Post-hype Machine Learning
Piero Savastano - pieroit
Why trust me?
- I have a fair experience in both academy and industry and worked on data for at least 10 years
- I was dealing with AI and data science when very few people were interested in it
Drop the expectations
We are currently in a bubble similar to the dotcom one.
When the bubble will pop, we will finally leave exaggerated expectations and profit from AI/ML as tools to partecipate in a data economy.
The bare minimum
you need to know about ML
- Implicit programming starting from data
- Statistical optimization, cyclic development
- Again, we are in a data economy
Data is king
- Data > algorithm
- Garbage in, garbage out
- Challenges in cleaning, integrating and standardizing data from different sources (also within an organization)
Good is enough
- “I want 99% accuracy”
- R&D is expensive
- Tip: start simply. Then search for a research lab or a startup that is already working on the problem, otherwise be prepared for a long term effort
The mith of the Data Scientist
- At least 4 different jobs:
- System admin (building and managing clusters)
- Researcher (testing and composing algorithms)
- Analyst (vertical domain expert)
- Developer (integrating in existing architectures)
- PR (viz and communications)
- If you find one, get ready to pay him big money
- A more mature approach is to set up a team
A step by step process
to dominate and take value
out of data
1. Build a team with a data scientist* and a domain expert
2. Clarify in great detail what are the problems to solve
3. Establish business-sound metrics
4. Build, validate and cycle
Thank you
&
good luck
Post-hypeMachine Learning
By Piero Savastano
Post-hypeMachine Learning
- 501