Yeah, I agree about smarter parameter search and training. That was a big theme at ICLR this year. Fwiw, here's the paper on RNN architecture generation refer

Yeah, I agree about smarter parameter search and training. That was a big theme at ICLR this year. Fwiw, here's the paper on RNN architecture generation referenced earlier. https://arxiv.org/abs/1712.07316 https://arxiv.org/abs/1712.07316