IAP-GmbH

Intelligent Analytics Projects

General Advanced Analytics

We offer advanced data analysis consultation with state-of-the-art analytics tools. Our statistical toolboxes include explainable artificial intelligence solutions among many other and are applicable to various data ranging from simple and small datasets over time series to complex high-dimensional and big data. Our field of expertise lies especially in artificial intelligence, classification and time series forecasting. One of our peculiarities is the interactive use of complex mathematical algorithms. The principle of human verifiable mechanics within the workflows and explainable results that are comprehensible to you is most important to us.
The following paragraphs will show some exemplary solutions in a descriptive way.
On the right-hand side, a published analysis of the income of the Germans is shown and it depicts four data generating processes according to our analysis which implies four different groups of people in Germany, which may be interpreted as social classes.

Proof of Concepts and Prototype Construction

We are able to extract knowledge from data and present it in comprehensible ways. In order to achieve this aim, we follow a workflow shown on the left which proved itself by experience. We develop our own tools for Data Mining explore intepretable machine learning tools for explaining the results in order to gain knowledge. The proof-of-concept (PoC) shown on the left only shows a simplified version. In reality, many processes are circularly and require sound communication with experts of the application field.

Predictive Maintenance

Utilizing intelligent data analysis, we enable our clients to plan maintenance by foresight. In order to do that, it is necessary to identify data patterns and outliers within priorly observed data. In one project, we used IOT data and pattern recognition to predict malfunctions of devices during lifetime and during production.

Automated Forecasts in Supply Chain Management

Management can be supported with smart computational algorithms in daily work. For example, machine learning methods can be automatized to forecast the demand in supply chain management (and in various other applications). Daily, weekly or monthly results can help the manager plan the next few steps with higher precision than humans can. The only requirement is a regular update with data and an adjustment of the algorithm to the data by an expert. The figure on the left shows the precision of the predictions of an algorithm versus the actual values for an AI prototype.

General Forecasting using Artificial Intelligence

We provide tools for scheduling a call center staff 24 hours, seven days or one year in advance, based on historical data and publicly accessible data (e.g., weather). The workforce management system is built with artificial intelligence tools and its forecasting quality lies above 90%, which beat all published statistical systems currently. The AI is able to automatically derive the demand of employees required per time unit (hours, days, weeks, months) from the past data. Such procedure can be applied for many other types of time series (such as electricity demand, sales).

Human in the loop for explainable Artificial Intelligence

Visualization of High-dimensional Data and Cluster Analysis

High-dimensional data and cluster analysis are two complex topics of our time. New developments especially in the field of medicine require thorough attention. But also, fields like hydrology and many other scientific and industrial fields have the need for new solutions. We recently developed neural networks and approaches using swarm intelligence in order to represent high-dimensional features on a 3D (2,5D) landscape. A topology preserving projection of the high-dimensional data allows a structure preserving representation which can be visualized and is human understandable.
Attached is the so-called cluster analysis which divides leukemia, cancer of the blood cells, into five types using genetic data consisting of 8000 features. Each point represents a patient. The valleys represent groups of similar patients and the mountains the differences between patients. This method, of course, can also be applied to customers. For example, customer discounts can be optimized through a strategic cluster analysis by identifying similarities beforehand. The applications are so widespread that, based on their quarterly financial reports, we have been able to group companies in terms of recommendations  (see ESANN 2019) for purchase or sale of shares.

 

 

 

Explainable Artificial Intelligence for Practical Applications

The explanations and reasons for decisions made by algorithms are firmly demanded in various applications. In case of biomedical applications concerning the health and/or treatment options for patients explainability and comprehensibility of algorithms is especially essential. This calls for systems that produce human-understandable knowledge and decision proposals out of the data such that domain experts can base their decision on these systems. There are different names for such systems: symbolic, knowledge-based, expert systems or recently “explainable AI” systems. Current systems of explainable artificial intelligence systems have either specific assumptions about the data or often do not follow the Grice’s maxims, i.e., the explanations are not meaningful and relevant to domain expert. Using the already researched insights about structures in data and swarm intelligence, an explainable machine learning system (XAI), allowing for user interaction and understandable both upon the Grice’s maxims and from the perspective of domain experts can be developed by our long years of experience in this field.

XAI