MMDS 2014 Talk - Distributing ML Algorithms: from GPUs to the Cloud
13 likes3,829 views
This document discusses distributing large-scale machine learning algorithms from GPUs to the cloud. It outlines Netflix's use of machine learning algorithms to power recommendations, describes how their data and needs have grown enormously over time, and discusses different levels of distributing and parallelizing machine learning algorithms in practice based on whether all the data is needed, what level of parallelization is appropriate, and what latency can be afforded.
![[Banko and Brill, 2001]
Norvig: “Google does not
have better Algorithms,
only more Data”
Many features/
low-bias models](https://image.slidesharecdn.com/mmds-xavieramatriain-140619140040-phpapp02/85/MMDS-2014-Talk-Distributing-ML-Algorithms-from-GPUs-to-the-Cloud-22-320.jpg)
More Related Content
![[PREMONEY 2014] Mayfield Fund >> Tim Chang, "Mobile Is The Future Of YOU: Why...](https://cdn.slidesharecdn.com/ss_thumbnails/09timchangmayfieldfund-140627080652-phpapp01-thumbnail.jpg?width=560&fit=bounds)
Similar to MMDS 2014 Talk - Distributing ML Algorithms: from GPUs to the Cloud (20)
MMDS 2014 Talk - Distributing ML Algorithms: from GPUs to the Cloud
- 1. Distributing Large-scale ML Algorithms: from GPUs to the Cloud MMDS 2014 June, 2014 Xavier Amatriain Director - Algorithms Engineering @xamat
- 2. Outline ■ Introduction ■ Emmy-winning Algorithms ■ Distributing ML Algorithms in Practice ■ An example: ANN over GPUs & AWS Cloud
- 3. What we were interested in: ■ High quality recommendations Proxy question: ■ Accuracy in predicted rating ■ Improve by 10% = $1million! Data size: ■ 100M ratings (back then “almost massive”)
- 4. 2006 2014
- 5. Netflix Scale ▪ > 44M members ▪ > 40 countries ▪ > 1000 device types ▪ > 5B hours in Q3 2013 ▪ Plays: > 50M/day ▪ Searches: > 3M/day ▪ Ratings: > 5M/day ▪ Log 100B events/day ▪ 31.62% of peak US downstream traffic
- 6. Smart Models ■ Regression models (Logistic, Linear, Elastic nets) ■ GBDT/RF ■ SVD & other MF models ■ Factorization Machines ■ Restricted Boltzmann Machines ■ Markov Chains & other graphical models ■ Clustering (from k-means to modern non-parametric models) ■ Deep ANN ■ LDA ■ Association Rules ■ …
- 10. 2007 Progress Prize ▪ Top 2 algorithms ▪ MF/SVD - Prize RMSE: 0.8914 ▪ RBM - Prize RMSE: 0.8990 ▪ Linear blend Prize RMSE: 0.88 ▪ Currently in use as part of Netflix’ rating prediction component ▪ Limitations ▪ Designed for 100M ratings, we have 5B ratings ▪ Not adaptable as users add ratings ▪ Performance issues
- 11. Ranking Ranking
- 12. Page composition
- 13. Similarity
- 15. Postplay
- 16. Gamification
- 17. Distributing ML algorithms in practice
- 18. 1. Do I need all that data? 2. At what level should I distribute/parallelize? 3. What latency can I afford?
- 19. Do I need all that data?
- 20. Really? Anand Rajaraman: Former Stanford Prof. & Senior VP at Walmart
- 21. Sometimes, it’s not about more data
- 22. [Banko and Brill, 2001] Norvig: “Google does not have better Algorithms, only more Data” Many features/ low-bias models
- 23. Sometimes, it’s not about more data
- 24. At what level should I parallelize?
- 25. The three levels of Distribution/Parallelization 1. For each subset of the population (e.g. region) 2. For each combination of the hyperparameters 3. For each subset of the training data Each level has different requirements
- 26. Level 1 Distribution ■ We may have subsets of the population for which we need to train an independently optimized model. ■ Training can be fully distributed requiring no coordination or data communication
- 27. Level 2 Distribution ■ For a given subset of the population we need to find the “optimal” model ■ Train several models with different hyperparameter values ■ Worst-case: grid search ■ Can do much better than this (E.g. Bayesian Optimization with Gaussian Process Priors) ■ This process *does* require coordination ■ Need to decide on next “step” ■ Need to gather final optimal result ■ Requires data distribution, not sharing
- 28. Level 3 Distribution ■ For each combination of hyperparameters, model training may still be expensive ■ Process requires coordination and data sharing/communication ■ Can distribute computation over machines splitting examples or parameters (e.g. ADMM) ■ Or parallelize on a single multicore machine (e.g. Hogwild) ■ Or… use GPUs
- 29. ANN Training over GPUS and AWS
- 30. ANN Training over GPUS and AWS ■ Level 1 distribution: machines over different AWS regions ■ Level 2 distribution: machines in AWS and same AWS region ■ Use coordination tools ■ Spearmint or similar for parameter optimization ■ Condor, StarCluster, Mesos… for distributed cluster coordination ■ Level 3 parallelization: highly optimized parallel CUDA code on GPUs
- 31. What latency can I afford?
- 32. 3 shades of latency ▪ Blueprint for multiple algorithm services ▪ Ranking ▪ Row selection ▪ Ratings ▪ Search ▪ … ▪ Multi-layered Machine Learning
- 34. Xavier Amatriain (@xamat) [email protected] Thanks! (and yes, we are hiring)