Day 14 #31DaysofML

ЁЯдФ How to pick the right #GoogleCloud #MachineLearning tool for your application?

Answer these questions
тЭУ What's your teams ML expertise?
тЭУ How much control/abstraction do you need?
тЭУ Would you like to handle the infrastructure components?

ЁЯз╡ ЁЯСЗ

@SRobTweets created this pyramid to explain the idea.
As you move up the pyramid, less ML expertise is required, and you also donтАЩt need to worry as much about the infrastructure behind your model.

To lear more watch this video ЁЯСЙ https://t.co/EqJNDmTfhV

#31DaysofML 2/10
@SRobTweets If youтАЩre using Open source ML frameworks (#TensorFlow) to build the models, you get the flexibility of moving your workloads across different development & deployment environments. But, you need to manage all the infrastructure yourself for training & serving

#31DaysofML 3/10
@SRobTweets Deep Learning VMs provide managed, click-to-deploy VMs for processing data & training the model
ЁЯФ╣ Popular ML frameworks pre-installed
ЁЯФ╣ Reduces the overhead of managing & allocating compute & storage required
ЁЯФ╣ But you figure out how youтАЩll serve those models

#31DaysofML 4/10
@SRobTweets Kubeflow - OS project for deploying ML workloads on #Kubernetes
ЁЯФ╣ Helps configure a multi-step ML pipeline including pre-processing data, training & serving the model
ЁЯФ╣ Run it on-premise or on any cloud
ЁЯФ╣ YouтАЩll still need to configure where itтАЩs managed

#31DaysofML 5/10
@SRobTweets AI Platform - managed service for all custom model needs
ЁЯФ╣ Includes tools for training & serving models, hosted notebooks, a data labeling service & more
ЁЯФ╣ Eg: take notebook code running on-premise with Kubeflow, and run it on GCP with AI Platform Notebooks

#31DaysofML 6/10
@SRobTweets BQML: Brings the power of ML closer to where the data is analyzed & make it accessible to data analysts
ЁЯФ╣ You donтАЩt have to write any of the underlying model code
ЁЯФ╣ Choose model type
ЁЯФ╣ Simple SQL queries to create & train the model & make predictions

#31DaysofML 7/10
@SRobTweets AutoML democratizes ML to build custom ML models regardless of ML expertise.
ЁЯФ╣ Use the UI to upload the data - images, video, text, or structured
ЁЯФ╣ Press "train" button
ЁЯФ╣ Model is available for prediction via an API
ЁЯФ╣ No need to deploy it yourself

#31DaysofML 8/10
@SRobTweets ML APIs: Easiest and fastest way to get started with AI
ЁЯФ╣ DonтАЩt need ML engineers or data scientists just some developers
ЁЯФ╣ Simple API request to pre-trained models for images, video, speech, text & translation
ЁЯФ╣ No need to supply any training data yourself

#31DaysofML 9/10
@SRobTweets ML APIs тЖТ https://t.co/XdR6oS5XrcтАЛ
AutoML тЖТ https://t.co/vbmIBiciLFтАЛ
BQML тЖТ https://t.co/Hs8zz57pcnтАЛ
AI Platform тЖТ https://t.co/zyYRq4HzT5тАЛ
Kubeflow тЖТ https://t.co/DNX7MftUb3тАЛ
Deep Learning VMs тЖТ https://t.co/9MG9KntYXbтАЛ
Tensorflow тЖТ https://t.co/G2xLT68gRXтАЛ

10/10

You May Also Like

The YouTube algorithm that I helped build in 2011 still recommends the flat earth theory by the *hundreds of millions*. This investigation by @RawStory shows some of the real-life consequences of this badly designed AI.


This spring at SxSW, @SusanWojcicki promised "Wikipedia snippets" on debated videos. But they didn't put them on flat earth videos, and instead @YouTube is promoting merchandising such as "NASA lies - Never Trust a Snake". 2/


A few example of flat earth videos that were promoted by YouTube #today:
https://t.co/TumQiX2tlj 3/

https://t.co/uAORIJ5BYX 4/

https://t.co/yOGZ0pLfHG 5/
Recently, the @CNIL issued a decision regarding the GDPR compliance of an unknown French adtech company named "Vectaury". It may seem like small fry, but the decision has potential wide-ranging impacts for Google, the IAB framework, and today's adtech. It's thread time! ЁЯСЗ

It's all in French, but if you're up for it you can read:
тАв Their blog post (lacks the most interesting details):
https://t.co/PHkDcOT1hy
тАв Their high-level legal decision: https://t.co/hwpiEvjodt
тАв The full notification: https://t.co/QQB7rfynha

I've read it so you needn't!

Vectaury was collecting geolocation data in order to create profiles (eg. people who often go to this or that type of shop) so as to power ad targeting. They operate through embedded SDKs and ad bidding, making them invisible to users.

The @CNIL notes that profiling based off of geolocation presents particular risks since it reveals people's movements and habits. As risky, the processing requires consent тАФ this will be the heart of their assessment.

Interesting point: they justify the decision in part because of how many people COULD be targeted in this way (rather than how many have тАФ though they note that too). Because it's on a phone, and many have phones, it is considered large-scale processing no matter what.
рдкреНрд░рд╛рдЪреАрди рдХрд╛рд▓ рдореЗрдВ рдЧрд╛рдзрд┐ рдирд╛рдордХ рдПрдХ рд░рд╛рдЬрд╛ рдереЗредрдЙрдирдХреА рд╕рддреНрдпрд╡рддреА рдирд╛рдо рдХреА рдПрдХ рдкреБрддреНрд░реА рдереАредрд░рд╛рдЬрд╛ рдЧрд╛рдзрд┐ рдиреЗ рдЕрдкрдиреА рдкреБрддреНрд░реА рдХрд╛ рд╡рд┐рд╡рд╛рд╣ рдорд╣рд░реНрд╖рд┐ рднреГрдЧреБ рдХреЗ рдкреБрддреНрд░ рд╕реЗ рдХрд░рд╡рд╛ рджрд┐рдпрд╛редрдорд╣рд░реНрд╖рд┐ рднреГрдЧреБ рдЗрд╕ рд╡рд┐рд╡рд╛рд╣ рд╕реЗ рдмрд╣реБрдд рдкреНрд░рд╕рдиреНрди рд╣реБрдП рдФрд░ рдЙрдиреНрд╣реЛрдиреЗ рдЕрдкрдиреА рдкреБрддреНрд░рд╡рдзреБ рдХреЛ рдЖрд╢реАрд░реНрд╡рд╛рдж рджреЗрдХрд░ рдЙрд╕реЗ рдХреЛрдИ рднреА рд╡рд░ рдорд╛рдВрдЧрдиреЗ рдХреЛ рдХрд╣рд╛ред


рд╕рддреНрдпрд╡рддреА рдиреЗ рдорд╣рд░реНрд╖рд┐ рднреГрдЧреБ рд╕реЗ рдЕрдкрдиреЗ рддрдерд╛ рдЕрдкрдиреА рдорд╛рддрд╛ рдХреЗ рд▓рд┐рдП рдкреБрддреНрд░ рдХрд╛ рд╡рд░рджрд╛рди рдорд╛рдВрдЧрд╛редрдпреЗ рдЬрд╛рдирдХрд░ рдорд╣рд░реНрд╖рд┐ рднреГрдЧреБ рдиреЗ рдпрдЬреНрдЮ рдХрд┐рдпрд╛ рдФрд░ рддрддреНрдкрд╢реНрдЪрд╛рдд рд╕рддреНрдпрд╡рддреА рдФрд░ рдЙрд╕рдХреА рдорд╛рддрд╛ рдХреЛ рдЕрд▓рдЧ-рдЕрд▓рдЧ рдкреНрд░рдХрд╛рд░ рдХреЗ рджреЛ рдЪрд░реВ (рдпрдЬреНрдЮ рдХреЗ рд▓рд┐рдП рдкрдХрд╛рдпрд╛ рд╣реБрдЖ рдЕрдиреНрди) рджрд┐рдП рдФрд░ рдХрд╣рд╛ рдХрд┐ рдЛрддреБ рд╕реНрдирд╛рди рдХреЗ рдмрд╛рдж рддреБрдореНрд╣рд╛рд░реА рдорд╛рддрд╛ рдкреБрддреНрд░ рдХреА рдЗрдЪреНрдЫрд╛ рд▓реЗрдХрд░ рдкреАрдкрд▓ рдХрд╛ рдЖрд▓рд┐рдВрдЧрди...

...рдХрд░реЗрдВ рдФрд░ рддреБрдо рднреА рдкреБрддреНрд░ рдХреА рдЗрдЪреНрдЫрд╛ рд▓реЗрдХрд░ рдЧреВрд▓рд░ рд╡реГрдХреНрд╖ рдХрд╛ рдЖрд▓рд┐рдВрдЧрди рдХрд░рдирд╛ред рдЖрд▓рд┐рдВрдЧрди рдХрд░рдиреЗ рдХреЗ рдмрд╛рдж рдЪрд░реВ рдХрд╛ рд╕реЗрд╡рди рдХрд░рдирд╛, рдЗрд╕рд╕реЗ рддреБрдо рджреЛрдиреЛ рдХреЛ рдкреБрддреНрд░ рдкреНрд░рд╛рдкреНрддрд┐ рд╣реЛрдЧреАредрдкрд░рдВрддреБ рдорд╛рдВ рдмреЗрдЯреА рдХреЗ рдЪрд░реВ рдЖрдкрд╕ рдореЗрдВ рдмрджрд▓ рдЬрд╛рддреЗ рд╣реИрдВ рдФрд░ рдпреЗ рдорд╣рд░реНрд╖рд┐ рднреГрдЧреБ рдЕрдкрдиреА рджрд┐рд╡реНрдп рджреГрд╖реНрдЯрд┐ рд╕реЗ рджреЗрдЦ рд▓реЗрддреЗ рд╣реИрдВред

рднреГрдЧреБ рдЛрд╖рд┐ рд╕рддреНрдпрд╡рддреА рд╕реЗ рдХрд╣рддреЗ рд╣реИрдВ,"рдкреБрддреНрд░реА рддреБрдореНрд╣рд╛рд░рд╛ рдФрд░ рддреБрдореНрд╣рд╛рд░реА рдорд╛рддрд╛ рдиреЗ рдПрдХ рджреБрд╕рд░реЗ рдХреЗ рдЪрд░реВ рдЦрд╛ рд▓рд┐рдП рд╣реИрдВредрдЗрд╕ рдХрд╛рд░рдг рддреБрдореНрд╣рд╛рд░рд╛ рдкреБрддреНрд░ рдмреНрд░рд╛рд╣реНрдордг рд╣реЛрддреЗ рд╣реБрдП рднреА рдХреНрд╖рддреНрд░рд┐рдп рд╕рд╛ рдЖрдЪрд░рдг рдХрд░реЗрдЧрд╛ рдФрд░ рддреБрдореНрд╣рд╛рд░реА рдорд╛рддрд╛ рдХрд╛ рдкреБрддреНрд░ рдХреНрд╖рддреНрд░рд┐рдп рд╣реЛрдХрд░ рднреА рдмреНрд░рд╛рд╣реНрдордг рд╕рд╛ рдЖрдЪрд░рдг рдХрд░реЗрдЧрд╛ред"
рдЗрд╕ рдкрд░ рд╕рддреНрдпрд╡рддреА рдиреЗ рднреГрдЧреБ рдЛрд╖рд┐ рд╕реЗ рдмрдбрд╝реА рд╡рд┐рдирддреА рдХреАред


рд╕рддреНрдпрд╡рддреА рдиреЗ рдХрд╣рд╛,"рдореБрдЭреЗ рдЖрд╢реАрд░реНрд╡рд╛рдж рджреЗрдВ рдХрд┐ рдореЗрд░рд╛ рдкреБрддреНрд░ рдмреНрд░рд╛рд╣реНрдордг рд╕рд╛ рд╣реА рдЖрдЪрд░рдг рдХрд░реЗред"рддрдм рдорд╣рд░реНрд╖рд┐ рдиреЗ рдЙрд╕реЗ рдпреЗ рдЖрд╢реАрд░реНрд╡рд╛рдж рджреЗ рджрд┐рдпрд╛ рдХрд┐ рдЙрд╕рдХрд╛ рдкреБрддреНрд░ рдмреНрд░рд╛рд╣реНрдордг рд╕рд╛ рд╣реА рдЖрдЪрд░рдг рдХрд░реЗрдЧрд╛ рдХрд┐рдиреНрддреБ рдЙрд╕рдХрд╛ рдкреМрддреНрд░ рдХреНрд╖рддреНрд░рд┐рдпреЛрдВ рд╕рд╛ рд╡реНрдпрд╡рд╣рд╛рд░ рдХрд░реЗрдЧрд╛ред рд╕рддреНрдпрд╡рддреА рдХрд╛ рдПрдХ рдкреБрддреНрд░ рд╣реБрдЖ рдЬрд┐рд╕рдХрд╛ рдирд╛рдо рдЬрдореНрджрд╛рдЧреНрдирд┐ рдерд╛ рдЬреЛ рд╕рдкреНрдд рдЛрд╖рд┐рдпреЛрдВ рдореЗрдВ рд╕реЗ рдПрдХ рд╣реИрдВред