It's got a wikipedia page so it must be "serious" :)
From my very quick look it seems like a field that isn't too new (90's). The paper Analysis of Neural Cryptography has Adi Shamir's name on it (the "S" in RSA and the Shamir from Shamir secret sharing), so there has at least been a very reputable cryptographer interested in the idea at one point. Searches on IACR's ePrint archive turn up very little (one hit with "neural" in the Anywhere field).
So, the field seems to be not very well explored and has not generated broad interest. That said, if you are taking the class, have to do some kind of project, and are interested in cryptography, why not find something related? If the project doesn't have to be too earth shattering, I'd try attacking some "classical" ciphers or even see if you can classify classical ciphers (Caesar, vigenere, etc) based on the ciphertext only. Another interesting area would be attacking something like Enigma with neural networks.
Just my 2 cents.