MERLOT: A Distilled LLM-based Mixture-of-Experts Framework for Scalable Encrypted Traffic Classification
This addresses scalable encrypted traffic classification for network security applications, representing an incremental improvement through a novel hybrid architecture.
The paper tackles encrypted traffic classification by proposing MERLOT, a mixture-of-experts framework using distilled large language models, achieving superior or competitive performance on 10 datasets while significantly reducing computational costs.
We present MERLOT, a scalable mixture-of-expert (MoE) based refinement of distilled large language model optimized for encrypted traffic classification. By applying model distillation techniques in a teacher-student paradigm, compact models derived from GPT-2-base retain high classification accuracy while minimizing computational costs. These models function as specialized experts in an MoE architecture, dynamically assigned via a gating network. Unlike generation-based methods, our approach directly classifies encrypted traffic using the final decoder token with contextual feature embedding as input. Experiments on 10 datasets show superior or competitive performance over the state-of-the-art models while significantly reducing resource demands, underscoring its effectiveness and robustness.