Machine learning (ML) has been flourishing in various fields, including image recognition, natural language processing, and even protein structure analysis. In recent years, it is getting attention in the optoelectronics field. Researchers not only use ML tools to help boost the research of optoelectronic devices but also try to invent new optoelectronic devices to build computers to help the application of ML in real life. In this dissertation, both directions are explored, including using ML to help design high-performing perovskite solar cells (PSCs) and synthesizing new materials to build new optoelectronic synapses for future neuromorphic computers for ML applications. First, ML is used to predict the bandgaps of perovskite materials and performances of PSCs, which shows that ML benefits the research of optoelectronic devices. Several promising findings are discussed based on ML's predictions to help guide the design of high-performing PSCs. Next, new optoelectronic synapses are fabricated, which can act as building blocks for neuromorphic computers. By applying heterogeneous nucleation principles to grow perovskite quantum dots (PQDs) on multi-wall carbon nanotubes (MWCNTs) and Graphene, new materials are synthesized and used to fabricate optoelectronic synapses. The potentiation of the synapses is realized by light pulses, and the depression is accomplished by electrical pulses. Using the properties of the device to do simulations, the ability of the new type of optoelectronic synapses to act as building blocks of optoelectronic neuromorphic computers is demonstrated. Finally, plasmonic OECTs are fabricated using a low-cost method called the nanoimprint method. Using glucose sensing as proof, this new type of OECT devices can significantly enhance the sensitivity of glucose sensing under light illumination. This new type of OECTs could be a new direction for optoelectronic synapses or work as building blocks for the human-machine interface.