use with Intel microprocessors. Certain optimizations not specific to Intel microarchitecture are reserved for Intel microprocessors. Please refer to the applicable product User and Reference Guides for more information regarding the specific instruction sets covered by this notice. Notice revision #

16 Components of Intel MKL 2017 New Linear Algebra Fast Fourier Transforms Vector Math Summary Statistics And More Deep Neural Networks BLAS LAPACK ScaLAPACK Sparse BLAS Sparse Solvers Iterative PARDISO* Cluster Sparse Solver Multidimensional FFTW interfaces Cluster FFT Trigonometric Hyperbolic Exponential Log Power Root Vector RNGs Kurtosis Variation coefficient Order statistics Min/max Variancecovariance Splines Interpolation Trust Region Fast Poisson Solver Convolution Pooling Normalization ReLU Softmax 16

17 Intel Math Kernel Library and Intel MKL-DNN for Deep Learning Framework Optimization Caffe Intel Math Kernel Library (Intel MKL) Deep Learning Frameworks Intel MKL-DNN Xeon Xeon Phi FPGA Intel MKL DNN primitives + wide variety of other math functions C DNN APIs Binary distribution Free community license. Premium support available as part of Parallel Studio XE Broad usage DNN primitives; not specific to individual frameworks Quarterly update releases Intel MKL-DNN DNN primitives C/C++ DNN APIs Open source DNN code* Apache 2.0 license Multiple variants of DNN primitives as required for framework integrations Rapid development ahead of Intel MKL releases * GEMM matrix multiply building blocks are binary

18 Deep Neural Network (DNN) Primitives in Intel MKL 2017 Operations Algorithms Activation ReLU Normalization batch, local response Pooling max, min, average Convolutional fully connected, direct 3D batched convolution Inner product forward/backward propagation of inner product computation Data manipulation layout conversion, split, concat, sum, scale 18

19 DNN Primitives in Intel MKL Highlights A plain C API to be used in the existing DNN frameworks Brings IA-optimized performance to popular image recognition topologies: AlexNet, Visual Geometry Group (VGG), GoogleNet, and ResNet Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such as SYSmark and MobileMark, are measured using specific computer systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. For more information go to *Other names and brands may be property of others Configurations: 2 socket system with Intel Xeon Processor E v4 (22 Cores, 2.2 GHz,), 128 GB memory, Red Hat* Enterprise Linux 6.7, BVLC Caffe, Intel Optimized Caffe framework, Intel MKL , Intel MKL 2017 Intel Xeon Phi Processor 7250 (68 Cores, 1.4 GHz, 16GB MCDRAM), 128 GB memory, Red Hat* Enterprise Linux 6.7, Intel Optimized Caffe framework, Intel MKL 2017 All numbers measured without taking data manipulation into account. 19

20 Product Portfolio Training Scoring Intel Xeon Processors Optimized for performance/tco Most widely deployed scoring solution Intel Xeon Phi Processors Optimized for performance Scales with cluster size for shorter time to model x86 architecture, consistent programming model for training and scoring Intel Xeon Processors + FPGA (discrete) Optimized for performance/watt Reconfigurable can be used to accelerate many DC workloads Programmable with industry standard OpenCL Cost reduction scenarios described are intended as examples of how a given Intel- based product, in the specified circumstances and configurations, may affect future costs and provide cost savings. Circumstances will vary. Intel does not guarantee any costs or cost reduction. Results have been estimated or simulated using internal Intel analysis or architecture simulation or modeling, and provided to you for informational purposes. Any differences in your system hardware, software or configuration may affect your actual performance. 20

21 Scientific/Engineering Web/Social Business Intel Data Analytics Acceleration Library (Intel DAAL) An industry leading Intel Architecture based data analytics acceleration library of fundamental algorithms covering all machine learning stages. Pre-processing Transformation Analysis Modeling Validation Decision Making (De-)Compression Outlier Detection Normalization PCA Statistical moments Variance matrix Pp-QR, SVD, Cholesky Apriori Sorting Ridge linear regression Naïve Bayes SVM Classifier boosting Kmeans EM GMM Collaborative filtering Neural Networks 21

22 Problem Statement Data sources Finance Social media Marketing IoT Mfg SQL stores NoSQL stores In-memory stores Connectors Data mining Recommendation engines Big Spark* data analytics MLLib Current common practice Run on state-of-art hardware Built Netlib-Java with a patchwork of math libs Under-exploiting hardware performance features JNI Breeze JVM Customer behavior modeling Netlib BLAS BI analytics Real time analytics Limited performance Many layers of dependencies Low ROI on HW investment Big data frameworks: Hadoop, Spark, Cassandra, etc. 22

23 Desired Solution Data sources Finance Social media Marketing IoT Mfg SQL stores NoSQL stores In-memory stores Connectors Data mining Recommendation engines Customer behavior modeling BI analytics Big data analytics Desired practice Optimized Run on state-of-art hardware Single library to cover all stages of data analytics Fully optimized for underlying hardware Real time analytics performance Simpler development & deployment High ROI on HW investment Big data frameworks: Hadoop, Spark, Cassandra, etc. 23

24 Where Intel DAAL Fits? Data sources Finance SQL stores Data mining Recommendation engines Customer behavior modeling BI analytics Real time analytics Social media Marketing NoSQL stores Connectors Intel Data Analytics Acceleration Library Building end-to-end data applications C++, Java, and Python APIs IoT Mfg In-memory stores Big data frameworks: Hadoop, Spark, Cassandra, etc. 24

25

26 Intel DAAL Components Data Management Interfaces for data representation and access. Connectors to a variety of data sources and data formats, such HDFS, SQL, CSV, KDB+, and user-defined data source/format Data Sources Compression / Decompression Numeric Tables Serialization / Deserialization Data Processing Algorithms Optimized analytics building blocks for all data analysis stages, from data acquisition to data mining and machine learning Data Modeling Data structures for model representation, and operations to derive model-based predictions and conclusions

27 Algorithms Batch Distributed Online Descriptive statistics Low order moments Quantiles/sorting Statistical relationships Correlation / Variance-Covariance (Cosine, Correlation) distance matrices SVD Matrix decomposition Cholesky QR Regression Linear/ridge regression Multinomial Naïve Bayes Classification SVM (two-class and multi-class) Boosting (Ada, Brown, Logit) Association rules mining (Apriori) Anomaly detection (uni-/multi-variate) Unsupervised learning PCA KMeans EM for GMM Recommender systems ALS Deep learning Fully connected, convolution, normalization, activation layers, model, NN, optimization solvers, 27

28 Processing modes in Intel DAAL Batch Processing Online Processing Distributed Processing R 1 D 1 D k D k- 1 Append D 1 D 3 D 2 D 1 S i,r i D 2 R 2 R R = F(D 1,,D k ) S i+1 = T(S i,d i ) R i+1 = F(S i+1 ) D k R k R = F(R 1,,R k ) 28

29 Key Features in Intel DAAL 2017 Neural Networks API Python API (a.k.a. PyDAAL) Easy installation through Anaconda Intel Distribution for Python 2017 Open source project on GitHub Fork me on GitHub: 29

30 Speedup SVM Performance Boosts Using Intel DAAL vs. Scikit-learn on Intel CPU X X Training Prediction Configuration Info - Versions: Intel Data Analytics Acceleration Library 2016 U2, scikit-learn ; Hardware: Intel Xeon E GHz, 24 cores, 30 MB L3 cache per CPU, 256 GB RAM; Operating System: Red Hat Enterprise Linux Server release 6.6, 64-bit. Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such as SYSmark and MobileMark, are measured using specific computer systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. * Other brands and names are the property of their respective owners. Benchmark Source: Intel Corporation : Intel s compilers may or may not optimize to the same degree for non-intel microprocessors for optimizations that are not unique to Intel microprocessors. These optimizations include SSE2, SSE3, and SSSE3 instruction sets and other optimizations. Intel does not guarantee the availability, functionality, or effectiveness of any optimization on microprocessors not manufactured by Intel. Microprocessor-dependent optimizations in this product are intended for use with Intel microprocessors. Certain optimizations not specific to Intel microarchitecture are reserved for Intel microprocessors. Please refer to the applicable product User and Reference Guides for more information regarding the specific instruction sets covered by this notice. Notice revision #

31 Intel DAAL+ Intel MKL = Complementary Big Data Libraries Solution Intel MKL C and Fortran API Primitive level Processing of homogeneous data in single or double precision Intel DAAL Python, Java & C++ API High-level Processing heterogeneous data (mix of integers and floating point), internal conversions are hidden in the library Type of intermediate computations is defined by type of input data (in some library domains higher precision can be used) Type of intermediate computations can be configured independently of the type of input data Most of MKL supports batch computation mode only Cluster functionality uses MPI internally 3 computation modes: Batch, streaming and distributed Developer chooses communication method for distributed computation (e.g. Spark, MPI, etc.) Code samples provided. 31

32 Highlights: Intel Distribution for Python* 2017 Focus on advancing Python performance closer to native speeds Easy, out-of-the-box access to high performance Python Prebuilt, accelerated Distribution for numerical & scientific computing, data analytics, HPC. Optimized for IA Drop in replacement for your existing Python. No code changes required Drive performance with multiple optimization techniques Accelerated NumPy/SciPy/scikit-learn with Intel Math Kernel Library Data analytics with pydaal, Enhanced thread scheduling with TBB, Jupyter* notebook interface, Numba, Cython Scale easily with optimized mpi4py and Jupyter notebooks Faster access to latest optimizations for Intel architecture Distribution and individual optimized packages available through conda and Anaconda Cloud Optimizations upstreamed back to main Python trunk 32

33 A two pronged approach for faster Python performance Accelerated Python Distribution + Performance profiling Step 1: Use Intel Distribution for Python* Leverage optimized native libraries for performance Drop-in replacement for your current Python Latest optimizations for Intel processors and compilers Step 2: Use Intel VTune Amplifier for profiling Get detailed summary of entire application execution profile Auto-detects & profiles Python/C/C++ mixed code & extensions Accurately detect hotspots. Line level analysis helps make smart optimization decisions fast! Available in Intel Parallel Studio XE 2017 suite 33

34 Tune Python + native code for better Performance Intel VTune Amplifier 2017, a performance analyzer in Intel Parallel Studio XE suite Challenge Full profile of Python + native applications Detect inefficient runtime execution Solution Accurately identify performance hotspots at line-level Auto-detect mixed Python/C/C++ code and extensions Focus your tuning efforts for most impact on performance Insert screenshot image Auto detection and performance analysis of Python and native functions Available in Intel VTune Amplifier 2017 Beta & Intel Parallel Studio XE 2017 Download beta at 34

35 Access multiple options for faster Python Included in Intel Distribution for Python* Accelerate with native libraries NumPy, SciPy, Scikit-Learn, Theano, Pandas, pydaal Intel MKL, Intel DAAL Exploit vectorization and threading Cython + Intel C++ compiler Numba + Intel LLVM Better/Composable threading Cython, Numba, Pyston Multi-node parallelism Mpi4Py, Distarray Intel native libraries: Intel MPI Integration with Big Data, ML platforms and frameworks Spark, Hadoop, Trusted Analytics Platform Better performance profiling Extensions for profiling mixed Python & native/jit codes Threading composability for MKL, CPython, Blaze/Dask, Numba

36

37 Intel Deep Learning Software Stack Intel Deep Learning SDK Intel Deep Learning SDK free tools to accelerate design, training and deployment of deep networks Caffe * Intel Optimized Frameworks * * * * Intel optimized frameworks single-node and multi-node CPU optimizations of the most widely used deep learning frameworks Intel Math Kernal Library (Intel MKL) Intel Data Analytics Acceleration Library (Intel DAAL) Intel MKL extracts max Intel hardware performance and provide common interface to all Intel accelerators Intel DAAL provides optimized building blocks for machine learning applications on Intel platforms Xeon Xeon Phi FPGA Intel libraries as path to bring optimized ML/DL frameworks to Intel hardware *Other names and brands may be claimed as property of others. 37

38 Intel Deep Learning SDK Deep Learning Training Tool Deep Learning Deployment Tool configure_nn(fpga/cve, ) allocate_buffer( ) fpga_conv(input,output); fpga_conv( ); mkl_softmax( ); mkl_softmax( ); INSTALL / SELECT IA-Optimized Frameworks PREPARE / CREATE Dataset with Ground-truth DESIGN / TRAIN Model(s) with IA-Opt. Hyper-Parameters MONITOR Training Progress across Candidate Models EVALUATE Results and ITERATE IMPORT Trained Model (trained on Intel or 3 rd Party HW) COMPRESS Model for Inference on Target Intel HW GENERATE faster inference with HW-Specific runtime INTEGRATE with System SW / Application Stack & TUNE EVALUATE Results and ITERATE Technical preview available at software.intel.com/deep-learning-sdk 38

39 Intel Deep Learning SDK Benefits Accelerate Your Deep Learning Solution Plug & Train/Deploy Maximum Performance Increased Productivity Simplify installation & preparation of deep learning models using popular deep learning frameworks on Intel hardware Optimized performance for training and inference on Intel Architecture Faster Time-to-market for training and inference, Improve model accuracy, Reduce total cost of ownership 39

40 Case Study: Cloud service provider* Illegal Video Detection Cloud provides illegal video detection service to 3 rd video cloud customer to help them detect the illegal videos Originally, Cloud providers adopted open source BVLC Caffe + OpenBlas as CNN framework, but saw poor performance Intel Optimized Caffe* Optimized Performance of Cloud * Illegal Video Detection - higher is better By using Intel Optimized Caffe plus Intel Math Kernel Library, got up to 30X performance improvement for training in production environment 10 0 BVLC Caffe + OpenBlas (Baseline) Intel Optimized Caffe + MKL * The test data is based on Intel Xeon E V3 processor Extract key frame Input Classification Pornogr aphic Sexual Normal Video Pictures CNN Network based on Caffe Picture Scoring Video Classification Cloud Illegal Video Detection Process Flow

41 Intel Data Analytics Acceleration Library (Intel DAAL) New High Performance Library for Big Data Analytics Library of Intel Architecture (IA) optimized analytics building blocks Key ingredient for proprietary and open source data analytics platforms and applications Fundamental building blocks for all data analysis stages Delivers forward-scaling performance and parallelism on IA Built upon Intel Math Kernel Library and Intel Integrated Performance Primitives Intel DAAL dramatically speeds up PCA computing speed to reduce data dimension from 4096 to 128 on big data platform Saves up to 4.57x Time (Second, the smaller the better) 480 Intel Data Analytics Acceleration Library (Intel DAAL) Big Data Platform: Spark * 0 Intel DAAL PCA Spark Native PCA The test data is based on Intel Xeon 2.4GHz, 2 sockets, 64 GB RAM, 2 node Hardware Layer: IA (Intel Xeon processor) Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such as SYSmark and MobileMark, are measured using specific computer systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. For more information go to

42 DNN Demo for Intel High Performance Library 42

43 Intel Nervana AI Academy Resources for Developers & Students 1. Intel Developer Zone for AI: Online community to reach 1.5M developers, data scientists and students with frameworks, tools, trainings and support. 2. Training: Partnership with Coursera to deliver series of online AI courses starting Q1 17 plus 100 AI workshops, webinars, meet-ups world wide. 3. Competitions: Intel has partnered with MobileODT, a company dedicated to enabling cervical cancer screening for every woman, everywhere, to host an AI Competition on Kaggle.com, starting February Student Developers: Intel Student Ambassadors leading AI Clubs and Workshops at universities world wide. 43

44 PARTICIPATE IN THE INTEL HPC DEVCON 2016 AND ACHIEVE GREATER INSIGHTS TO DEVELOP A SMARTER FUTURE! 44

45

46

47

48 Legal Disclaimer & INFORMATION IN THIS DOCUMENT IS PROVIDED AS IS. NO LICENSE, EXPRESS OR IMPLIED, BY ESTOPPEL OR OTHERWISE, TO ANY INTELLECTUAL PROPERTY RIGHTS IS GRANTED BY THIS DOCUMENT. INTEL ASSUMES NO LIABILITY WHATSOEVER AND INTEL DISCLAIMS ANY EXPRESS OR IMPLIED WARRANTY, RELATING TO THIS INFORMATION INCLUDING LIABILITY OR WARRANTIES RELATING TO FITNESS FOR A PARTICULAR PURPOSE, MERCHANTABILITY, OR INFRINGEMENT OF ANY PATENT, COPYRIGHT OR OTHER INTELLECTUAL PROPERTY RIGHT. Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such as SYSmark and MobileMark, are measured using specific computer systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. Intel, Pentium, Xeon, Xeon Phi, Core, VTune, Cilk, and the Intel logo are trademarks of Intel Corporation in the U.S. and other countries. Intel s compilers may or may not optimize to the same degree for non-intel microprocessors for optimizations that are not unique to Intel microprocessors. These optimizations include SSE2, SSE3, and SSSE3 instruction sets and other optimizations. Intel does not guarantee the availability, functionality, or effectiveness of any optimization on microprocessors not manufactured by Intel. Microprocessor-dependent optimizations in this product are intended for use with Intel microprocessors. Certain optimizations not specific to Intel microarchitecture are reserved for Intel microprocessors. Please refer to the applicable product User and Reference Guides for more information regarding the specific instruction sets covered by this notice. Notice revision #

49

50 Case Study: Cloud service provider* Illegal Video Detection Cloud provides illegal video detection service to 3 rd video cloud customer to help them detect the illegal videos Originally, Cloud providers adopted open source BVLC Caffe + OpenBlas as CNN framework, but saw poor performance Intel Optimized Caffe* Optimized Performance of Cloud * Illegal Video Detection - higher is better By using Intel Optimized Caffe plus Intel Math Kernel Library, got up to 30X performance improvement for training in production environment 10 0 BVLC Caffe + OpenBlas (Baseline) Intel Optimized Caffe + MKL * The test data is based on Intel Xeon E V3 processor Extract key frame Input Classification Pornogr aphic Sexual Normal Video Pictures CNN Network based on Caffe Picture Scoring Video Classification Cloud Illegal Video Detection Process Flow

51 Intel Data Analytics Acceleration Library (Intel DAAL) New High Performance Library for Big Data Analytics Library of Intel Architecture (IA) optimized analytics building blocks Key ingredient for proprietary and open source data analytics platforms and applications Fundamental building blocks for all data analysis stages Delivers forward-scaling performance and parallelism on IA Built upon Intel Math Kernel Library and Intel Integrated Performance Primitives Intel DAAL dramatically speeds up PCA computing speed to reduce data dimension from 4096 to 128 on big data platform Saves up to 4.57x Time (Second, the smaller the better) 480 Intel Data Analytics Acceleration Library (Intel DAAL) Big Data Platform: Spark * 0 Intel DAAL PCA Spark Native PCA The test data is based on Intel Xeon 2.4GHz, 2 sockets, 64 GB RAM, 2 node Hardware Layer: IA (Intel Xeon processor) Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such as SYSmark and MobileMark, are measured using specific computer systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. For more information go to

52 DNN Demo for Intel High Performance Library 52