Systems and methods for statistics based and topology based interoperable AI/ML model monitoring

A method implemented by a wireless transmit/receive unit (WTRU), the method comprising:

• receiving configuration information, the configuration information comprising an indication of one or more topology methods used to evaluate a topology of a dataset, a metric associated with at least one topology method, and an identification associated with each of the one or more topology methods;
• selecting a topology method from the one or more topology methods indicated by the configuration information;
• determining topological features of the dataset in an input data space using the selected topology method, and determine topological features of the dataset in a latent space using the selected topology method;
• comparing the topological features of the dataset in the input space to the topological features of the dataset in the latent space to determine whether a WTRU-side model satisfies a metric associated with the selected topology method;
• and
• sending an interoperability report, wherein the interoperability report comprises an identification associated with the selected topology method and an indication of whether the WTRU-side model satisfies the metric associated with the selected topology method;
• wherein a determination of whether the WTRU-side model satisfies the metric associated with the selected topology method comprises training the WTRU-side model such that the topological features of the dataset in the input space and the topological features of the dataset in the latent space satisfy the metric associated with the at least one topology method.

A wireless transmit/receive unit (WTRU) comprising:

• a processor configured to:
• train a WTRU-side model of a two-sided model for channel state information (CSI) compression, wherein the WTRU-side model is trained independently from a network-side model to satisfy topology-related criteria;
• determine a data topology for training the WTRU-side model using a single method Mi, wherein the method Mi is configured to:
• identify the topology of received input data;
• extract topological features from the received input data;
• map the input data prior to evaluating connectivity in the data topology;
• evaluate connected components of the input data using a distance measure and apply a weighting factor;
• enforce a pre-defined topology in a latent space of the WTRU-side model to match a topology defined by at least one of the network-side model or by a standard;
• and
• transmit an indication related to topological compatibility testing of the WTRU-side model with respect to the network-side model, wherein the indication includes an identity of the method used for compatibility testing and topological features of a test data set in the latent space.