Computer Vision News - October 2022

25 Ashley Bruce electrodes, there are many possibilities for arranging them. It is not usually technically feasible to find a solution. “Ashley approached this as a greedy optimizationproblem, whereoneelectrode is placed after another,” Michael explains. “We used a computational model of bionic vision to help predict what the vision would look like for a given electrode placement. By iterating over that, Ashley found a mathematically proven optimal solution.” Greedy optimization is just that – a greedy approach to optimization. Each electrode is taken one at a time, its best placement is found, and it stays there. Then the same is done with the second electrode, taking a greedy approach to placing the electrodes in their optimal positions on the implant. “We don’t place the electrode if it doesn’t improve the final result,” Ashley adds. “When you start getting higher ρ and λ values, or when you get too high in the number of electrodes, each electrode might not add as much to the next iteration. Therefore, we considered a small upper value to ensure we were still increasing our results; otherwise, that electrode was not helpful.” Before settling on the greedy approach, Ashley and Michael went through several other options. They started exploring biological methods, including particle swarm optimization , which looked promising. “We spent a lot of time on that, but there was no guarantee that the minimum we found was optimal, and it took way too long for something that wasn’t even a fully optimal solution,” Ashley recalls. “When that didn’t give us the results we wanted, it