I presented Forecasting A brief history of Machine Learning and a look to its future at Data Science Speakers in November 2018
This presentation focused in summary of the material of the Google Cloud Platform Specialization in Machine Learning and a personal view of what we will be expect in the coming years
Below are the slides and the material that I went through
And here is the transcript:
Machine Learning dates quite a long way back to the 19th Century. The first type of machine learning was a Linear Regression and one of the first uses was the size of growth of vegetables, the size of humans and even the position of planets. Humans have been always interested to find patterns and relationship of the things that happen around us so the first attempt to establish these patterns by rules were by approximating them with a straight line
But not all patterns are that simple and not everything can fit with just a straight line
After the end of World war 2, in the 40s the perceptron model arrived, the precursor of the neural networks. This model tries to mimic the way that neurons works in the brain and by using the basics concepts of Linear Regression these model are further combined with non linear transformations to fit more complex patterns.
Around the time that the man landed on the moon, these kind of models got more complex by adding more layers of these transformations..
..and well sometimes these models got quite complex that time to compute them was really slow, it was still the 60s!
When Michael Jackson was dancing thriller on the 80s a there was a new wave of machine learning models. First the decision trees tried to explain patterns by using more simple rules. For example this model explains the patterns of the people that died in the Titanic by just splitting the data with yes and no rules and using gender, cabin class and age
Also in these way of new models, Support Vector Machines arrived with models that tries to create the best boundaries to find patterns in the data
Now in the new millennium with more processing power and also morecapacity to store more data new models arrived. First the Random Forest that are a collection of hundreds of decision trees improved the accuracy of prediction
In the recent years the neural networks are back again but now more complex than ever in the shape of deep neural networks. These kind of models are trying to solve the most difficult problems right now like developing driverless cars, talk with your Alexa like if you were talking with a human and even finding a cure for cancer
What is the future of ML? Well I think it is a bright future because
-There is more data being collected by all the devices around like our smart phones, Alexa speakers and fitness trackers
-There is more computer power and cloud services that provides the platform to develop machine learning at a really low cost
-There is more people using and developing Machine Learning with open source software and global collaboration that is speeding the research and development of new algorithms
Hopefully you have enjoyed the history of how a machine learns!
For can find more details about Data Science Speakers in here