Data Science &
What are Data Science and Machine Learning?
Data science is concerned with the collection and exploitation of data in science and technology. Data is only useful to the extent that raw data can be structured into patterns that allow useful information to be extracted. Machine learning is concerned with equipping machines with an ability to detect patterns in data that are characteristic of the data source rather than a chance occurrence in the particular data. This ensures that the pattern can be exploited to extract information from further data from the same source in a variety of ways depending on the type of data/application and the type of pattern.
There is now a rich, well-engineered set of methods that can achieve this goal for specific well-defined Machine Learning tasks such as classification (identifying one of a small set of possible categories for an object), regression (learning a real-valued function), etc. Hence Machine Learning is a part of data science. Machine Learning is becoming a key component of data science in that without the ability to detect patterns, data will only be useful if patterns are pre-coded or implicit in the structure of the data, something that becomes increasingly unlikely as data sources diversify and become heterogeneous. Machine Learning (ML) has also brought about a major revolution in the advancement of Artificial Intelligence (AI).
This has been effected through what might be termed ‘soft Artificial Intelligence’ where the introduction of pattern learning has enabled systems to perform in ways that seem to imply real intelligence. Examples are the breakthrough with MoGo, the artificial Go playing system that has started to match human level performance, something held as an open core AI challenge until very recently. Following this success ML has also been a key part of combining deep learning with reinforcement learning to play Atari games. ML here is an enabling technology that helps systems perform as if they were intelligent.
Future of Machine Learning
Machine Learning is already playing a very important role in society through the many online services that we use that make use of the technology. Examples are the ranking of responses to search queries, the placement of advertisements on web pages, and recommendations of news on news websites. This is only the beginning of the use of the technology and of the impact it can make through more advanced AI systems as discussed above.
Given the extraordinary advance effected by combining two well understood components in the Atari games system, it is likely that further advances will be rapidly forthcoming. There is no doubt that the technology is already able to deliver very significant value in many different applications and use cases. The range of potential applications will undoubtedly increase as further advances are made. Perhaps the key to securing further advances is understanding how to engineer more complex systems with machine learning components.
There is a need for both theoretical frameworks and practical experience to both understand and reliably predict properties of the composite systems. They are numerous theoretical challenges that arise once machine learning is applied in varied and changing environments that are no longer covered by current frameworks. Knowledge about the practical implementation of systems that learn from interaction with users is something that is largely owned by large corporations such as Google, Amazon and Facebook. This is not a desirable situation for societies that could benefit hugely from this expertise being more widely available. It is certainly the case that learning from interacting with users can deliver a significant competitive advantage for telcos wanting to understand their subscribers’ interests and motivations in order to reduce churn and increase average return per user (ARPU).
Extracting value with Machine Learning: the Scentrics case study
Scentrics enables users to create, store, share and distribute content while guaranteeing privacy. This is achieved through a novel key distribution architecture that enables a balance to be struck between privacy and legal disclosure. The viral nature of this service promises to create a rapid uptake both for delivered applications and additional services that telcos are planning from bespoke enterprise solutions through social media to mobile money.
Scentrics security means that message contents and social media posts will no longer be accessible to the systems providing these services. This will disrupt their ability to use machine learning methods to profile Scentrics users in order to direct advertisements and other content as discussed above. There will be a need to extract relevant features from the messages and other content in order to apply these techniques.
Scentrics will be able to offer a service to such businesses that extracts these features without compromising the users’ privacy, hence satisfying both the business requirement to profile users as well as users’ demands for privacy. Partnerships that enable sharing of features in order to maximise the value of customer information hold the promise to deliver to users both useful information and the level of privacy that they desire, while placing telcos in a key brokering role.