YEN-HUAN LI2021-05-052021-05-05202015206149https://www.scopus.com/inward/record.url?eid=2-s2.0-85089211934&partnerID=40&md5=23de0c61a3a5e7f006dce210c0b52353https://scholars.lib.ntu.edu.tw/handle/123456789/559329The number of measurement outcomes in positron emission tomography (PET) is typically large, rendering signal reconstruction computationally expensive. We propose an online algorithm to address this computational issue. The per-iteration computational complexity of the proposed algorithm is independent of the number of measurement outcomes and linear√ in the signal dimension. The algorithm has a rigorous Oleft( {1/sqrt k } right) convergence rate guarantee, where k denotes the iteration counter. Numerical experiments on synthetic data-sets show that the algorithm can be significantly faster than expectation maximization and stochastic primal-dual hybrid gradient method. The proposed algorithm is based on an equivalent stochastic optimization formulation, the Soft-Bayes algorithm for online portfolio selection, and standard online-to-batch conversion. © 2020 IEEE.online portfolio selection; online-to-batch conversion; Positron emission tomography; Soft-Bayes; stochastic optimization[SDGs]SDG8Audio signal processing; Gradient methods; Maximum principle; Numerical methods; Optimization; Positrons; Signal reconstruction; Speech communication; Computational issues; Expectation - maximizations; Iteration counters; Numerical experiments; On-line algorithms; Positron emission tomography (PET); Stochastic optimizations; Synthetic datasets; Positron emission tomographyconference paper10.1109/ICASSP40776.2020.90532302-s2.0-85089211934