Information Classification Using Local, Global Models

USPTO published patent application US20260099726A1 titled 'Method, Apparatus, Device and Medium for Information Classification.' The application discloses a machine learning technique involving training a local classification model according to a first training objective to reduce association between multiple feature representations, then sending model parameters to construct a global classification model. The approach addresses dimensional collapse by applying decorrelation on feature representations.

For technology companies and AI developers, this patent application represents potential prior art in machine learning classification methods. While patent applications do not immediately restrict development activities, once granted, the patent could affect how companies implement distributed or federated learning classification systems. Organizations developing similar decorrelation-based machine learning approaches should review the claims for potential licensing or design-around considerations.