Teaching the cloud to think

Teaching the Cloud to ThinkIntro to Machine Learning with Azure

Josh Gillespie

In t roduct ions

Ground ru les

Formal Definition

Machine learning is a scientific discipline that deals with the construction and study of algorithms that can learn from data. Such algorithms operate by building a model based on inputs and using that to make predictions or decisions, rather than following only explicitly programmed instructions.

In formal (F lashy)Machine learning is the science of getting computers to act without being explicitly programmed

Informal (Mundane)

Machine Learning is turning data sets into software.

Software is called a “model” (or network, or graph, etc.).

Model “describes” the data set.

Use the model to generalize and make predictions about new data.

Example

y = mx + b

So isn ’ t th is just s tat i s t ics?

Machine learning is actually a software method. It's a way to generate software. So, it uses statistics but it's fundamentally... it's almost like a compiler. You use data to produce programs.

- John Platt, Distinguished Scientist at Microsoft Research

Example

y = mx + b

Summar izedMachine Learning is a computer program where the task performance measurably improves with experience.

WarningBuilding Machine Learning systems is slow, time-consuming, and error prone work.

Example applications

• The Post Office

• Self-driving cars

• Search Engines

• Skype/Cortana, Siri, Google Now.

Let ’s bu i ld someth ingRestaurant Recommendations

Thought experiment

Model Development

• Acquire data

• Prep Data

• Manipulation

• Training

• Scoring

• Evaluation

• Tuning

• Offline

• Re-implemented in another language

• Data Plumbing

• Verification

• Monitors, metrics, logging

• A/B testing

• Scaling/High availability

Azure Machine Learn ingWhat it is, what it is not.

What it is What it is not

• Fully managed service

• Browser based “ML Studio”

• Workflow-based experiments

• Drag/drop/connect

• Large library of common tasks

• Many algorithms built in

• Can run R scripts

• Parallel execution

• A silver bullet

• Magical

• A cloud-based PhD in data science

• Fast

• Generally available

WarningBuilding Machine Learning systems is slow, time-consuming, and error prone work.

Azure Stud io MLTour & Demo

Discuss ion

Thank [email protected]

@jcgillespie

http://awaitwisdom.com