In AI in production, a crucial challenge emerges: non-stationarity of data, which compromise the effectiveness and reliability of algorithms.
ML cube Platform addresses it by providing by proactively detecting changes in data and operating conditions that could degrade model accuracy and potentially cause production issues.
It helps AI developers to improve the reliability of their AI systems deployed on Siemens Industrial Edge.
Continuously tracks model performance, inputs, image metrics, and outputs in real time, generating alerts for anomalies or degradation to ensure reliability and quick response.
Identifies and explains why drift occurs by highlighting which image features or data segments changed, helping teams quickly understand and address root causes.
Groups data into evolving clusters based on patterns, enabling detection of shifts in data segments and uncovering hidden changes affecting model performance.
The ML cube Platform monitored the input images and correctly identified an unexpected seasonality drift, which was causing a model performance deterioration. The dynamic clustering and drift explainability modules allowed the user to understand the source of the drift.
The ML cube Platform monitored of the input images and the label prediction, correctly identifying drift due to changes in lighting condition of the environment. Segmented monitoring allowed to identify issues in specific production lines helping understand where errors are most likely.