Optimal strategic quantizer design via dynamic programming
Anju Anand
Abstract:
This research topic is on a quantization setting with misaligned objectives for the encoder and the decoder. We motivate the problem using a toy example to show the intricacies of the strategic quantization problem. Using the example, we show that iteratively optimizing decoder and encoder mappings may not converge to a local optimum. We then propose a dynamic programming based optimal optimization method inspired by early works in quantization theory. We extend this approach to variable rate (entropy coded) quantization and we present numerical results of the proposed algorithms. Further, we do complexity reduction of the algorithm with an assumption on the decoder’s distortion function. A conference paper on this work was accepted at IEEE’s Data Compression Conference 2022, titled “Optimal strategic quantizer design via dynamic programming”.