Machine Learning (ML) is one of the main drivers of digital transformation. It not only enables deeper insights about the users of our products or the visitors of our shops and portals but also predicts their future behavior and is able to take decisions on these predictions.
ML is part of Artificial Intelligence (AI) which tries to mimic human intelligence with machines. It is also called the "weak" or "narrow" part of AI because it tries to solve one or a few specific problems at a time, f.e. recognizing faces on pictures or extracting the meaning from speech or text.
In ML you do no longer program rules but instead train the system to learn patterns by feeding it input data. ML creates algorithms that are able to make predictions on new data. You can combine this intelligence with sensors and take automated decisions if you like i.e. automatic control of temperature or light in your home.
The term "Deep Learning" is a part of ML and is used for newer neural networks. These deep neural networks have multiple hidden layers between the input and the output layer; each layer providing learning output to the next one. Thus, Deep Learning is used for training complex pattern matching problems like face or language recognition. For training deep neural networks the amount of input data and efforts for preprocessing is much higher than "normal" ML.
You can find a good commonly understandable introduction into ML in the slide deck from Jason Mayes:
The purpose of ML is to automatically learn patterns, apply them on new data, predict the desired information and optionally take decisions or actions from these predictions. Raw data input may originate from text, pictures, videos, speech, behavioral or master data; data may be delivered in discrete files or as data streams, i.e. from sensors.
Two main areas are predestined for ML usage at Haufe Group:
1) Product view: How can ML support in creating data-driven or self-optimizing products?
2) Marketing view: How can ML support Marketing teams to better understand the users/visitors and their context and provide the next best action to them?
Other application areas like self driving vehicles, medical diagnosis, games are mentioned under "3. others".
Mostly all current initiatives are in the state "Discovery" - proofing the value for us and our customers.
|Topic||Area of application||Customer Need / Use Case / User experience||Haufe initiative|
|semantics||search algorithm, extract semantic context and return most relevant results||UX: Get quick and relevant answers/solutions to my email@example.com|
|word embedding||UX: get relevant help comments for creating contracts||word embedding in tenancy law firstname.lastname@example.org|
|entity recognition, natural language processing (NLP)||convenient document/bill capture and assignment||lexoffice email@example.com|
|speech recognition, device control via language (Siri, Google now, Cortana, Amazon Echo)||Users want to interact with devices in a fast, natural and efficient way. Next evolutionary step: natural spoken language||?|
|product usage||finding patterns in product usage to prioritize or filter content for the user||UX: improved user experience||?|
|personal digital assistants||Train personal digital assistants to||Alexa skill for reading Haufe Newsfeed - HCP firstname.lastname@example.org|
|Outlier detection, anomaly detection||Log analysis, root cause analysis for bugs||UX: less bugs, experience of a bug-free app||Lexware offline log analysis email@example.com|
|correction of user input data in apps||UX: User is satisfied, because he gets a more accurate tax return||smartsteuer.de firstname.lastname@example.org|
|Topic||Area of application||Customer Need/Use Case / User experience||Haufe initiative|
|Clustering / classification / regression (Datamining)||identify segments of users with the same behavior, predict their probability and take actions on that i.e. churn prediction||get more relevant information or offers dependent from the customer's context||"Next Best Action" - SME email@example.com|
|Decisioning||optimized A/B-Testing with contextual multi-armed bandits||display most relevant content/offer dependent from the customer's context||"Next Best Action" - SME firstname.lastname@example.org|
|Fraud detection||prevent fraud by identifying potential visitors using ML||intern: avoid additional costs||Lexware Shop email@example.com|
|Automotive: self driving vehicles (cars, trucks)|
|Affective computing - recognition of emotions in faces, speech|
|Picture recognition (automatic tagging)|
|Security (public places, traffic)|
|Games: intelligent agents (deep reinforcement learning) i.e. "Alpha Zero"|
|Smart home: ML combined with sensor technique|
The following topics are intended to make you curious and to provide you with new possible ideas to continue investing on your own.
If you need help or consultancy, want to contribute ideas, need ML support for your own idea or cell kick-off, want to discuss topics or get development support in ML issues, feel free to contact the people mentioned in the HG initiatives above or contact