What are artificial intelligence and machine learning?
Artificial intelligence (AI) refers to solutions that suggest an intelligence that is typically only to be expected from humans. It can be divided into weak and strong AI. Weak artificial intelligence means that the AI can only master one specific use case.
Within the area of weak AI, machine learning (ML) refers to all learning processes that learn from existing data in order to apply this knowledge to new data. In principle, however, a classic algorithm can also be classified as artificial intelligence, provided it emulates human intelligence.
If human intelligence is artificially modeled, then this falls under the term artificial intelligence.
Machine learning refers to the use of learning processes to recognize patterns in existing data and apply the resulting pattern recognition to new data for decision-making.
Artificial neural networks are the architectural paradigm behind deep learning and many other facets of machine learning. The inspiration for this lies in the structures of the biological brain.
Buzzword-Factor
Experts predict that machine learning will be part of almost all software solutions in the near future and will also change society.
Entry hurdle
The use of machine learning requires major changes in companies, such as a data-oriented culture with the corresponding roles and complex infrastructure for collecting and processing this data.
Added value
Machine learning creates solutions that were previously not possible through pure specification. The insights gained and the significantly improved user experience provide a decisive competitive advantage.
What are the specialist areas of application for machine learning?
In particular, the machine learning solutions described in the press can rarely be used in a business context. AlphaGo and AlphaStar, for example, are specifically designed for game mechanics. In contrast, msg's generic use cases provide a modular construction kit. Depending on the business context, the right modules for the business case can be identified and combined.
Experience shows that machine learning can be used sensibly and with added value in almost any process or touchpoint. In most cases, the result is a significantly improved user experience. However, there are also many cases in which better decisions can be made and new business models can be developed.
Important for all areas of application: Despite all the fascination with artificial intelligence, the business case must not be lost sight of. After all, there are often existing digital solutions that are already good or even better than artificial intelligence. In some cases, the additional effort required to develop and introduce AI may not be worth it. However, in the case of new business models or improved decisions, considerable competitive advantages can be gained that pay for themselves more quickly.
TechDOSSIER: Advanced Machine Learning
Our TechDOSSIERs summarize important topics and trends in a product-neutral and compact way. We prepare complex technology aspects according to the information needs of executives and managers by explicitly highlighting certain characteristics of the topics and trends. Among other things, we provide definitions, describe the potential, name external influencing factors, set out an application scenario and explain the barriers to entry.
Machine Learning Catalogue: Modular construction kit for machine learning solutions
The Machine Learning Catalogue summarizes the building blocks for machine learning solutions in a structured manner. It lists the common and successful algorithms, specifies use cases, lists supporting techniques and assigns them to the typical task areas as well as various input and output data. This makes it possible to quickly decide which building blocks are useful, possible and combinable in a particular scenario.
What training and further education opportunities are available for machine learning?
We offer you both planned and tried-and-tested training courses on machine learning and deep learning, as well as individual coaching and workshops tailored to your needs. Get in touch with us!
- Target group: Technical experts, business analysts, prospective data analysts, managers
- Duration: 0.5 days (4 hours)
- Learning objective: Understand the benefits and limitations of machine learning, know typical application scenarios of machine learning, identify and integrate application scenarios in the customer context
- Contents: Introduction and motivation, areas of application and limitations, overview of ML functions and how ML solutions work, generic use cases and their function and benefits, areas of application and limitations, exercise to develop innovative and machine learning-based solutions in an individual context
- Prerequisites: none
- Target group: prospective data scientists, prospective data engineers, developers, technically interested people
- Duration: 2 days (16 hours)
- Learning objective: Understand the theoretical basics of machine learning, use machine learning in practice using examples
- Contents: Introduction and motivation, areas of application and limitations, learning methods from supervised to unsupervised to active learning, functional building blocks from value prediction to classification to feature recognition and others, methods and algorithms from linear regression to nearest neighbor to support vector machines and deep neural networks, machine learning pipeline and data science, overview of exemplary technology stacks, exercises for the most relevant functional building blocks and methods using a technology stack, overview of current trends such as TPU and SoC but also universal neural networks and marketplaces for models
- Prerequisites: none
- Target group: prospective data scientists, prospective data engineers, developers, technically interested people
- Duration: 2 days (16 hours)
- Learning objective: Understand the theoretical basics of machine learning, use machine learning in practice using examples
- Contents: Introduction to the function of neural networks, deep learning and network architectures, reinforcement learning with neural networks, convolutional neural networks
- Prerequisites: Machine Learning - Foundation
What is Natural Language Processing?
Natural language is the most intuitive interface for us to exchange information. Computers usually impose a very structured way of exchanging information via input masks. Natural Language Processing is trying to change this. But it not only plays a role as an interface between man and machine, it also ensures that we can access a large amount of data that is only available in natural language. NLP solutions consist of one or more of the following functions that make up an NLP pipeline.
How can language processing be simplified in machine learning?
With Holmes, we provide an open source library developed in Python for analyzing German and English texts. Holmes compares text passages with each other and searches for statements with the same or similar meaning. Sentences that obviously mean the same thing to a human reader can have completely different superficial grammatical structures on the one hand and use different terms for similar ideas on the other.
Holmes provides a general toolkit to meet the challenges of semantic text analysis. It supports use cases that become possible in the first place. Holmes is ideal for simplifying the configuration of chatbots, but can also be used as a basis for intelligent semantic search or for supervised document classification.
How does project management work in machine learning projects?
Machine learning projects differ from traditional projects in their approach. Instead of specifying technical algorithms and functions, the learning data that contains the decisive information for solving a problem must be identified. The specific use of machine learning can then be used to determine whether a technically usable output can be generated from this data.
This requires a very data-driven approach and an iterative process in which the machine learning solution is gradually refined. In contrast to traditional programming, the choice of technology and algorithms is not clear at the beginning, but must be found through experienced trial and error. This can lead to dead ends and unpredictable effort, even though our modular machine learning catalog already supports this step very well.
Our E3 approach provides a simple, understandable and reliable basis for this. We proceed in three phases: Explore, Evaluate, Execute. The business case always remains in view and we work stringently towards the goal. The approach can be used in agile, hybrid and plan-driven project organizations.
Is the machine learning scenario possible based on the data?
- What about the demand for and availability of the data?
- What statements can be made about the quality, structure and labels of the data?
- Are quantity and variability given?
- Do legal requirements such as the GDPR prevent the use of the data?
Is the integration and use of the machine learning scenario technically feasible?
- How can success and quality be measured?
- What quality is required?
- What are the costs of false negatives and false positives?
- What about sensitivity and specificity?
- Does the prognosis offer added professional value?
- How can forecasting be integrated into processes and applications?
- Which architecture and technology is the right one?
- What does the design of the machine learning pipeline look like?
- Is it online versus batch learning?
- Is it development versus runtime technology?
- How can the data sources be connected?
- How can they be integrated into the existing application landscape?
How are implementation and integration carried out?
- How can the machine learning model be perfected?
- How can the machine learning pipeline be implemented?
- How can the model and pipeline be integrated into applications and processes?
- How can an automatic data supply be ensured?