Powerful. Pragmatic. Explainable.
aivis works close to real-world problems, thus minimizing the effort from data import to result dramatically
Industrial data-driven challenges are demanding
Thousands of parameters, highly dynamic environments, and the highest standards demand the maximum performance, flexibility, and usability from an machine learning solution. Most conventional methods are just not up to the task: They are either too slow, too expensive, too complicated to use, or just don’t deliver the required quality and therefore limit the industrial success.
aivis is different
Developed for demanding industrial requirements, aivis delivers great results incredibly fast, can cope with almost any data complexity, and offers explainability. It helps industrial applications to get the most out of their data. Compared to other AutoML solutions, aivis is faster, provides better results, and requires less training data. It easily handles thousands of parameters and offers full explainability.
aivis is supported by strong and passionate partners who bring aivis into various industrial applications and enable good and data-driven decisions.
aivis opens up a whole new access to your data. Being close to industrial applications, it enables you to get the most out of your data fast and efficient. No excessive data preparation and no data science expertise needed.
Minimal data preparation
The only data preparation necessary is the data conversion into the aivis CSV format. Everything else, aivis does for you: aivis works on any amount of unprocessed, uncleaned, and unsynchronized raw data. Also, no prior exclusion of supposedly irrelevant signals and data or any insertion of pre-knowledge is required.
Actionable and explainable results
aivis always presents fully transparent results by providing a report that shows which signals or parameters contribute to which extent. Therefore, aivis’ results are easily understandable and interpretable to the data owner and domain experts.
Focus on the right questions
Once you have converted your data, you are ready to go. An educated guess is already enough to start the first calculation. Fast and easy to use, aivis allows you to explore your data’s information and the possibilities that come with it in an almost playful manner. This way, you can fully concentrate on asking the right questions without being distracted by how to answer them.
Simple to use
Working with aivis requires no data science expertise. In particular, you will not have to choose between available mathematical approaches regarding the training or model building. Instead, all you need is a general understanding of your processes and your data. An intuitive and straightforward UI helps you through every step from data upload to starting aivis calculations.
Quickly first results
With aivis Insights, you achieve first valuable results in the quickest possible way. With very little initial effort, you can let aivis create an Insights report giving you a purely data-driven answer to your question. In many cases, this already deepens the understanding of your processes significantly and enables spontaneous efficiency improvements.
aivis combines the best of different worlds being fast, flexible, stable, and transparent at the same time.
aivis can easily cope with thousands of parameters. This makes it especially useful in very complex environments where human comprehension quickly reaches its limits.
Before starting with resource-intensive operations, aivis first rotates the data in the right direction by data segmentation. This makes all following operations much easier and faster.
A data segmentation based on stochastic differential geometry creates data segments with similar properties and known relevant influences. This allows aivis only to learn relevant behavior and natively prevents overfitting.
No hyperparameter tuning
The theory behind aivis allows a geometrical interpretation of hyperparameters enabling their direct determination. Therefore, aivis eliminates the need for choosing and tuning hyperparameters.
Unsupervised model building
aivis builds trained models without any manual intervention. In particular, it requires no selection of mathematical approaches or filtering of irrelevant data.
The preceding data segmentation reveals the contribution of each signal or parameter, allowing excellent transparency and explainability of each result.
The initial clustering of the data into segments with similar behavior greatly simplifies the model building, enabling excellent quality.
Since aivis models natively include the system’s different types of behavior into the training, they are accurate over a long time without any adaptions.
aivis works close to real-world problems, thus minimizing the struggle to translate your question back and forth into something that mathematical approaches can solve.
The preceding data segmentation allows for independent model building for each segment. This is a very data-efficient approach that only requires little data to develop and train a complete model.
aivis is an industry-oriented and industry-agnostic machine learning technology that builds on the entirely new mathematical theory of Geometrical Kernel Machines: Kernel machines extended by methods from Stochastic Differential Geometry and Quantum Field Theory.
The first step aivis always performs on the data is a data segmentation: Stochastic Differential Geometry is used to find a representation of the data that reveals all dependencies and relationships. Based on that, a proprietary universal cluster algorithm clusters similar geometries. The result is a segmentation of the data where each segment pools similar relationships and importances of the features.
Defining the metric
Next, for each segment the metric is determined that best describes its dependencies. In a way, this can be interpreted as calculating the hyperparameters, which eliminates the need for choosing and tuning hyperparameters.
Also, aivis builds and trains a model for each segment. This is accomplished by using methods from Quantum Field Theory: The labels are interpreted as vacuum states of a scalar quantum field whereas the hyperparameters are interpreted as vacuum expectation values of background fields (e.g. Higgs field). Finally, the models are combined to a strong ensemble predictor.
In summary, the preceding segmentation brings two vital benefits: First of all, it is relatively easy to train a model on the segmented data since aivis initially revealed all significant importances and relationships. Second, the results are transparent and interpretable since it is clear which segment contributes to the current prediction to which amount.
Due to it’s unique nature and mathematical approach, aivis provides great features and benefits when compared to other AutoML solutions and providers.
For a medium sized problem, aivis trains a model within a few minutes, where other approaches require at least a few hours and more. Even for complex approaches, aivis rarely needs more than one hour.
Less data required
Neural networks in particular need very much training data in order to achieve an acceptable result. For aivis, a small amount of data is already sufficient.
Durable and reliable
Conventional models tend to lose their accuracy on slight changes in the system’s behavior, requiring constant adaptions. aivis models have already learned the underlying behavioral changes resulting in better accuracy over a much longer time.
Easier to use
aivis requires no data science expertise. All steps usually required in the machine learning process (feature engineering, algorithm selection and hyperparameter tuning, etc.) are taken care of.
Only to know what is happening is not enough. Understanding why it is happening is of equal importance. This fundamental realization forms the basis for every interaction with aivis making each result transparent and explainable.
Closer to real-world problems
A large part of data analytics is usually the translation of the problem into something that can be solved by a theoretical Machine Learning approach – and the retranslation. This effort can be avoided entirely when using aivis.
Conventional approaches can handle up to a few hundred parameters and require a large effort to do so. aivis can easily handle thousands of parameters without any additional effort.
Many approaches, especially neural networks, tend to overfit since they struggle to distinguish between noise and data. aivis uses data segmentation to avoid this problem from the very beginning.
Greater data tolerance
aivis can cope with many of the commonly known data collection problems. In particular, it accepts data voids and non-equidistant timestamps on time-series data without problems.
Due to its data efficiency and high speed, aivis is particularly well suited for huge datasets as they can be expected in industrial environments.