TensorFlow Everywhere
for Everyone
Google Cloud Next Extended Singapore
25 August 2018
About Me
- Machine Intelligence / Startups / Finance
-
- Moved from NYC to Singapore in Sep-2013
- 2014 = 'fun' :
-
- Machine Learning, Deep Learning, NLP
- Robots, drones
- Since 2015 = 'serious' :: NLP + deep learning
-
- & GDE ML; TF&DL co-organiser
- & Papers...
- & Dev Course...
About Red Dragon AI
- Google Partner : Deep Learning Consulting & Prototyping
- SGInnovate/Govt : Education / Training
- Products :
-
- Conversational Computing
- Natural Voice Generation - multiple languages
- Knowledgebase interaction & reasoning
Outline
whoami= DONE- TensorFlow ecosystem picture
- TensorFlow is evolving
- Demo : TensorFlow/Keras
- AutoML description & demo
- Wrap-up
TF Ecosystem
- 2017-06 version...
TF options
- TensorFlow (original flavour)
- TensorFlow Lite
- TensorFlow.js
- Eager Mode
Ingesting Data
tf.dataAPIs- Run data ingestion in the graph
- Benefit : async with GPU
- Bigger Benefit : TPUs
TPU 'High Level'
- v. fast
- TPUs live in 'pods'
TPU Detail
- TPUs live in 'pods'
- Execution graph must be transmitted across network
-
- Needs ingestion in the graph
- Storage buckets faster than SSD...
Making it easier
- Cloud Vision API
- AutoML
Raw ML in the Cloud (FREE)
- Google Colab
-
- Collaborative Jupyter notebook
- Switch on $Free GPU
- .. Preemptible + 12hr limit
- Can connect to Drive and GCP Storage too
Raw ML in the Cloud ($)
- Google Deep Learning VM images
-
- Latest Nvidia Drivers
- TensorFlow specially compiled
- PyTorch support on equal footing
Code
- Using Keras / TensorFlow and Transfer Learning
https://github.com/
mdda/deep-learning-workshop/
notebooks/2-CNN/5-TransferLearning/5-ImageClassifier-keras.ipynb
Load Directly into Colab
AutoML
- For people that :
-
- Have data
- Have money, but no specialists
- Want "AI" results
- AutoML builds :
-
- A usable model, tuned to your data
- Fully Automatically : Just add Money
AutoML Picture
- For people with data, requirements and money
AutoML Data
- Drag-and-drop interface
AutoML Train
- Pick budget + 'Start Training' interface
AutoML Evaluate
- Nice interface
AutoML Serve
- Easy to interface. Probably scalable.
AutoML Demo
- (if there's time)
AutoML Advantages
- Easy-to-Use (in theory)
- Need to know almost nothing about ML
- Everything is done for you :
-
- Training, evaluation, metrics
- API is immediately available
AutoML Disadvantages
- Classification Only
- No control over what is happening :
-
- Can't choose train/test splits
- Not easy to understand
- Might be costly ~ $20/hr (beta)
AutoML Conclusions
- Fantastic if you need a model but don't know TensorFlow
- Even if you do know TF : Great for baselines
- Probably won’t win a Kaggle competition
- Could be expensive if you are retraining a lot
Wrap-up
- TensorFlow ecosystem is EXPLODING
- Initial promise starting to be delivered
- AutoML allows much broad adoption of 'AI'
Deep Learning
MeetUp Group
- MeetUp.com / TensorFlow-and-Deep-Learning-Singapore
- Next Meeting :
-
- 4-Sept, hosted at Google :: "Explainability"
- Typical Contents :
-
- Talk for people starting out
- Something from the bleeding-edge
- Lightning Talks
Deep Learning
Developer Course
- JumpStart module is Module #1 of 5
- Plan : Advanced modules in October/November
- Each 'module' will include :
-
- Instruction
- Individual Projects
- Support by SG govt
- Location : SGInnovate
- Status : TBA
Deep Learning : Jump-Start Workshop
- First part of Full Developer Course - last one oversubscribed
- Dates + Cost : Sept 12-13, S$600 (much less for SC/PR)
-
- 2 week-days
- Play with real models & Get inspired!
- Pick-a-Project to do
- Regroup on subsequent week-night(s)
- SGInnovate - Talent - Talent Development -
Deep Learning Developer Series
RedDragon AI
Intern Hunt
- Opportunity to do Deep Learning all day
- Work on something cutting-edge
- Location : Singapore
- Status : SG/PR FTW
- Need to coordinate timing...
- QUESTIONS -
Martin.Andrews @
RedCatLabs.com
Martin.Andrews @
RedDragon.AI
My blog : http://blog.mdda.net/
GitHub : mdda