TB_lite
( Slimmed TensorBoard )
TensorFlow & Deep Learning SG
19 December 2019
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
- Red Dragon AI...
About Red Dragon AI
- Google Partner : Deep Learning Consulting & Prototyping
- SGInnovate/Govt : Education / Training
- Research : NeurIPS / EMNLP
- Products :
-
- Conversational Computing
- Natural Voice Generation - multiple languages
- Knowledgebase interaction & reasoning
Outline
whoami
= DONE
- What is TensorBoard?
-
- Why did I build
tb_lite
?
-
- Wrap-up
TensorBoard FTW
- Visualisation Tool :
-
- Metrics
- Structure
- Weight distributions
- Performance
Marketing Animation
TensorBoard WTF
- Metrics is must-have
- Over optimised (?)
-
- Sometimes says "No"
- Particularly with PyTorch interop
- Need for consistency
- Need for better colours
Marketing Animation
Error: Category Mis-match...
Deep Learning
MeetUp Group
MetaInit
Quick Paper Shoutout
- Google Poster at NeurIPS : MetaInit
- Idea:
-
- Initialisation of networks is tricky
- Standard methods don't account for structure
- Create metric for 'good initialisation'
- Train initial weight norms for that
- No need for
BatchNorm
- Works on MNIST ... ImageNet
Deep Learning : Jump-Start Workshop
Deep Learning
Developer Course
- QUESTIONS -
Martin @
RedDragon . AI