[KongchangAI]
ModelS1Mal集成方案

S1Mal

An ensemble malware detection approach based on the System One decision model that feeds atomic behavior question answers and classification scores derived from CLM-8B state embeddings into a gradient-boosted tree for combined judgment.

Timeline (last 90 days)

Nov 17

The study is the first to systematically evaluate the performance of System One decision models in Windows PE malware detection, and proposes an ensemble scheme named S1Mal.

Unverified50%
Nov 17

The S1Mal scheme feeds the atomic answers together with the CLM-8B A2 scores into a gradient-boosted tree.

Unverified50%
Nov 17

The S1Mal ensemble scheme achieves an AUROC of 0.9946.

Unverified50%
Nov 17

At an actual FPR of 1.45% on the test ROC curve, S1Mal detects 93.47% of malware.

Unverified50%

All Facts (4)

Source Articles