Enhancing QAOA Ansatz via Multi-Parameterized Layer and Blockwise Optimization

The contribution about QAOA in the conference ICASSP 2026 (see this previous post) has been published here. I’ve also updated the blog, with a brief discussion abut the main ideas behind the paper. In particular, in this paper we focuse on multi-parameterized layers and advanced parameters’ update, in particular on the effect that an extra layer at the end of a standard QAOA circuit has, and on possible (blockwise) optimizations of the parameters of the extended circuit, including ablation studies. Concretely, applied to the MaxCut problem across diverse graph families, the proposed architecture achieves, with an order of magnitude smaller circuit depth, approximation ratios comparable to QAOA, using a single cost–mixer layer, thus reducing gate counts by up to fivefold. Moreover, ablation studies prove that blockwise fine-tuning is crucial to deliver higher-quality solutions at shallower depth, offering a practical, quantum hardware-efficient alternative for signal processing applications. Most of the code is an adaptation of this repository.