Reconfigurable Neural Network Circuit for Inline Packet Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing network analysis techniques based on deep learning with neural networks are inadequate for real-time analysis due to their reliance on large packet datasets, leading to delayed intrusion detection and inadequate network protection.
Innovation Solution
A reconfigurable neural network circuit that performs inline network analysis by adaptively processing packets at line rate, utilizing a pipeline architecture with feature computation, scaling, and imputation to determine predicted network characteristics such as anomalies or congestion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional neural networks process large numbers of packets (over 10,000 packets) to determine network characteristics, then measurement precision is improved, but loss of time increases and productivity decreases
Solution Approach 1:
The system performs preliminary feature extraction on packet data before full neural network processing. By extracting relevant features (such as packet size, protocol type, source/destination addresses) in advance, the system prepares processed data that can be quickly analyzed, eliminating the need to process raw packet streams in real-time and reducing detection delays while maintaining accuracy
Solution Approach 2:
The invention extracts only the essential features from packet data that are relevant to intrusion detection, rather than processing entire packets. This extraction approach isolates the critical information needed for analysis, reducing the data volume from over 10,000 packets to a manageable feature set that can be processed rapidly while preserving detection precision
2Reliability
If traditional neural networks analyze large packet datasets, then reliability of network protection is improved, but productivity decreases
Solution Approach 1:
The system segments the network analysis process into distinct stages: packet filtering, feature extraction, and neural network classification. Each segment handles specific tasks with optimized resources, allowing the system to maintain high reliability through comprehensive analysis while achieving high productivity by processing packets through specialized sub-systems that operate in parallel
Solution Approach 2:
The system performs partial processing on all packets (extracting key features) and full processing only on suspicious packets identified by the feature extraction stage. This partial action approach ensures reliable detection of intrusions while maintaining high overall productivity by avoiding exhaustive analysis of every packet
Data Source
AI summary
Described herein are a device and a method for performing a network analysis. In one aspect, the device includes a feature extraction circuit, an input processing circuit, and a reconfigurable neural network circuit. In one aspect, the feature extraction circuit receives a raw packet stream, and obtains temporal statistics of a flow, according to a first packet attribute or a first flow attribute of the raw packet stream. In one aspect, the feature extraction circuit generates a feature data including one or more statistical features based on the temporal statistics of the flow. In one aspect, the input processing circuit scales the feature data to generate an adjusted feature data. In one aspect, the reconfigurable neural network circuit performs computations corresponding to a neural network on the adjusted feature data to determine a predicted network characteristic.