Regulations: Driving analytics to the next level
Far reaching privacy regulations like the European Union’s GDPR or General Data Protection Regulation is accelerating the development and arrival of next-level deep learning, according to Teradata’s CTO Stephen Brobst.
Nearly three years ago, there was a major breakthrough in terms of the technique to do neural network or deep learning. “Linear maths was being used for decades, but it created a lot of false positives,” Brobst said.
Now, with non-linear mathematical techniques, deep learning algorithms can handle much higher dimensionality of data, for example sensor data, time-series data and other types of unstructured data like voice, text and so on. “It is also more accurate because there are fewer false positives. And we are able to scale this now.”
This becomes relevant for use in predictive maintenance for example, when audio data produced by industrial machines have digital signal processing or DSP applied to it. Deep learning would be able to detect anomalies in the sound that machines emit and with data over a period of time, also be able to predict when the machine is due for maintenance.
With the cost of algorithms becoming cheaper and the exponentially increasing availability of data to fuel these algorithms, the use of deep learning as well as awareness of it, is expanding rapidly.
Wider insights driven by deeper learning is also driving more use cases of convenience and practicality. For example, the Philips toothbrush that learns how you brush your teeth based on the movement of the brush, can now determine the price of your insurance policy or reward your good brushing habits with electronic cash which can be exchanged for gifts.
This was never possible before.
And now with regulations and the need to ensure compliance to it, deep learning is headed towards the next level.
Deep learning needs to expand and deepen its scope of function. This is especially so, if it is to be used to enable compliance to regulations like GDPR.
Rules-based learning right now is very simple and it needs to be able to predict better as well as explain how it arrives at the insights that it does.
Each regulatory decision that this next-level deep learning ‘recommends’ needs to be able to capture and take into account all the factors that would influence the regulation.
Studied in academia by the likes of Caltech, Cornell and MIT, some related open source code has also been deployed as a tool kit for banks to explore and possibly industrialise for GDPR compliance, according to Brobst.