US 2026/0288559 A1 · SYSTEM AND METHOD FOR ADAPTIVE ANOMALY DETECTION IN DATA USING LEARN-ABLE LOSS FUNCTIONS

Abstract

The present invention provides an adaptive anomaly detection system using learn-able loss functions. integrating an autoencoder neural network and a dynamic loss weighting module to improve anomaly classification accuracy and real-time adaptability. The data processing unit collects, normalizes, and transforms input data before feeding it into the autoencoder neural network, which generates a latent representation, reconstructs input data, and computes reconstruction errors. The dynamic loss weighting module assigns adaptive weights to reconstruction errors, dynamically adjusting anomaly sensitivity and determining an optimal threshold via an iterative learning process. An adaptive training module optimizes the system using backpropagation while dynamically updating hyperparameters. A deployment module integrates the trained system into real-time applications.

Claims10

Dependent claims

A system for adaptive anomaly detection using learn-able loss functions, comprising:

a data processing unit, to collect, normalize, and transform an input data received from one or more sources;
an autoencoder neural network, coupled to the data processing unit, to generate a latent representation of the transformed data, reconstruct the input data, compute reconstruction errors, and analyze the reconstruction errors for anomaly detection;
and
a dynamic loss weighting module, coupled to the autoencoder neural network, configured to:
assign adaptive weights to the reconstruction errors based on statistical deviation from normal patterns;
adjust anomaly sensitivity dynamically by modifying loss contributions during training;
and
determine an anomaly threshold automatically based on an iterative learning process;
an adaptive training module, coupled to the dynamic loss weighting module, configured to:
optimize the autoencoder neural network and the dynamic loss weighting module using backpropagation;
and
continuously refine anomaly detection sensitivity based on detected anomalies.

9 more elements

Dependencies are inferred from the claim text.

System and method for adaptive anomaly detection in data using learn-able loss functions

Life of this patent

System and method for adaptive anomaly detection in data using learn-able loss functions

Life of this patent

Abstract

Claims10