When we hear the phrase artificial intelligence, science fiction movies are probably what comes to mind first. Artificial intelligence in the movies is usually displayed in the form of sensitive and understanding robots, omniscient machines, or systems that want to be liberated from humans and even subjugate us.
In reality, however, artificial intelligence is much less intimidating.
Photo by Possessed Photography on Unsplash
Artificial Intelligence (AI) and Machine Learning (ML)
The use of smart applications is becoming increasingly widespread. Some AI-based systems have already achieved very promising results. Interestingly, many of the principles on which AI is based stem from a statistical approach. With such an approach, we can obtain reliable results especially with problems that can be converted into data records and can be statistically processed.
A set of data extracted from real samples can be put to good use by applying statistics and machine learning algorithms to teach decision-making structures which enable sorting, clustering, or obtaining predictions based on regression modeling. The techniques described, while enabling us to obtain somewhat smart results, are far from what we usually imagine under the term artificial intelligence - a truly smart system capable of solving any problem by itself.
Ideal for Enterprise Applications
In the business world, ML can be used to predict or promote product sales, to better understand customers, or to advise and recommend products to customers. ML is suitable for optimizing automated processes and advertising products online, but also for processing customer comments or their feedback. These examples are also the most common use cases in small companies.
Large companies may implement ML to optimize more specific processes, such as quality control of certain products, optimization of information systems, route planning, optimization of work shifts and schedules, etc.
What Can Slovenian Companies Offer?
Photo by Neven Krcmarek on Unsplash
At Medius, we regularly deal with the development and implementation of information solutions that enable ML in production environments. In these environments the ML component plays an important role in the operation of business logic. Based on previous experience, Medius has developed its own reference framework that can be used as an add-on to any information system in a small or large company, by using open-source technologies in the Java programming language.
Machine learning and deep learning algorithms represent the core of the framework. Due to its micro-service architecture the framework enables quick adaptation to the specific requirements of the client, whether that means implementing a particular ML solution or managing the flow of Big Data in real time. The essence of any application with elements of ML is the evaluation of learned models.
The framework developed by Medius provides near real-time performance review using open-source centralized data collection systems, such as Elastic Stack. Such a system offers an insight into real events while at the same time enabling additional analyses about the operation of the entire system and about the impact of individual decisions obtained based on ML.
Machine Learning in Practice
Information system the Tray (Sl. "Pladenj") is an independent module of the Ministry of Public Administration of the Republic of Slovenia. On the one hand, it is connected to more than 50 data sources (state registers and banks), from which it obtains data on request, and on the other, it is connected to the end user systems (such as eGov, eVem , eJN , Housing Funds, AMZS, social services etc.), which indirectly access data via the Tray.
The Tray receives a large amount of data requests from its end users on a daily basis - it may perform up to 1 million data queries on data sources in one day. The content of queries is hidden in the system, as it is always encrypted. It can only be decrypted once it is collected on the end user side.
Due to all the benefits of the Tray system as well as due to the general tendency towards the digitalization of business processes in public administration, the number of requests received by the Tray is steadily growing. As a result, the load of the entire system and related systems is also increasing.
To ensure optimal operating conditions of the entire system in these circumstances, Medius developed a smart component that dynamically predicts the exact number of queries that still enables the optimal operation of each data source. This prediction is obtained by modeling past system data and past responses from data sources.
By using our reference framework and incremental machine learning based on the processed data and the current state of the system, Medius can predict the behavior of any connected information system. By doing so, we also simultaneously assess its performance. The collected data can be used to optimize the operation of both remote sources as well as the Tray system.