Livoa
Discord
Pricing
English
Sign In
Integrated Hybrid CNN-Transformer Architecture for Behavioral Malware Detection
Input: API Call Sequences
Input: API Call Sequences
2D Transformation Process
Sequence Encoding
Tokenization Map to Embeddings
Matrix Formation
Padding/Trucation Fixed Length
Mapping to 2D Calls, Cols * Eibeddings
2D Feature Map
Hybrid CNN-Transformer Model
CNN Encoder
Long-Range Dependencies, Attention
Classification Head
Feature Aggregation Relevance (Candid/Spatial Featpts)
Fully Connected Layer
Dropout/Regularization Prevent Overfitting
Malware/Activation (REU/GELU)
Explainability Module (XAI)
LIME
Sequence Encoding
Local Surrogate Surrogate Model
Shecay with (Linear/Tree locally)
SHAP
Input Sequence (API Calls)
Shapley Value Computation
(Feature-wise contribution:(+GELU))
SHAP
Input Sequence (API Calls)
Fully Connected Transformer Outputs
Prevent Overfitting
LRP
Input Sequence (API Calls)
Backppsgate Relevance Dismniniity API cation Back/GELU
Softmax Owith Probviisabiity
project
by nani
Use this design
0
0 uses
