3.4. Discussion 3 37 period of rehabilitation that users of visual prostheses typically undergo to improve their ability to use the prosthetic vision (Chen et al., 2009a; Erickson-Davis & Korzybska, 2021; Rassia & Pezaris, 2018). In the light of these of these differences it is important to recognize that our RL benchmark measures an idealized scenario, i.e., the information content that can be retrieved by anoptimal decoder. Still, the evaluation of task-relevant information content in a confined benchmark can provide a useful early-stage indication of the functional visual quality. If not used in isolation of human-centered experiments, the results of our experiments can be regarded as a valuable basic evaluation of the taskbased informativity of the prosthetic percept. This computational evaluation framework can serve as an augmented simulation strategy in parallel to experiments with sighted subjects and prosthesis users. 3.4.3. Conclusion The results of several mobility experiments with virtual implant users were in line with prior behavioural results of sighted subjects. These results indicate that the framework can be used for early-stage prototyping of software and hardware parameters of the implant design. Although the presented experiments are basic examples, and there remain important differences between human subjects and computational agents, our results indicate that the RL-framework can provide a fast and inexpensive addition to existing simulation research and clinical experiments.
RkJQdWJsaXNoZXIy MTk4NDMw