Download as PDF, PPTX
Uploaded byCarol McDonald
Introduction to machine learning with GPUs
The document provides an introduction to machine learning concepts including supervised and unsupervised learning. It discusses classification and regression as examples of supervised learning techniques and clustering as an example of unsupervised learning. It also provides an overview of deep learning using neural networks and examples of convolutional neural networks and recurrent neural networks. The document emphasizes how GPUs have accelerated machine learning by enabling parallel processing.
More Related Content
Similar to Introduction to machine learning with GPUs
Introduction to machine learning with GPUs
- 1.
- 2.
- 3. 3 Introduction to MachineLearning Supervised Machine Learning Unsupervised Machine Learning Deep Learning GPU Accelerated Machine Learning Agenda 3
- 4. 4 Agenda Introduction to MachineLearning Supervised Machine Learning Unsupervised Machine Learning Deep Learning
- 5.
- 6.
- 7. 7 Problems with hard coded Rules • Rules are manual, uses a human expert – difficult to maintain – give a one size fits all decision! (2 times overdose same as 38 times) • Machine learning uses data and statistics – can give sorted probabilty, can precisely match/target individuals
- 8. 8 What is Machine Learning? Train Algorithm BuildModel Features Predictions f(X)
- 9.
- 10. 10 Apache Spark Distributed Datasets Distributed Dataset Node Executor P4 Node Executor P1 P3 Node Executor P2 partitioned Partition 1 8213034705, 95, 2.927373, jake7870, 0…… Partition 2 8213034705, 115, 2.943484, Davidbresler2, 1…. Partition 3 8213034705, 100, 2.951285, gladimacowgirl, 58… Partition 4 8213034705, 117, 2.998947, daysrus, 95…. • Data read into Memory Cache • Partitioned across a cluster • Operated on in parallel • Cached in memory for iterations
- 11. 11 GPUs speed up Multi core servers for parallel processing Cluster of GPUs1 million times faster than Cray-1
- 12.
- 13.
- 14. 14 Agenda Introduction to MachineLearning Supervised Machine Learning Unsupervised Machine Learning Deep Learning
- 15. 15 Supervised and Unsupervised Machine Learning Machine Learning Unsupervised • Clustering • Collaborative Filtering • Frequent Pattern Mining Supervised • Classification • Regression Label
- 16.
- 17.
- 18.
- 19. 19 Form of ML that: • Identifies which category an item belongs to • Uses supervised learning algorithms – Data is labeled Classification: Definition Sentiment
- 20. 20 Credit Card Fraud Example • What are we trying to predict? – This is the Label or Target outcome: – Fraud or Not Fraud • What are the “if questions” or properties we can use to predict? – These are the Features: – Is the amount spent today > historical daily average? – Are there transactions close in time at locations far apart ? – Are the number of transactions today > historical average? – Are there new state or foreign purchases? Credit Card Transaction Features Number of Transactions last 24 hours Total $ Amount last 24 hours Average Amount last 24 hours Average Amount last 24 hours compared to historical use Location and Time difference since Last Transaction Average transaction fraud risk of merchant type Merchant types for day compared to historical use Features derived From Transaction History
- 21. 21 Decision Tree For Classification • Tree ofdecisions about features • Estimates IF THEN ELSE questions • Gives probability of a correct decision Is the amount spent in 24 hours > average Is the number of states used from > 2 Are there multiple Purchases today from risky merchants? YES NO NoYES Fraud 90% Not Fraud 50% Fraud 90% Not Fraud 30% YES No
- 22. 22 • Random Forest model consists of multiple decision trees from subsets of data • Final Prediction is combined average output of all trees • Improves accuracy • Gradient Boosted Decision Trees • Iteratively uses residual error to improve multiple trees. Final prediciton weighted sum. • XGBoost distributed faster version of GBDT Random Forest, Gradient Boosting Decision Trees, XGBoost
- 23. 23 Examples • Retail Example: – Predict price, sales • Telecom: – Predict customer will churn • Healthcare Example: – Probability of readmission • Marketing – Predict probability customer will click on add
- 24. 24 Supervised Learning House Price Prediction Example • What are we trying to predict? – This is the Label or Target outcome: – The house price $ • What are the “if questions” or properties we can use to predict? – These are the Features: – The size of the house (square meters)
- 25. 25 Label: House Price Y X Feature: housesize (square meters) Data point: price, size House Price = intercept + coefficient * house size y = a + bx House Price Regression Example
- 26.
- 27. 27 Regression Predicts a Numeric Value • Regression predicts a numeric value (eg price) • Retail Example: – Sales based on an event • Healthcare Example: – Days of hospital stay
- 28. 28 Agenda Introduction to MachineLearning Supervised Machine Learning Unsupervised Machine Learning Deep Learning
- 29. 29 Supervised and Unsupervised Machine Learning Machine Learning Unsupervised • Clustering • Collaborative Filtering • Frequent Pattern Mining Supervised • Classification • Regression Label
- 30. 30 Unsupervised Algorithms use Unlabeled data Customer GroupsBuild ModelTrainAlgorithm Finds patterns New Customer Purchase Data Use Model Similar Customer Group Contains patterns Recognizes patterns Customer purchase data
- 31. 31 Unsupervised Machine Learning: Clustering Clustering group news articlesinto different categories
- 32.
- 33. 33 Uses of Clustering Clustering Examples include: • Grouping: – Similar search results , text – Similar customers – Patient similarity – Similar Products • Anomaly detection: finding outliers, what’s outside of the groups not similar
- 34.
- 35.
- 36. 36 Clustering: Example Group similar objects Use MLlib K-means algorithm 1. Initialize coordinatesto K clusters centers (centroid) 2. Assign all points to nearest cluster center (centroid)
- 37. 37 Clustering: Example Group similar objects Use MLlib K-means algorithm 1. Initialize coordinatesto center of clusters (centroid) 2. Assign all points to nearest centroid 3. Update centroids to center of assigned points
- 38. 38 Clustering: Example Group similar objects Use MLlib K-means algorithm 1. Initialize coordinatesto center of clusters (centroid) 2. Assign all points to nearest centroid 3. Update centroids to center of points 4. Repeat until conditions met
- 39. 39 Association, Co-Occurrence, Market Basket Recommendations • Retail – Products which are purchased together • Take action: – Store layouts – Which products to put on specials, promote, coupons… • Healthcare – Patients like mine cohorts
- 40. 40 Agenda Introduction to MachineLearning Supervised Machine Learning Unsupervised Machine Learning Deep Learning
- 41.
- 42.
- 43.
- 44.
- 45.
- 46.
- 47.
- 48.
- 49.
- 50. 50 Deep Learning Features • Advantage: – Features do not have to be predetermined • Disadvantage: – Decisions are a black box Feature Decisions ?
- 51.
- 52.
- 53.
- 54. 54 Agenda Introduction to MachineLearning Supervised Machine Learning Unsupervised Machine Learning Deep Learning GPUs
- 55. 55 GPUs have beenresponsible for the advancement of deep learning in the past several years https://developer.nvidia.com/deep-learning-software
- 56. 56 cuDF cuIO Analytics Data PreparationVisualizationModel Training cuML Machine Learning cuGraph Graph Analytics PyTorch, TensorFlow, MxNet Deep Learning cuxfilter, pyViz, plotly Visualization Dask GPU Memory RAPIDS End-to-End Accelerated GPU Data Science https://developer.nvidia.com/blog/building-an-accelerated-data-science-ecosystem-rapids-hits-two-years/
- 57. 57 � All systems utilizethe same memory format � No overhead for cross-system communication � Projects can share functionality (eg, Parquet-to- Arrow reader) Source: From Apache Arrow Home Page - https://arrow.apache.org/ GPU DataFrame and Apache Arrow
- 58. 58 Data processing challenges with CPU-powered spark Hadoop brought scale-out processing to data analytics ETL and traditionalmachine-learning workloads continued to be written with tools like Scikit-Learn or multi-CPU distributed solutions like Spark. Limited Horizontal Scalability 2030202020102000 Data Processing Requirements CPU Gap in Data Processing Data to Analyze CPU Processing Hadoop Era Spark Era
- 59.
- 60. 60 BENEFITS OF GPUACCELERATED SPARK 3.0 Accelerate data science pipelines without code changes One pipeline, from ingest to data prep to training Data preparation and model training are both GPU-accelerated Infrastructure is consolidated and simplified
- 61. 61 NVIDIA innovations in Spark 3.0 • RAPIDS Accelerator for Spark 3.0 GPU Acceleration of: • Spark Data Frames • Spark SQL • Shuffle Accelerating Spark pipelines end-to-end
- 62.
- 63. 63 • Developers https://developer.nvidia.com/ • RAPIDS https://developer.nvidia.com/rapids • Deep Learning https://developer.nvidia.com/deep-learning • https://news.developer.nvidia.com/oak-ridge-national-laboratory-coronavirus- research/ • https://developer.nvidia.com/blog/jumpstarting-ai-with-covid-19-ct-inference- pipeline-and-clara-deploy-quickstart-vm/ • https://news.developer.nvidia.com/ • https://news.developer.nvidia.com/new-resource-for-nvidia-developers-access- technical-content-through-nvidia-on-demand/ • https://blazingsql.com/ TO LEARN MORE
- 64. 64 Integrations, feedback, documentationsupport, pull requests, new issues, or code donations welcomed! APACHE ARROW GPU OPEN ANALYTICS INITIATIVE https://arrow.apache.org/ @ApacheArrow http://gpuopenanalytics.com/ @GPUOAI RAPIDS https://rapids.ai @RAPIDSAI DASK https://dask.org @Dask_dev Join the Movement Everyone Can Help!
- 65. 65 GITHUB DOCKER https://github.com/rapidsai ANACONDA NGC https://anaconda.org/rapidsai/https://ngc.nvidia.com/registry/ nvidia-rapidsai-rapidsai https://hub.docker.com/r/ rapidsai/rapidsai/ RAPIDS How Do I Get the Software?