TesnsorFlowJs.pdf

Transcript

1. Hiren Dave Machine Learning in Browser With TensorFlow.js
2. What is a Machine Learning?
3. Data Training Prediction End Machine Learning in Simple Terms
4. Why ML in Browser?
5. Proprietary + Confidential No Drivers / No Installs Just Downloads Source: Lorem ipsum dolor sit amet, consectetur adipiscing elit. Duis non erat sem $ python –v $ pip install pandas $ pip install django $ pip install numpy $ pip3 install flask $ pip install tensorflow $ pip install sklearn $ python –v $ pip install pandas $ pip install django $ pip install numpy $ pip3 install flask $ pip install tensorflow $ pip install sklearn <script src="https://cdn.jsdelivr.net/npm/@tensorflo w/tfjs@1.0.0/dist/tf.min.js"></script>
6. Proprietary + Confidential Highly Interactive Source: Lorem ipsum dolor sit amet, consectetur adipiscing elit. Duis non erat sem $ step 1 – loss – 5.0681 (10 sec/step) $ step 2 – loss – 5.0681 (10 sec/step) $ step 3 – loss – 5.0681 (10 sec/step) $ step 4 – loss – 5.0681 (10 sec/step) $ step 5 – loss – 5.0681 (10 sec/step) $ step 6 – loss – 5.0681 (10 sec/step) $ step 7 – loss – 5.0681 (10 sec/step) $ step 8 – loss – 5.0681 (10 sec/step)
7. Proprietary + Confidential Access to Sensors Source: Lorem ipsum dolor sit amet, consectetur adipiscing elit. Duis non erat sem GPS Microphone Camera Gyroscope
8. Proprietary + Confidential Can Work Offline / Data Privacy Source: Lorem ipsum dolor sit amet, consectetur adipiscing elit. Duis non erat sem
9. Proprietary + Confidential Web Is Every Where Source: Lorem ipsum dolor sit amet, consectetur adipiscing elit. Duis non erat sem
10. History of TensorFlow.js
11. deeplearn.js • Release in August 2017 • GPU Accerelerated via WebGL • Allows Inference and Training in Browser
12. What it Offers?
13. With TensorFlow.js you can • Create models directly in browser • Import pretrained models for inference • Retrain imported models
14. TensorFlow.js Architecture
15. Proprietary + Confidential API Layers API OPs API Browser WebGL Keras Model TensorFlow Saved Model
16. A Quick Example
17. Let’s understand Basic Terms
18. What is a Tensor • It is n – dimensional Array • Each element in Tensor has same data type • Retrain imported models
19. const tensor = tf.tensor(0.1).variable();
20. What is an Epoch? • When you have huge data • Divide it in batches for training • Feed batches one by one for training • When all batches are done, that’s the EPOCH
21. What are the loss functions? • You have huge data for training • Feed data one by one for training • How to ensure model’s output is correct • A loss function maps decisions to their associated costs.
22. Create / Load Model in Browser
23. <script src="https://cdn.jsdelivr.net/npm/@tensorflow/tfj s@1.0.0/dist/tf.min.js"></script>
24. Using TensorFlow.js with NodeJs
25. npm install @tensorflow/tfjs-node
26. A Quick Example
27. Save Convert Load End Use Python Model with TensorFlow.js Python TensorFlow.js
28. Use Existing Model
29. Let’s See Some ML Examples
30. Text Classification • Detect text toxicity • Use prebuilt model
31. Object Detection • Detect Object Through Webcam
32. Image Classification
33. TensorFlow.js Limitations
34. Hiren Dave, AbhayTech Solutions LLP. @hjdave Thank you!
35. Q&A