Leveraging AI on the Cloud to transform your business Florida Business Analytics Forum 2018 at University of South Florida 1
My (unusual) Neural networks at NOAA path to Google 2
DNNs solved image analysis My ��� �he��� c�� �ow �� le����d �� a n���a� n����r� ... 3
After 4 4 years years managing infra ... 4
I discovered the power of cloud. Eve�� t�� ��ek�! 5
Mac���� Le�r���g Where do you go if you want to be part of two revolutions? Clo�� ���pu���g 6
7
Leveraging AI on the Cloud to transform your business 8
When you hear machine learning , you probably think of... 9
The most % of Type of # of network # of deployed network layers weights models common ML MLP0 5 20M 61% models at MLP1 4 5M Google LSTM0 58 52M operate on 29% LSTM1 56 34M structured data CNN0 16 8M 5% CNN1 89 100M 10
Machine Learning is a way to use standard algorithms to derive predictive insights from data and make repeated decisions algorithm Predictive insight decision data 11
Baseball California Football New York 12
Machine learning scales better than hand-coded rules query = ‘Giants’ user location = user location = user location = ‘Bay Area’ ? ‘New York’ ? ‘other’ ? results about results about results about SF Giants NY Giants giants 13
Search RankBrain machine learning for search engines (a deep neural #1 network for search ranking) improved improvement to ranking quality performance in 2+ years significantly 14
#1 ML can be used to solve many problems for which you are writing rules today 15
Is this machine learning? What’s needed for ML? 16
Is this machine learning? What’s needed for ML? 17
Is this machine learning? What’s needed for ML? 18
#2 Machine Learning is how you personalize applications and reach the long tail 19
“It's not who has the best algorithm who wins, it's who has the most data” Andrew Ng 20
Stage 1: Leads (1000s) Conventional methods are about filtering down the Stage 2: data you happen Products (100s) to have Stage 3: Customers (10s) 21
Machine Learning is about accounting for more diverse factors 22
Can now capture data from many sources 23
Games and social media analytics Advertising campaign optimization Big data Sensor data analysis is changing Transportation and logistics many industries POS-Retail Analytics Web Logs, Machine Logs, Infrastructure monitoring Mobile application analytics 24
8.4 Billion The number of connected things in use in 2017, up 31% from 2016* We’re generating more data than ever before 25
#3 Design systems with the expectation that you will have more data next year 26
What happens when you collect petabytes and exabytes of data? 27
Typical Big Data Processing Time to Understanding Businesses can not derive value Monitoring Programming from data if they Resource Performance are focused provisioning tuning on building infrastructure Handling Utilization growing scale improvements Deployment & Reliability configuration 28
Towards serverless data analysis and processing Auto ML NoSQL Dataflow Spanner Pub/sub Dataflow Cloud Cloud ML Spanner GCS Dataproc BigTable BigQuery GCS Datastore Dataflow F1 Engine 2002 2004 2006 2008 2010 2012 2013 2016 2018 29
Big Data Processing with Google Cloud Platform Time to Understanding Spend Time on ‘What’ not ‘How’ Programming Focus on insight, not infrastructure 30
#4a Use a platform that lets you forget about infrastructure 31
The ML marketplace is moving towards increasing levels of ML abstraction Custom image model Build off NLP API to Use Vision API as-is to Use Dialogflow to to price cars route customer emails find text in memes create a new shopping experience 32
#4b ML is software -- learn to make buy-vs-build decisions 33
#1 ML can be used to solve many problems for which you are writing rules today #2 Google Cloud* can help ML is how you personalize applications and reach the long tail you transform your #3 business with AI Design systems with the expectation that you will have more data next year #4 *In Tampa, Tom Howe: thowe@google.com Use a platform that lets you forget about infrastructure and offers great pre-built models @lak_gcp 34
Thank you. cloud.google.com 35
Recommend
More recommend