4 ms·
Not to sound negative – this is a cool paper – but isn't this in the grey zone of being actual research? It feels like a fairly trivial application of things al
by tw1010 9y ago
Not to sound negative – this is a cool paper – but isn't this in the grey zone of being actual research? It feels like a fairly trivial application of things already known since way back to work in simulation.
- nl 9y agoCitation needed? I'm not aware of any work doing this combination - but I don't follow the simulation space closely. My impression is that NNs haven't been used broadly successfully in simulation until a couple of years ago.
- mike_n 9y agoyes, model-based (ie - learning physics and predicting future states and planning accordingly, rather than just choosing actions that you've learned will give you rewards) RL has been in the literature for many years, see Sutton's DYNA for example. But I think the model-based part is still considered relatively hard in real-world environments due to the large search space of possible actions/consequences. And note that here they didn't actually use the model at runtime, they just used it as an 'expert' (rather than getting a human to provide guidance) to give a model-free agent policy a head start (see DAGGER algorithm) and reduce the number of training samples.
- AstralStorm 9y agoI'd they figure out how to change these models, say, an NFA representation of one, we might have something very interesting... Best of both worlds?