Get started
Do the next step and choose your preferred way to use aivis and gain access!
Gain access to aivis
There are multiple ways how to access and use aivis. Just choose the option that suits you best.
Insights App
Use the aivis Insights app to gain access to aivis fast and simple. The intuitive web UI helps you on every step to get you started smooth and efficient.
Project
Your challenges are rather complex and you prefer our aivis experts to take care of it? No problem, just contact us and get to know us in our free initial web meeting.
SDKs (coming soon)
You are used to work with jupyter notebooks or similar? You prefer direct access to aivis from within your code rather than using our web UI? Then get access to our aivis SDKs!
IIoT Platforms
Why import your data if you are already connected to one of our IIoT platform partners. Make use of a direct data connection and start using aivis in no time.
How to use aivis
Independently of which option you choose to gain access to aivis, the following steps are always the same.
Step 2: Start AutoML
Simply define your goal and start aivis. No data science expertise required.
Unsupervised model building and training requiring no model selection or hyperparameter-tuning.
Native handling of up to thousands of parameters and huge data amounts with results usually within few minutes up to 1 hour at max.
Autonomous distinction between relevant and irrelevant data per use case and defined goal.
Selecting goals based on predefined, real-world use cases.
Step 3: Use the Report
Consume the ease-to-understand report, unveiling root causes, correlations & relationships to better understand and improve your processes.
Each report can be quickly understood by a domain expert.
The reports are great for decision-making and deriving direct actions.
Each report provides precise information about the quality and reliability of the results.
Each report follows strict scientific guidelines and requirements.
Step 4: Deploy the Model
Deploying the independent, fast and light-weight model within various environments.
Model accompanied by an report explaining the model.
Continuous benchmarks proof top quality of aivis models.
Ground-breaking methods prevents models from overfitting.
Models stay accurate over a long period of time.
FAQ
You have questions? Maybe our FAQ can clearify things up for you. If not, just use the contact form to ask us directly.
General
You should use aivis if you have any process data that you are allowed to use with external services like aivis. If you think that the data might contain interesting or useful information that can help to improve your processes, you should give it a try.
You should not use aivis if your data contains critical content or personal information, and you do not have the right to use it with data analytics services like aivis.
Yes! aivis was particulary designed for process engineers, since we believe, that those being closed to the data should be able to work with it as well without the need to talk to data scientists first.
Yes! aivis is an excellent tool for data analysts since it requires almost no data preparation compared to other AutoML services and provides incredible speed and quality and exceeding the complexity others can handle by far.
Yes! aivis is an excellent tool for data analysts since it requires almost no data preparation compared to other AutoML services and provides incredible speed and quality and exceeding the complexity others can handle by far.
Data
You use historical data to learn from past behavior and build and train models for e.g. prediction models, virtual sensors, anomaly detectors, or similar.
A trained and ready-to-go model can be deployed to consume live data streams. By doing so, they can live predict or warn if abnormal behavior was detected.
To find hidden dependencies, relations, root causes, etc. you need historic time-series or tabular data. For deployments of models, you need live time-series or tabular data.
Time-series data contains data records that usually come from a sensor or similar and are associated with a timestamp, which indicates when the associated value was measured:
timestamp,Sensor_1
2021/08/31 01:32:31,0.23
2021/08/31 01:32:32,0.22
2021/08/31 01:32:33,0.24
2021/08/31 01:32:34,0.25
…
Tabular data entries are usually created when an event or state is logged. It may contain timestamps or be time-independent. One record can contain countless columns, each associated with this event.
speed,torque,pressure,temperature,variable_1
31,0.22,4.1,293,42
34,0.23,4.2,293,42
38,0.24,4.4,293,43
45,0.26,4.5,293,43
349,0.28,4.7,293,43
…
This depends on the option you choose, how to use aivis. Via an IIoT platform partner, there is no need for an extra data import. The insights app, on the other side, provides a guided wizard to upload your data to an AWS S3 storage connected to the aivis app. Alternatively, you can use your own AWS S3 storage and grant access to aivis.
Yes! Every dataset uploaded to aivis is strictly confidential, not saved permanently, not duplicated, remains entirely under your control, and will be used for your purposes only. If required, an NDA can be set into place.
The amount of data can be anything between a few kilo bytes (kb) and multiple terra bytes (tb).
Deployment
aivis models are independent, light-weight pieces of software, that are executable and deployable independently of aivis.
aivis models have no special requirements at all. We strive, to make aivis model as small and light-weight as possible, so they can be deployed in any environment – on-edge, on-cloud or on-premise.
aivis models are usually deployed by your IT.
Performance
Working fully automated, aivis requires almost no data preparation and provides clear and actionable results with unseen speed and quality.
Compared to others, it is faster, provides better results, requires significantly less data preparation, is easier to use, and always provides clear and explainable results.
The first step aivis performs on the data is always data segmentation: We use Stochastic Differential Geometry to represent the data so that all dependencies and relationships become apparent. Based on that, we use our proprietary universal cluster algorithm to cluster similar geometries. The result is a segmentation of the data where each segment pools similar relationships and importances of the features.
For each segment, we then determine the metric that best describes its dependencies. In a way, this can be interpreted as calculating the hyperparameters, which eliminates the need to manually choosing hyperparameters.
Next, aivis builds and trains a model for each segment using methods from Quantum Field Theory: We interpret the label as the vacuum states of a scalar quantum field and the hyperparameters as vacuum expectation values of background fields (e.g. Higgs field). Finally, we combine the models to a strong ensemble predictor.
In summary, the preceding segmentation brings us two vital benefits: First of all, it is relatively easy to train a model on the segmented data since aivis has already revealed all significant importances and relationships. Secondly, the results are quite transparent and interpretable since it is clear which segment contributes to the current prediction and to which extent it does so.
Other AutoML platforms usually act more like a search engine for suitable approaches to solve a particular problem. They iterate through countless conventional and generic methods and try to pick the most promising one per dataset and problem.
In contrast, aivis already is the perfect match for industrial problems and needs. It provides easily understandable results and clearly identifying root causes. Additionally, superior handling of input data is expressed by the ability to handle the extent of training data that is to be expected in industrial applications, as well as very forgiving requirements in regards to the necessary data structure.
No. aivis is completely independent of any external libraries.
aivis can handle thousands of signals. We have not yet found a limit on the number of signals.