Machine Learning Researchers Suck at Math

In comparison to e.g. theoretical physicist.

  1. They suck at defining terms

  2. When they do, they generally try to reinvent the wheel.
    1. e.g. computational graphs/networks. They have no precise, useful definition

  3. They are vague
    1. E.g. they don’t specify the distributions they work with

    2. They misuse/misinterpret well-defined terms such as bias

  4. Their notation is awkward and outdated.

  5. They’re trend oriented, failing to recognize ML is no longer in the explore phase of it’s lifecycle. The foundations need to be refactored.

    • mainly due to laziness.

  6. They are not up-to-date on contemporary methods.

    • e.g. late adoption of the totally obvious fact that NNs live in function spaces