US 12,750,066 B2 · System and method for network weight compression and intrusion detection

Abstract

A system and method for neural network weight compression with intrusion detection capabilities that optimizes model storage and transmission while providing security. The system analyzes weight characteristics to identify statistical properties within different neural network layers, generates optimized encoding schemes based on the analysis, and creates reference distributions for security verification. The compression process employs a multi-resolution approach that produces a progressive representation with base and enhancement layers, enabling flexible deployment across diverse computing environments. Security markers and statistical fingerprints can be embedded throughout the encoded representation, allowing for detection of unauthorized modifications during transmission or deployment. The system monitors encoded weight streams, measures distribution divergence against reference baselines, and generates alerts when statistical anomalies indicate potential tampering. This approach achieves superior compression ratios while maintaining model performance and providing robust protection against increasingly sophisticated attacks targeting neural network weights.

Claims13

Dependent claims

A system for neural network weight processing, the system comprising:

one or more hardware processors configured for:
receiving a neural network model comprising a plurality of weight tensors;
analyzing weight characteristics to identify statistical properties within the neural network model;
generating one or more encoding schemes based on the analyzed weight characteristics;
creating reference probability distributions for security verification;
encoding the weight tensors using the one or more encoding schemes to produce an encoded weight stream forming a compressed representation of the neural network model;
incorporating security information comprising embedded security markers within the compressed representation;
outputting the compressed neural network model;
verifying cryptographic integrity of the encoded weight stream using the embedded security markers;
computing statistical distributions of the encoded weight stream;
measuring distribution divergence between the computed statistical distributions and the reference probability distributions;
determining whether the measured distribution divergence exceeds a defined threshold indicating potential tampering;
and
generating a security alert when potential tampering is detected.

11 more elements

A method for neural network weight processing, comprising the steps of:

receiving a neural network model comprising a plurality of weight tensors;
analyzing weight characteristics within the neural network model;
generating one or more encoding schemes based on the analyzed weight characteristics;
creating reference probability distributions for security verification;
encoding the weight tensors using the one or more encoding schemes to produce an encoded weight stream forming a compressed representation of the neural network model;
incorporating security information comprising embedded security markers within the compressed representation;
outputting the compressed neural network model;
verifying cryptographic integrity of the encoded weight stream using the embedded security markers;
computing statistical distributions of the encoded weight stream;
measuring distribution divergence between the computed statistical distributions and the reference probability distributions;
determining whether the measured distribution divergence exceeds a defined threshold indicating potential tampering;
and
generating a security alert when potential tampering is detected.

10 more elements

Dependencies are inferred from the claim text.

System and method for network weight compression and intrusion detection

Life of this patent

System and method for network weight compression and intrusion detection

Life of this patent

Abstract

Claims13