4 ms·
Ask HN: Can you train multiple neural networks to work together?
I’ve been getting into ML recently, and started studying Neural Networks. The work seems to be mostly about training nets on a specified dataset, at least in the intro lessons I’ve covered. Training on a dataset is interesting, but what idea i wanted to do a more complex task, like classify something in multiple mediums, I.e images and sound.
Would it be possible to string together multiple neural networks trained on different datasets all feeding forward to some master net that operates on the bunch and comes up with a more complex decision?
- db48x 8y agoIt's a computer program; you can implement any features you want.
- thedevindevops 8y ago>multiple neural networks trained on different datasets all feeding forward to some master net What you have described is just a larger neural net or at least is indistinguishable from one.
- hsikka 8y agoThe idea is that these are trained independently with no interconnectivity, a la https://en.m.wikipedia.org/wiki/Modular_neural_network https://en.m.wikipedia.org/wiki/Modular_neural_network
- yesenadam 8y agoA GAN is a fascinating way of using 2 neural nets together, or rather against each other e.g. pix2pix used conditional GANs. https://en.wikipedia.org/wiki/Generative_adversarial_network https://en.wikipedia.org/wiki/Generative_adversarial_network