Hello. You could check recent publications of Rosemberg DB studying zebrafish. Good luck

It depends on how you choose to define memory, or what memory module you are investigating.

For instance, Saucier and colleagues investigated location memory in rats. They did this by familiarizing the rats with objects placed in an arena. Later, the objects were moved, and "memory" was measured by time spent investigating moved objects. I see no reason why this could not be replicated with fish or aquatic animals.

As another example, B'shary and colleagues have done much research on cleaner fish. Among their studies were the ability to distinguish between visiting fish and residential fish. Making this distinction means the fish have some sort of memory for what is local and not local.

Yet another example, from Mattioli, is that goldfish can learn to discriminate two different compartments when searching for food, meaning that they can remember which compartment is the food source. Clearly goldfish remember more than three seconds.

It really depends on how you choose to define memory. The examples I listed are just a few out of many ways this can be done, but hopefully this might help inspire you.

I have linked to the articles I mentioned below.

Article Sex differences in object location memory and spatial naviga...

Article Adult Cleaner Wrasse Outperform Capuchin Monkeys, Chimpanzee...

Article Substance P facilitates memory in goldfish in an appetitivel...

I would suggest you works by Culum Brown from Macquerie University, Australia.

https://www.researchgate.net/profile/Culum_Brown/publications

You could always do whatever you want and validate it. But it is important to define operationally what memory would be. Also generally speaking it is a good idea to choose an instrumental behavior very easy to record in terms of frequency or latency. Also you could choose a good escape behavior (which means a behavior that interrupts a particular aversive stimulus.