I attended the ohio devfest by GDG (google developer group)
https://ohiodevfest.com/
Get talking with Actions on Google
The keynote was about Actions for google by
and
You can make a action for google app using dialogflow templates
You say something to google assitant which
INVOKES your app. It relays the message to your app which detects the
INTENT of the the user and pairs it with an
ACTION.
From there it does the action and returns a response to the user.
The speakers demoed their movie maven app which helped the user find a movie based off a genre.
You can reserve keywords which the google assitant will understand and do a implicit invocation of your app.
Predicting the Internet of Things With Tensorflow
Taron foxworth talked about internet of things, machine learning.
What is IOt 95% of all consumers wil lhave a chip by 2020. as avg devices become more intelligence, consumer will start to expect it.
microcontroller : tiny computer run 1 program again and again.
gateway can be as small as a pi or a big server serving one purpose connecting things.
arduino connected to piezo buzzer making a vibration. demo trying to show increse vibration and at certain level stop. platform around notifcation and control. macine learning track subtle changes not perceptible by humans. machine learning is glorified statitstics, tweaking own statistical parameters. process iterate it over and automate the adjustments. Keras
- Training a model: model is a function which takes input which you define. It is very strict. gives you output. very consistent. Training looks at data in supervised way. looking at all data and trying and trying and trying and see a pattern. if you have examples for all your cases. machine has to break first.
- Building the model : qualitative -> quantitative . Monday to Sunday becomes 0 to 7. break complex to simple which machine can understand. reframing your question very important.
Convolutional Neural Networks : what group of pixels I have seen before. Other thinigs low confidence.
let training determine what the right answer is. By playing with the training, number of layers, number of data points.
Keras is an open source neural network library written in Python. It is capable of running on top of MXNet, Deeplearning4j, Tensorflow, CNTK or Theano.[
keras
https://www.tensorflow.org/serving/
for production you upload to google and google scales it. just an api call.
https://github.com/tensorflow/models
models can train other models. just looking at numbers.
data scientist trains the model, build the model , dev ops take to production because it is just api from that point..
machine learning not on cloud running on edge. what happens if inernet goes. need to control the pipleine.
everyday examples of AI and ML
self healing machines. as a mchine if I know I am failing, I can change my arm.
Actions on Google Code Lab
https://codelabs.developers.google.com/codelabs/androidthings-assistant/#0
Using Popular Machine Learning Models and Techniques
https://github.com/ksasso/MLmodels_inR_GDG2017
have 100 people predict your height,
prediction error
resampling methods
result are not sample specific
k fold ramdom sample
tree based method for regression and classification
take the data and segment it ba
https://en.wikipedia.org/wiki/Decision_tree_learning
Favorite Resouces: Machine Learning Concepts
https://www.r-bloggers.com/in-depth-introduction-to-machine-learning-in-15-hours-of-expert-videos/
https://www.coursera.org/learn/machine-learning
https://www.datacamp.com/
https://www.analyticsvidhya.com/blog/2017/09/common-machine-learning-algorithms/
https://www.womeninanalytics.org/
fireside chat
tools most excited about : kotlin, VR, lithtml firebase probabilistic programming dso tensorflow
tech coming in near future : autonomous vehicles, tensor flow on mobile devices, block chain
3d print
https://techcrunch.com/2015/10/03/the-blockchain-might-be-the-next-disruptive-technology/
favorit goog api
tensorflow
google home
places api
firebase
vision analysis api
go to stack overflow and look for unanswered questions and solve them
