Maximum Entropy Approach to Massive Graph Spectrum Learning with Applications

We propose an alternative maximum entropy approach to learning the spectra of massive graphs. In contrast state-of-the-art Lanczos algorithm for spectral density estimation and applications thereof, our does not require kernel smoothing. As choice function associated bandwidth heavily affect resulting output, mitigates these issues. Furthermore, we prove that smoothing biases moments density. Our can be seen as information-theoretically optimal a smooth graph density, which fully respects moment information. The proposed method has computational cost linear in number edges, hence applied even large networks with millions nodes. showcase on problems similarity counting cluster graph, where outperforms existing iterative approaches both synthetic real-world