[D] What are some ways to improve your *science* for a top-tier ML conference?

There seemed to be some consensus in the responses to yesterday’s post asking about ways to improve your paper for a top-tier ML conference—the primary thing you should do is ensure your research makes a top-tier worthy contribution.

That itself seemed like it might be worthy of further exploration, so I thought I would continue the discussion: What traits characterize a high value research idea or insight? What are good strategies for pursuing one?

conference machinelearning ml science top