Research Based Approaches

After a few years as a researcher you will find your idealism waning and your general energy in life diminishing. You can revive your spirit by entering the race to be come a big-shot.

The best way is to find a gold mine of paper generation and exploit it as fast as possible. Once you have sucked the mine dry, articulate publicly a high principle that „it is not sufficient to merely <insert your method> to become a good researcher“.  You can use this principle to good effect when you become an editor.

Classical examples that have been used successfully in the past are:

  1. Apply neural networks to everything in the known universe (a more up-to-date name for this activity is „deep learning„).
  2. Bound the VC dimension of every known function class.
  3. Kernelise any known linear statistical estimator. (There are actually still a few nuggets left in this mine).
  4. Apply variational inference or expectation propagation to every possible probabilistic model in the exponential family.
  5. Take every problem expressible in terms of a linear program, and extend it to semi-definite programming, second order cone programming, etc.
  6. Find a closely related but little known field (such as economics or game theory), move in and make a killing.
  7. Apply support vector machines to everything in the known universe.
  8. Apply Bayes‘ rule to everything in the known universe.
  9. For theoreticians, re-do every classical piece of analysis with the latest model of machinery (in 2002 it was Rademacher averages. update: still true in 2007).