Deep learning in your browser

How javascript could be used to harness the power of deep learning algorithms to understand the world around us. This presentation will help the audience to understand deep learning in a simple way using their existing knowledge in javascript, implementing and solving funny and real examples.

Transcript

1. Deep Learning in your browser
2. Kia ora I'm a Solutions Architect at Solnet PMC at Apache Foundation A Brazilian living in #welly, NZ You can check me @igorcosta Or drop me email at igor.costa@solnet.co.nz
3. What is A.I. ?
4. A.I. is when a machine can do things in a way that is indistinguishable from human behavior. Algorithm
5. This is not a fully awareness A.I.
6. In a much humbler reality like • Movie recommendations • Driverless cars • Drones • Natural Language processing • Predictive analysis • Image Recognition • Speech Recognition • Classification
7. A.I. today is a symphony of buzzwords
8. What is Machine Learning? Instead of writing thousands of use cases against a dataset to parse, extract and learn. You simply teach the algorithm to understand from the data. The output of that is a value between 0 and 1
9. https://github.com/thinkroth/Sentimental Sentiment Analysis
10. Recommendation https://github.com/axiomzen/Alike
11. Clustering https://github.com/anpandu/nodejs-text-summarizer
12. Machine Learning popular algorithms • Probabilistic outcomes - Naive Bayes, Gaussian Mixture Models, Logistic regression • Linear Classification problems - Decision Trees, KNN, SVM with Linear Kernel • Sequential Modeling/Time Series - Hidden Markov Model, Recurrent Neural Nets • Feature Learning - Autoencoders, Convolutional Neural Networks and other Deep Net Architectures • Nonlinear Classification Problems - Random Forest, SVM with RBF/Polynomial Kernels, Neural Networks • Clustering - K-Means and it's different implementations https://github.com/yomguithereal/talisman* * Collection of ML algorithms implemented in JS
13. What is Deep Learning? It's a branch of Machine Learning that mimics the human neuron with high level of data. The output of that is a value of Anything from Image to text or vice-versa.
14. playground.tensorflow.org
15. Image recognition example
16. Microsoft Cognitive Vision API https://www.microsoft.com/cognitive-services/en-us/computer-vision-api Google Vision API https://cloud.google.com/vision/ Nodejs - https://github.com/tejitak/node-cloud-vision-api Nodejs - https://github.com/viane/microsoft-computer-vision Output: You're fired, The Wall
17. To more complex applications Darknet Yolo Open CV + Deep learning https://pjreddie.com/darknet/yolo/
18. How an algorithm in Deep learning works?
20. To make Deep learning works you will need + Numerical operations are very efficient like 100x faster than CPU + Single Machine, no complication over splitting the load - Significant Memory constraint we can't train large data models
21. To make Deep learning works you will need
22. To make Deep learning works you will need
23. To make Deep learning works you will need + Supports much larger models like the latest Youtube label model with 21 Million labeled videos + Super Scalable - Too much complicated - Too expensive even in a cloud world.
24. Most popular toolsets • Tensorflow • Caffe • Keras • Theano https://github.com/aymericdamien/TopDeepLearning
25. How hard is to master in Deep learning?
26. Learning Deep Learning without a PhD
27. What if ... I could use Javascript on client side to training my networks?
28. ConvnetJS https://github.com/karpathy/convnetjs MindJS https://github.com/stevenmiller888/mind Synaptic https://github.com/cazala/synaptic
29. Why Javascript? Sometimes is faster, specially for small dataset
30. Thank you
31. Resources Great collection of articles about Deep Learning with practical examples https://medium.com/tag/deep-learning Gallery of Deep learning examples http://deeplearninggallery.com/ Youtube Channel with lots of introductions about Deep Learning https://www.youtube.com/channel/UCWN3xxRkmTPmbKwht9FuE5A Curated list of Deep Learning papers https://github.com/terryum/awesome-deep-learning-papers A collective list of dataset and resources for Deep Learning https://github.com/ChristosChristofidis/awesome-deep-learning APi.ai create chatbots for anything like (slack, Google Home, facebook messager) https://api.ai/ Deep learning by Google https://cloud.google.com/blog/big-data/2017/01/learn-tensorflow-and-deep-learning-without-a-phd