A Human-in-the-Loop Based ML Framework to Estimate User's QoE on Cloud Gaming Using Active Learning

Autor(es)

Carvalho, Marcos ; Soares, Daniel ; Macedo, Daniel F.

• Tipo de documento: Artigo Científico
• Fonte: 2024 JOINT EUROPEAN CONFERENCE ON NETWORKS AND COMMUNICATIONS & 6G SUMMIT, EUCNC/6G SUMMIT 2024; v. N/A, p. 6-pg., 2024-01-01.

Resumo

Data collection for Quality of Experience (QoE) models can be quite costly in terms of time and effort. Data-driven techniques such as supervised machine learning demand a large number of labeled instances to generate precise models. On the other hand, frequently nagging users to provide inputs is not an option, since it reduces the user's engagement with a service. This paper explores Active Learning (AL) to create a user (human) in the loop learning process: learning occurs in steps, where in each step the AL technique suggests new instances to be labeled by humans. We evaluate our proposal in an emulated network, where users label QoE for a cloud gaming application, followed by a comparative analysis between random sampling (non-AL strategy) and Batch Mode Expected Model Change Maximization (BEMCM) AL strategy. BEMCM reduces the number of instances required to achieve a certain level of precision by 38.4%, while the model error decreases 37.1%. (AU)

• Processo FAPESP: 18/23097-3 - SFI2: fatiamento de infraestruturas de internet do futuro
• Beneficiário: Tereza Cristina Melo de Brito Carvalho
• Modalidade de apoio: Auxílio à Pesquisa - Temático
• Processo FAPESP: 20/05182-3 - PORVIR-5G: programabilidade, orquestração e virtualização em redes 5G
• Beneficiário: José Marcos Silva Nogueira
• Modalidade de apoio: Auxílio à Pesquisa - Temático