Opportunistic Information Bottleneck for Edge-Assisted Goal-Oriented Communications

The Information Bottleneck (IB) is a key tool for Edge-Assisted Goal-Oriented Communications (GOCs), enabling compact data representation while meeting specific inference performance requirements. However, its practical application is challenging due to the reliance on a functional optimization problem solvable only in limited cases (e.g., Multivariate Normal regression). The recently proposed Opportunistic Information Bottleneck (OIB) addresses this by adapting the Multivariate Gaussian IB solution to general inference tasks based on Deep Neural Networks. This work demonstrates OIB's advantages in optimizing resource allocation for Edge-Assisted GOCs, showing that, compared to competitive approaches, it can dynamically minimize energy consumption in mobile User Equipments (UEs) while meeting learning accuracy and latency constraints.