It just occurred to me, that after a couple of years tracking Deep Learning developments, that nobody has even bothered to create a map of what’s going on! So I quickly decided to come up with a Deep Learning roadmap. A word of warning, this is just a partial map and doesn’t cover the latest developments. Many of the ideas I write on this blog isn’t even covered by this map. Anyway, here’s a start of this and hope people start coming out of their labs to further expand on it.
The “Unsupervised Learning” part is from a talk by Russ Salakhutdinov and the “Reinforcement Learning” part is from a talk by Pieter Abbeel.
There are a ton of other ideas that are coming out of the edges as well as the center of this diagram. Also, I did not show the connections between the 3 center concepts. For example, you can use CNNs for Value Iteration and GAN and VAEs use DL networks. It’s a wild world in the Deep Learning space and you just never know how all of this gets re-arranged.
I’ve got a higher level map that starts off with this:
that possibly can stitch everything together in one “grand unified theory”. This is how I think it will all play out:
Unsupervised learning is the the ‘dark matter’ where we need a lot more clarity. It’s my conjecture that meta-learning (with context) is the approach to this. There is some evidence that is developing, but I cannot know for sure. Modular Deep Learning is already in the cards. There is sufficient evidence that this works well. Market Driven Coordination is still early stages, but I believe that the only real way forward is to have diverse architectures working on the same problem and “markets” are a known decentralized way to coordinate actions.
There’s still a lot to be done though and we just in the early stages of Deep Learning evolution:
One additional key issue outside of unsupervised learning is the need to bridge the semantic gap between connectionist and symbolic architectures.
If you think there’s a demand for more clarity in Deep Learning, then support this kind of effort by buying the “Deep Learning Playbook”.