MUMBAI, India, Oct. 5 -- Intellectual Property India has published a patent application (202641115415 A) filed by Vellore Institute Of Technology on September 26, 2026, for Stateful Automated Preprocessing With Adaptive Categorical Encoding And Zero-Skew Online Inference For Machine Learning Pipelines.
Inventors include Anindita Kundu; and Shashi Bhushan Vijay.
The application for the patent was published on October 02, 2026, under issue no. 40/2026.
Abstract: ABSTRACT STATEFUL AUTOMATED PREPROCESSING WITH ADAPTIVE CATEGORICAL ENCODING AND ZERO-SKEW ONLINE INFERENCE FOR MACHINE LEARNING PIPELINES A data processing system (100) for automated machine learning comprises a stateful preprocessing apparatus (126) configured to eliminate training-serving skew and prevent runtime pipeline failures caused by categorical schema drift. The apparatus comprises a schema detection module (202) configured to classify columns as numerical, categorical, or datetime, a transformer fitting engine (204) configured to fit statistical parameters strictly on a training partition, an adaptive encoding module (206) configured to apply one-hot encoding or frequency encoding based on cardinality thresholds, and a serialization engine (208) configured to persist the complete fitted state as a single atomic byte-stream artifact (214). An asynchronous online serving engine (180) dynamically loads the serialized artifact and executes transformations guaranteeing mathematical parity between offline training and online inference. A drift handling module (210) detects unseen categorical values and automatically reverts encoding strategies in real-time without crashing. (Fig. 1)
Disclaimer: Curated by HT Syndication.
