Design novel workflow for ROP attack detection with 0 overhead by only processing incoming data.
Generate ROP gadget chains.
Train CNN for real gadget chain detection.
Final evaluation shows our workflow can achieve over 99.3% detection rate with <0.01% false positive rate.