You are on page 1of 19

Introduction to

ML & DL On cloud
Shiva Bansal
Bigdata ML Architect (HP Inc.)
PREREQUISITE

 Some background of Statistics and Probability


 Some Programming experience
 Some data wrangling and visualization skills
WHAT IS AI??
 Artificial Intelligence is research area to teach the machine to behave
like human.

AI

Others Like
Machine Machine
Robotic, fuzzy
Learning Reasoning
logic

Reinforcement Natural Language


Deep Learning
Learning P&U
WHAT IS ML??

Machine Learning
Process of teaching machine to learn & improve from experience without being explicitly
programmed

Model represents mathematical equation


derived by analyzing data patterns during
machine learning process
i.e
WHY ML ??
Data

Human
Business
• Volume • Proactive • Bias
• Variety • Automation • Knowledge
• Velocity • Experience
• Competition
BUSINESS USE CASES

Banking E com Medical Telecom


Money Laundering by Recommendation Visual Data Network optimization
identifying fraud based on purchase processing for by analyzing traffic
transaction
History Tumor detection behavior

Anomaly Detection
ML CATEGORIES
ML MODEL DEEP DIVE

food travel expense


10 20 70
20 30 110 Expense(y) = 2 * food + 2 * travel + 10
25 25 110
100 120 400

Outlier
ML MODEL DEEP DIVE (CONT.)

• Model is trying to predict accumulated


savings over weeks

• points represents actual observation

• Line depicts best possible equation


satisfying model

• Correlation coefficient (r) represents


model accuracy or error
WHAT IS DL??

Deep Learning
Deep Learning is subset of machine learning that tries to imitate the working of human brain by building
brain like deep neural network.
DL MODEL COMPONENTS

Model components

• Input Layer i.e. dense


• Hidden Layer i.e. dense
• Neurons
• Activation function (relu, softmax)
• Bias/weight
• Loss function
• Optimizer
• metrics
MACHINE LEARNING PIPELINE
Model
Deployment
building/retraining

YES

Design Data Meeting


Business Problem Data Collection & Model Training & Business
Hypothesis Preprocessing &
Formulation Integration validation Goals
Testing Visualization

No

Feature
Engineering

Refine Problem/
New data
integration
WHY CLOUD

No Fixed Cost Easy to setup

Rich ML Deployment &


Libraries Maintenance
ML CLOUD PROVIDERS

Cloud Agnostic
CLOUD SERVICES MAPPING

AWS Azure GCP


• EC2 • VM • Compute Engine
• S3 • Blob Storage • Cloud Storage
• Lambda • Azure functions • Cloud functions
• RDS • SQL Database • Cloud SQL
• Sagemaker • ML Studio • Cloud Datalab
• Cloudwatch • Azure monitor • Stackdriver
• Athena/Glue • SQL Datawarehouse • Bigquery
• Kinesis • Stream Analytics • Cloud dataflow
SAMPLE ARCHITECTURES
CLOUD NATIVE ML APPROACH
CLOUD AGNOSTIC ML APPROACH
Q&A

You might also like