Change Detection

On edge machine learning we are constantly dealing with non-stationary environments. These non-stationary environments can range from changes in temperature or humidity in sensors to changes in the way a machine, such as a gas turbine, is operating.

Time dependent data often experience variations on trends. Ekkono’s change detector can detect those variations so that the model can either adapt to the new trends or signal an alarm when a change occurs. Through our experience with different customers in many different industries, we have discovered that detecting a malfunction or a change in the behaviour of the device ahead of time is time and money-saving. 

At Ekkono, we have built a change detector that can monitor data distributions in real-time, and signal when a change is happening. The uniqueness of the change detector is that it can be calibrated and run both incrementally in a microcontroller and batch-trained on a standard server. We provide two types of change detectors. The first type is one dimensional and runs directly over the sensor value. The second type is multidimensional, and runs on top of a machine learning model, being able to detect concept drift on multivariate data. 

Related content

More Research from Ekkono

Federated Learning

Federated Learning

Federated Learning is a machine learning technique to train a global model from different clients or devices while keeping the data decentralized.  A central server is in charge of coordinating the transfer of the models’ data between the server and the clients,...

Incremental Learning

Incremental Learning

Incremental Learning is a key feature when dealing with tiny and embedded devices. Many have claimed that only inference can be performed on embedded devices. At Ekkono, we have built a library that can train and run machine learning algorithms not only on big...

Conformal Prediction

Conformal Prediction

Many machine learning systems involve making predictions, through estimating the value of a dependent variable with an a priori unknown ground truth. Verification of such predictive systems, in particular when applied in high-risk applications, is crucial. Traditional...