What's New in ASCMO-STATIC & ASCMO-DYNAMIC V5.17

General

• The ETAS ASCMO installation now comes with a docker image of ASCMO to run calculations e.g. in the cloud.

  ASCMO Docker Container

• Checkboxes in tables can be copied and pasted (Ctrl+C/V). The state of the checkbox is only changed when the checkbox is clicked and not when the cell is selected.

• Automated Machine Learning now shows the RMSE/Complexity plot with optional log scaling for x and y axis. Log scaling for the y-Axis is default.

  Automated Machine Learning

ASCMO-STATIC

• Automated Machine Learning now can use ASC Compressed models.

  Model Setting: ASC Compressed

• Automated Machine Learning now supports Gaussian Process Linear Extrapolation option.

  Model Setting: ASC Gaussian Process

• ASC Compressed models now can utilize a Matern kernel.

  ASC Compressed Model

• Gaussian Process Models (ASC GP, Compressed, ASC GP-SCS) can now be exported to the Embedded AI Coder.

  Overview: Exports Supported by Model Type

• Gaussian Mixture Model (GMM) has been added as a model type for ASCMO-STATIC. The model represents the training data as a mixture of Gaussian distributions and can be used to calculate a fit score for new data points.

  Gaussian Mixture Model (GMM)

• Anomaly detection outputs can now be added to ASCMO-STATIC projects. The output can be assigned a Gaussian Mixture Model and used to monitor whether the current input data is within the distribution of the training data. . The model can be exported to c-code with the Embedded AI Coder.

  Add Outputs for Anomaly Detection

• List optimizations can now be exported as a job to be calculated on another computer.

  Optimization menu > Global Optimization > Extras menu > List Optimization > Export Current Settings

  Global Optimization

• Calibration menu> Operating Points allows to lock OPs with a formula, e.g., speed > 2000 & Torque > 10.

  Operating Points Manager

  Lock/Unlock OPs with a Condition

• 1/y and 1/sqrt(y) transformations are no longer chosen automatically (AutoML and Automatic Transformation).

ASCMO-DYNAMIC

• Model Monitoring has been added for supported trained Recurrent Neural Network models. Model Monitoring creates a monitoring project that uses a Dynamic Gaussian Mixture Model to estimate whether model predictions can be trusted for the current input data. The model itself together with the monitor model can be exported with the Embedded AI Coder to c-code.

  Open Model Monitoring Project

  Model Configurations: Dynamic Gaussian Mixture Model

  Model Menu (ASCMO-DYNAMIC)

• Added Anomaly Detection Demo to start screen.

ASCMO-STATIC ExpeDes

• Increased the limit to allow generation of 10 million points. This is intended for use cases where a large number of points are initially generated, with many then being removed by a constraint.

ASCMO-DYNAMIC ExpeDes

• Gradients are now less restrictive. Settings which have been rejected in the past are now allowed with a warning and better warning messages with tips.