method: DeepTrust2025-04-04
Authors: Daniel Pulido, Daniel Gibert
Affiliation: Artificial Intelligence Research Insititute (IIIA-CSIC)
Email: danielpulidocortazar@gmail.com
Description: DeepTrust is compound of two Multilayer Perceptron models which leverage adversarial training and fuzzy labelling distilled by a Random Forest Model. Both models are configured in a cascade fashion governed by an Isolation Random Forest that decides which model to use based on anomaly detection at the embedding level of the models.
To download the pretrained model: https://drive.google.com/drive/folders/1MzppCM60UBRjTAZ5jBm32Pfo0if21YmX?usp=sharing
method: Baseline - SecSVM2024-09-04
Authors: .
Description: Reproduction of the detector from Demontis et al. “Yes, machine learning can be more secure! a case study on android malware detection.” IEEE TDSC 2017.
method: Baseline - Multilayer Perceptron2025-03-18
| Detection Rate | False Positive Rate | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Date | Method | 100 manipulated features | 50 manipulated features | 25 manipulated features | Clean data | False Positive Rate | |||
| 2025-04-04 | DeepTrust | 19.92% | 33.84% | 43.28% | 78.32% | 0.98% | |||
| 2024-09-04 | Baseline - SecSVM | 5.44% | 10.64% | 14.96% | 75.36% | 0.44% | |||
| 2025-03-18 | Baseline - Multilayer Perceptron | 2.64% | 4.00% | 6.80% | 68.48% | 0.32% | |||
| 2025-03-10 | SVM-CB (b=0.2, n=100) | 2.32% | 3.28% | 4.16% | 75.36% | 0.82% | |||
| 2025-03-10 | SVM-CB (b=0.8, n=100) | 2.32% | 3.28% | 4.16% | 75.36% | 0.82% | |||
| 2025-02-20 | Continual-Positive Congruent Training | 0.24% | 1.20% | 4.56% | 59.60% | 0.40% | |||
| 2024-09-04 | Baseline - DREBIN | 0.00% | 0.08% | 0.48% | 77.28% | 0.36% | |||