Adaptive Machine learning or how to analyze environment´s volatility

The environment is constantly changing. Although we may think the environment is immutable, that the data collection on people and the environment hardly changes, there is nothing further from the truth. Data changes so rapidly that predictions based on Machine learning techniques become soon obsolete. Why?

According to Gartner´s latest trends curve, one of the technologies that will flourish in the next years will be the Adaptive machine learning, also known as Adaptive automatic learning. This technology allows for continuous learning with online machine learning models and in real-time. This capacity allows machine learning models to adapt themselves to the constantly changing world.

Due to this high adaptation capacity, this technology is particularly useful for autonomous cars training. These cars must be able to incorporate new data in real-time, analyze it and make decisions based on this data.

However, the use of this technology goes beyond autonomous cars (which Gartner gives over 10 year-period+ to see one). Adaptive autonomous learning in real-time requires efficient reinforcement learning, in other words, how an algorithm must continually interact with its environment to maximize its reward. These algorithms would work for agriculture, online marketing, smart cities, and even financial institutions or any other industry using IoT.

In these changing environments is not possible to collect all the generated data, organize and quantify it and train a “traditional “machine learning model as it must be retrained and make decisions in real-time.

The idea that emerged in the 50s

Although it may sound very innovative, the truth is back in the 50s B.F. Skinner developed a training machine focused on skills development. The idea was to monitor students´ progress in programmed learning. The machine adapted the questions based on the students´ previous correct answers and provided them the capacity to learn at their own pace.

It was not until 1970 that this idea was applied to artificial intelligence. However, as happened with many other things back then, the development of the adaptive automatic learning was slowed down due to computing capacity, capture and data storing. Simply, they were not ready yet.

Despite all this, Adaptive automatic learning did not die there. In the last decades, these technologies became more agile, scalable and easier to use although not widely used.

Nowadays, there are a myriad of fields for using adaptive machine learning, besides autonomous cars or robots. In the financial field, this technology allows for online data incorporation, for example, to monitor the stock market. It is also being used for location prediction (very useful for autonomous cars), advertising and in insurance field.

According to Gartner, adaptive machine learning is still a challenge as systems based on self-learning and autonomy will require considering aspects such as privacy, ethical aspects, and security. On the other hand, companies will apply continuous learning for autonomous and smart decision-making.

This doesn´t mean current methodologies based on machine learning must be changed as once the off-line algorithms are trained, the adaptive learning allows to improve, contextualize and personalize even more existing machine learning models.