![](https://www.yuriburger.net/img/leap2019-4.png)
Notes (not a complete recap today) of day 4 of the LEAP event (day 1, day 2, day 3).
Machine Learning Fundamentals
The real measure how well a ML model is performing is how well it works with data it has never seen. The training phase could use 70% of data and the Test phase the remaining 30%.
![](/2019/02/01/leap-2019-day-4/images/machine-learning-algorithm-cheat-sheet-small_v_0_6-01.png)
https://docs.microsoft.com/en-us/azure/machine-learning/studio/algorithm-cheat-sheet
More information here: Hackaton Blogpost
Neural Networks are one type of algorithm. Deep Learning is based on Neural Networks.
Activation Function to get the values between 0 and 1 and are an important feature of the neural networks. They basically decide whether a neuron should be activated or not. Whether the information that the neuron is receiving is relevant for the given information or should it be ignored.
![](/2019/02/01/leap-2019-day-4/images/dcn-rnn.png)
More for you to study on:
- http://www.asimovinstitute.org/neural-network-zoo/
- https://news.microsoft.com/apac/features/ai-for-earth-helping-save-the-planet-with-data-science/
How do we test a trained RNN model (like the word RNN or the character RNN)? We measure perplexity per word!
“PerplexityΒ is a measurement of how well a probability distribution or probability model predicts a sample. It may be used to compare probability models. A low perplexity indicates the probability distribution is good at predicting the sample.” Source: https://en.wikipedia.org/wiki/Perplexity
Deep Learning requires large datasets. Transfer Learning tries to counter this by taking a different approach by using a big source of data that is already trained (for example from Imagenet). Chop of the last layer and add your own inputs and train from there.
![](/2019/02/01/leap-2019-day-4/images/20190131_091004.jpg)
More information: check these out
Introduction to New Azure Machine Learning Service
The Azure Machine Learning Service is brand rew, was released December ‘18. It is considered a code first service. You can use one of the deep learning frameworks and transfer into ONNX (Open Neural Network Exchange)
- TensorFlow
- PyTorch
- Scikit-Learn
- MXNet
- Chainer
- Keras
Demo: https://github.com/maxluk/dogbreeds-webinar
AI & Cognitive Services
AI is about amplifying and augmenting human ingenuity. Azure Cognitive Services is here for you if you want to start today. Microsoft offers prebuilt AI for your business: infuse apps, websites and bots with human-like intelligence.
Start AI with ethics (“with great power comes great responsibility”) https://aka.ms/ai-ethics
![](/2019/02/01/leap-2019-day-4/images/untitled.png)
Demo on AI: https://aidemos.microsoft.com/
Create your own Computer Vision projects: https://customvision.ai
Create a bot in 3 minutes: https://www.qnamaker.ai/
Analyze video content: https://www.videoindexer.ai
Support for 6 key AI capabilities through containers:
- Key Phrase Extraction
- Language Detection
- Sentiment Analysis
- Face and emotion detection
- OCR / Text Recognition
- Language Understanding
https://docs.microsoft.com/en-us/azure/cognitive-services/cognitive-services-container-support