What is Machine Learning and how does it work? But even more importantly, what problems can ML solve for you and your company?
Once you have understood the potential use cases, we will briefly describe the main challenges in the world of Big Data.
Why is deploying ML models so hard and how can Cloud Computing help?
Many MLaaS options are available on the market (AWS, Google, Azure, BigML, etc.). We will see how they compare to each other and which may best fit your needs.
Whenever MLaaS is not enough, you can build your own ML models. We will briefly explain why Serverless is a great deployment strategy for this use case and what problems and limitation arise with it.
Furthermore, we will put these ideas into practice and build a model for Sentiment Analysis, based on Python (scikit-learn), and trained with a public dataset by Stanford University.
3. What is Machine Learning?
Back to 1959 (Arthur Samuel)
How computers learn from Data
clda.co/serverless-‐workshop Workshop | 東京
How to solve decision problems
4. Machine Learning pipeline
Training Predic6on
batch real-‐Ame
Feature
extrac6on
batch
data informaPon
features ML models
clda.co/serverless-‐workshop Workshop | 東京
9. What problems can ML solve for you?
Supervised
Learning
Unsupervised
Learning
classifica'on
regression
clustering
rule extrac'on
?
170
cm
gro gro
A, B C
clda.co/serverless-‐workshop Workshop | 東京
10. What problems can ML solve for you?
Supervised
Learning
Unsupervised
Learning
classifica'on
regression
clustering
rule extrac'on
?
fraud detecPon
170
cm
gro gro
A, B C
price of a stock over Pme
purchase likelihood
user segmentaPon
clda.co/serverless-‐workshop Workshop | 東京
12. Generated data started growing ~10 years ago…
“90% of the data in the world today has been
created in the last two years alone” -‐ IBM
“300+ hours worth of video content is being
uploaded to the site every minute” -‐ Youtube
clda.co/serverless-‐workshop Workshop | 東京
13. … and it keeps geKng bigger!
clda.co/serverless-‐workshop Workshop | 東京
14. Big data challenges
Manual exploraPon is not an opPon
Data-‐driven decisions are a must
Distributed/parallel compuPng
The curse of dimensionality
clda.co/serverless-‐workshop Workshop | 東京
17. Why is deploying ML models a challenge?
clda.co/serverless-‐workshop Workshop | 東京
18. Why is deploying ML models a challenge?
1. Prototyping != ProducPon-‐ready
2. We need ElasPcity
4. MulP-‐model architectures
3. Too many nice-‐to-‐have features
clda.co/serverless-‐workshop Workshop | 東京
5. Avoid lack of ownership
19. Machine Learning as a Service (MLaaS)
Amazon
Machine Learning
Azure
Machine Learning
Google
PredicAon API
IMB
Watson AnalyAcs
BigML
Workshop | 東京clda.co/serverless-‐workshop
20. Amazon Machine Learning
AmazonML
One of the first MLaaS soluPons (1 year old)
Great service for classificaPon and regression
Only linear models (linear & logisPc regression + SGD)
No support for advanced scenarios yet
Workshop | 東京clda.co/serverless-‐workshop
21. AmazonML @ Cloud Academy
Workshop | 東京clda.co/serverless-‐workshop
clda.co/7-‐day-‐free
(no credit card required!)
22. Serverless compuAng to the rescue!
Transparent scalability, elasPcity and availability
Developer-‐friendly maintenance (versioning + aliases)
AWS
Lambda
Event-‐driven approach & never pay for idle
1 funcPon = 1 model
A/B tesPng via composiPon
clda.co/serverless-‐workshop Workshop | 東京
![[特別企画]
Serverless Machine Learning
Workshop
clda.co/serverless-‐workshop
10/04/2016 東京](https://image.slidesharecdn.com/serverless-machine-learning-workshop-tokyo-161004023553/85/Serverless-Machine-Learning-Workshop-1-320.jpg)
