Machine learning

Make big data manageable!

Machine Learning (ML) is a sub-area of artificial intelligence. By recognising patterns in existing data sets, IT systems are able to find solutions to problems on their own.

Prototypes and demonstrators available

Cross-industry deployment

Suitable for SMEs?

State of play

At present, a large number of machine learning implementation and development projects can be found. The deep learning sub-category in particular highlights many new developments, for example in the fields of facial recognition or speech recognition.

If you want to start your own development project, use the programming language Python is recommended. The open source availability of numerous programming libraries, such as Google TensorFlow, scikit-learn or Theano, makes it possible to quickly integrate and exploit machine learning techniques.

Recognising patterns in large amounts of data requires both high storage capacity and advanced knowledge in machine learning. Cloud providers (Amazon, IBM, Google, Microsoft, etc.) are therefore now offering model recognition as a ‘machine learning as one service’ within their platforms.

Technology and deployment

Description of technology

As the fourth industrial revolution advances, business and process-related data are increasingly being collected. It is therefore becoming increasingly difficult for humans to interpret and evaluate them. On the basis of existing data, machine learning serves to identify estimations and patterns within the structures, thus enabling solutions and conclusions to be developed.

The Deep Leaning sub-category also describes a method for the production of artificial neuronal nets. They are perceived in their function and structure as a human brain. By reading large datasets, important features can be extracted and classified automatically. Smart evaluation and forecasting are thus possible.

Picture: Based on Towards Data Science

Possible use scenarios

There are currently many applications related to machine learning. Machine learning can be used anywhere in order to analyse large amounts of data and search them according to patterns.

In everyday environments, the most well-known applications are Amazon’s and Netflix’s recommender services, spam mail sorting, speech recognition (Siri, Cortana) or facial recognition (Facebook).

In the production environment, quality assurance, customer satisfaction, autonomous systems or predictive maintenance can continue to benefit machine learning.

There are also first fields of application in the office environment. For example, chatbots can be used to communicate. You can ask questions and receive information, recommendations for action or proposed solutions.

Gradual introduction

Step 1: Definition of the application scenario

Provide a concrete description of the objectives. What do you expect from using machine learning? Define criteria that are crucial for solving the problem.

Step 2: Ml-mapping

Record your problem so that it can be handled by machine learning methods and algorithms. Reduce the overall complexity to individual, easily manageable problem cases.

Step 3: Edit data

Data-collection:

A crucial step is to ensure the appropriate data structure. Do you currently collect the correct data/characteristics?

Data exploration:

Based on your experience, conduct an evaluation of the data in order to determine whether the data has been processed in such a way that it is possible to analyse it. This step is particularly important to understand which categorisations of the data they should carry out or which input variables will be needed for the algorithms. If necessary, you may carry out preliminary studies of the changes resulting from the variation of individual data.

Step four: Modelling

On the basis of the problem definition, different algorithms for data analysis can be used (Decision Trees, Random Forest, etc.). If you already have skills in your business, you can now implement Machine Learning Tasks. If this is not the case, it is recommended to commission experts or, where appropriate, to use machine learning-as-a-service functionalities in cloud platforms.

Step 6: Evaluation

The ML model now needs to be continuously monitored in its application. It is necessary to verify whether the datasets have been correctly evaluated in relation to the objective. Adjustable screws leading to an improvement of results:

— Increase in the volume of data

— Clear categorisation

— Reducing complexity

— Improving data quality

Opportunities for SMEs

Identification of complex relationships

Increased efficiency

Identifying potential for optimisation

Making predictions

contact

Do you need support in setting up your business?

Use our technology radar to keep a look at the main technologies relevant to SMEs!

Back to technology radar