The digital revolution requires an ever-increasing number of repetitive and targeted decisions.
The digital revolution is faster and more comprehensive than the Industrial Revolution at the beginning of the last century. It requires less capital, and focuses on intellectual and digital innovation, which is affordable to many. The innovations produced have therefore had a much more immediate impact on the market. A wide variety of services are now offered by the leaders of this digital revolution, including Google, Amazon, Facebook and Apple. This has resulted in a rush to set up data-oriented cultures and predictive algorithms, to increase company knowledge and automate more business processes.
The best way to manage these is through machine-learning algorithms.
Artificial Intelligence (AI), and in particular the development known as machine learning, promise to automate mundane tasks such as driving a vehicle or recognizing individuals in a crowd. Indeed, more and more, these techniques mimic human behavior: learning, classification, correlation, prediction, decision-making. They reduce the number of scenarios and help to anticipate events by learning automatically from the past. Without these new technologies, it would be impossible to predict all possible scenarios in advance.
Machine learning will not replace humans
I do not think machine learning will ever fully replace humans. Instead, its added value is to offer “augmented knowledge”. This is especially useful for more qualitative work, particularly repetitive and targeted but complex tasks, such as credit granting, predictive maintenance, and fraud detection.
Machine learning techniques are not new
SAS has offered them in its predictive modeling solution (data mining) since the 1990s. They rely on both traditional statistical techniques, such as logistic regression, and innovative techniques such as random forests and neural networks. These techniques are therefore mature and well understood, but also take advantage of the latest evolutions in technology. These include Hadoop to access huge data tanks and the ability to compute in parallel on multiple clustered servers.
So what innovations has SAS made for 2017, to promote this new culture of “augmented knowledge” using predictive algorithms?
There are seven major areas of SAS innovations for machine learning:
- Exploitation of all types of available digital data: texts, images, SQL or NoSQL databases: Hadoop can store and process petabytes of data without format constraints at a very low marginal cost. SAS offers to process this data from the Hadoop cluster without changing the format, loading and processing them in-memory.
- The availability of all known analytical techniques supporting all types of decision-making. Addressing business problems like demand planning may require a combination of several predictive techniques like segmentation, prediction, recommendation, forecasting, operational research and so on. SAS offers to process all of these analytical techniques in-memory for better performances.
- The ability to ease the collaboration of three different roles within the company (IT, data scientists and business units): People are at the center of the machine learning process. Predictive models are only effective if operational processes benefit. Having data scientists work in their ivory tower is not going to be effective. SAS offers to IT, data scientists and business units to work together on a common process through an integrated analytical platform.
- The provision of predictive algorithms to application developers from a public API: Integrating the augmented knowledge created by machine learning into operational processes rely on opened APIs to these SAS technologies for both data scientists and application developers regardless of the programming language used (Python, R, Java REST…). SAS provides public APIs to its analytical platform.
- Unlimited computing power for parallel predictive modeling. SAS has revised the foundations of its analytical platform, originally developed for large servers with shared memory. It has now developed to work on distributed servers, lowering machine costs and offering much more power through scale out on multiple servers.
- Running algorithms anywhere: at the heart of databases, server memory, connected objects, or public or private cloud. Deploying the results of machine learning could take several months, because of the rules of integration of new functionalities, which sometimes require a rewrite. SAS offers the ability to augment the company’s knowledge at the source of information, at the heart of a machine, database, or application.
- Governance of algorithms to ensure their reliability and robustness over time. SAS technologies automate the deployment of predictive algorithms in real time if needed, while respecting the rules of security and integration. This level of automation could not be done without a solid governance of the predictive models to assure that this automation is completely reliable from a business perspective. SAS offers the solution to govern the full analytic life cycle with high confidence.
- Machine learning is a new technological revolution. To take full advantage of it, companies could benefit of those 7 areas of SAS innovations available now with the new SAS® Viya ™ in-memory analytical platform. This analytical platform is particularly suitable for new business processes relying on experimentation and agility while guaranteeing security and robustness.
Additional reading: Josefin Rosén and Britta Skriver discussing on Machine learning and the evolving intelligence landscape.Download a SAS Best Practices eBook: The Machine Learning Primer