A deep neural network hierarchical structure analysis method based on side channel timing characteristics
By acquiring the timing signals of the graphics card side channel and constructing context-dependent representations, the problem of parsing the hierarchical structure of deep neural networks under complex network structures and noise interference is solved, achieving stable and reliable hierarchical structure recognition and parameter acquisition.
Patent Information
- Authority / Receiving Office
- CN Β· China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANGHAI UNIVERSITY OF FINANCE AND ECONOMICS
- Filing Date
- 2026-01-28
- Publication Date
- 2026-06-16
AI Technical Summary
Existing technologies struggle to accurately analyze the hierarchical structure of deep neural networks from continuous side-channel timing features under complex network structures and diverse conditions, and their robustness is insufficient, especially when subjected to noise interference and changes in execution modes, resulting in unstable analysis results.
By acquiring the side-channel timing signals during the deep neural network inference process executed by the graphics card, a timing representation containing context-related information is constructed. The neural network is used to parse the hierarchical structure, and the structural parameter information is determined by combining the execution interval and the hierarchical type.
It achieves stable parsing of the hierarchical structure of deep neural networks under complex network structures and noise interference conditions, improves the completeness and reliability of the parsing results, expands the scope of application, reduces the dependence on additional prior information, and enhances robustness.
Smart Images

Figure CN122221918A_ABST