Journal Club [Daniel Kapla] (30 mins)

Fusing sufficient dimension reduction with neural networks

Daniel Kapla, Lukas Fertl, EfstathiaBura

https://www.sciencedirect.com/science/article/pii/S0167947321002243

Notes:

• ...

TensorFlow.js Workshop - Part I [Jeya] (90 mins)

Outline (tentative)

1. Why client-side TensorFlow?

2. Tensors

3. Image Tensors

4. Pretrained models and model zoos

5. Converting Python models into TF.js models

6. Training models

7. Backpropagation

8. Perceptrons and multilayered perceptrons (manifold learning)

9. Convolutions and pooling

10. Supervised learning (convolutional neural networks)

11. Unsupervised learning (autoencoders)

12. Generative adversarial networks

13. Tabular data analysis

14. Transfer learning

15. Federated learning

16. Big data analytics

Today's session:

1/3 - https://observablehq.com/@jeyabbalas/getting-started-with-tensorflow-js

2/3 - https://observablehq.com/@jeyabbalas/tensors

3/3 - https://observablehq.com/@jeyabbalas/image-tensors

Notes

• https://www.w3.org/TR/webnn, https://webmachinelearning.github.io/webnn-intro - W3C is standardizing interoperability with the underlying hardware.

• https://github.com/google/jax/issues/1472 (Praful)

• https://transcranial.github.io/keras-js-docs/ (Monjoy)

• https://blog.tensorflow.org/2018/04/a-gentle-introduction-to-tensorflowjs.html (Daniel) predecessor of TF.js)

• https://brain.js.org

• the toxicity model via script tag loading - https://cdn.skypack.dev/@tensorflow-models/toxicity

•