![](http://i.imgur.com/opYvKPd.jpg)

This post is an attempt at a high level drone's view of the machine learning ecosystem and an experimental educational post format. Machine learning evolution is a significant contributor to tectonic shifts in the structure of social and economic systems. You should pay attention.

Machine learning ecosystem is huge and amorphous, it has no well defined boundaries and machine learning can be found in many facets of our hypermodern life. 

## ML computer scientists and developers

Machine learning specialists are in high demand and their salaries are among the highest in the computer science industry in the USA: [$120k - $180k](https://www.glassdoor.com/Salaries/us-machine-learning-salary-SRCH_IL.0,2_IN1_KO3,19.htm) per year.

## Researchers and academy

Due to the superior salary and benefits offered by tech companies, there is a [leakage](https://www.ft.com/content/298e2ac0-b010-11e6-a37c-f4a01f1b0fa1) of researchers and scientists from the academy. This is not that great and can lead to progress slowdown in the field due to the reduction of openly published studies.

## Learning platforms

I find it telling that two (Coursera, Udacity) of the three (edX) largest massively open online courses platforms have some machine learning related history. 

Initially, [Udacity](http://udacity.com) was launched by Sebastian Thrun and Peter Norvig as an experimental course at Stanford University on artificial intelligence. Udacity launched as a generic courses platfrom but since evolved to specialize in courses related to machine learning and artificial intelligence. A return to its' roots so to say.

Sebastian Thrun has contributed in such projects as Google Self Learning Car, DAPRA Self Driving Car, Challenge Ways, Google Street View, Google Glass, Google X.

[Coursera](http://coursera.com) was co-founded by Andrew Ng, an American-Chinese scientist, who led the Artificial Intelligence Group as part of Baidu, a professor at Stanford University. One of the first courses launched on Coursera was Machine Learning taught by Andrew Ng. 

## Startups and megacorporations

Machine learning is embedded in the organizational DNA of megacorporations. In many respects, megacorporations are the driving force behind most recent machine learning advancements. Megacorps also capture a lot of value created by ML systems. As a prominent example, at Sundar Pichais' Google, artificial intelligence (but it's really machine learning) is [everywhere](https://www.fastcompany.com/3065420/at-sundar-pichais-google-ai-is-everything-and-everywhe).

The angel.co startup database has [2057] entries (https://angel.co/machine-learning) for startups working on machine learning related products and services.

There were [24 machine learning start up acqusitions](https://index.co/market/machine-learning/acquisitions) by megacorps in 2016 tracked by index.co. At the end of March 2017, 9 ML startups have been acquired.

Investments in the ML industry have also shown a significant [increase](https://venturebeat.com/2017/02/12/before-you-build-another-machine-learning-startup-read-this/).

### Machine learning in online search

[RankBrain](https://www.bloomberg.com/news/articles/2015-10-26/google-turning-its-lucrative-web-search-over-to-ai-machines) is a machine learning system that helps Google to process search results and provide more relevant search results for users. The main objective of this [algorithm](https://thenextweb.com/artificial-intelligence/2016/09/02/machine-learning-just-got-more-human-with-googles-rankbrain/) is to help to provide quality results for rare or even one-of-a-kind queries. If the machine sees an unknown request before it, it tries to extrapolate what the user had in mind or which words or phrase have a similar meaning. 

### Machine learning in communications

[Google's Neural Machine Translation System](https://arxiv.org/abs/1609.08144) is a current Google Translate algorithm that is based on neural networks and represented a new approach to automated translation.

[Skype Translator](https://blogs.technet.microsoft.com/machinelearning/2014/12/25/skype-translator-puts-machine-learning-to-the-test/) is an automated real time translation system  available for 8 languages in a voice mode enabled by machine learning algorthims.


### Machine Learning in Finance

Machine learning in finance is applied, among other usecases, to increase the accuracy of price predictions if assets, stocks and currencies. A lot of hedge funds are betting on machine learning based trading vs human traders. According to market research firm Preqin, [1,360](https://www.wired.com/2016/01/the-rise-of-the-artificially-intelligent-hedge-fund/) hedge funds make a majority of their trades with the help from computer models.

An interesting fact: about a third of Goldmans Sachs employees are [developers](http://www.businessinsider.com/goldman-sachs-has-more-engineers-than-facebook-2015-4). Though only some of them are working with machine learning, it's an important indicator of how computer science and finances are intertwined with each other in the hypermodern world.


### Machine Learning in politics

[Cambridge Analytica](https://www.extremetech.com/extreme/245014-meet-sneaky-facebook-powered-propaganda-ai-might-just-know-better-know
) is a relevant example of application of machine learning algorithms for political purposes and propaganda. 

## Crypto currency

Machine learning and crypto currenciea are mostly orthogonal. Though there are most certainly a lot of attempts to predict exchange rate fluctuations using machine learning.

## Chip manufacturers

Machine learning is done with GPUs (graphics processing units). The CPU architecture has a set of tradeoffs that are not suitable for massively parallelized computations (as used in machine learning). Unlike the CPU, the exponential growth of GPU performance per 1 conventional unit of money continues. Nvidia and AMD are two major players in this field with Nvidia having a significant lead.

The latest trend of interest is Google [adding ASICs to the Google Cloud platform](https://cloudplatform.googleblog.com/2016/05/Google-supercharges-machine-learning-tasks-with-custom-chip.html) specifically designed for machine learning. 

## Cloud solutions

Large cloud services companies are providing specialized GPU servers that mainly used for machine learning: [Amazon](https://aws.amazon.com/about-aws/whats-new/2010/11/15/announcing-cluster-gpu-Instances-for-amazon-ec2/), [Google](https://cloud.google.com/gpu/).

Megacorporations and startups provide MLaaS (machine learning as as service): preinstalled, preconfigured, optimized systems that can reduce initial time and effort investment by an order of magnitude.

* [Google ML engine](https://cloud.google.com/ml-engine)
* [Microsoft Azure ML](https://azure.microsoft.com/en-us/services/machine-learning/)

## Open source 

A non-exhaustive list of popular programs and libraries for machine learning:

* [Tensorflow](https://www.tensorflow.org/) by Google
* [Keras](https://keras.io/)
* [NTLK](https://github.com/nltk/nltk)
* [Scikit-learn](http://scikit-learn.org/stable/index.html)
* [Prediction.io](http://prediction.io)