Axiomatic data science

Mathematician by background, I love simplicity and applications to the real world. Most of my work experience is in the data science field. Starting from my learnings, I would like to share my perspective about the thought process, applicable to both data science and other fields.

We all have cognitive biases. For instance, we often try to confirm our beliefs by seeking supportive information. This tendency, known as “confirmation bias”, prevents us from re-examining what is true. Also, facing different perspectives, this behavior implies a close mind, which can easily generate conflicts.

As a result of biases, we often over-complicate our thought process. While this is probably unavoidable, there are ways of mitigating it. What if, instead, we could have a though process so simple to be hardly disputable? This is especially relevant in the field of data science, involving decisions based on complex data.

Mathematics could help to simplify our thought process. The field is not always considered, primarily due to its reputation of being though, since it requires unnatural abstract thinking. What if we could find a compromise, applying maths to the real world while using its core principles?

The basics of maths are axioms, that could be considered base hypotheses. Then, using a “proof” thought process, theorems can be derived from the axioms.

How does this apply to the real world? First, there are several similarities with data science. Perhaps the assumptions, real-world axioms, are often mixed with common sense, and the thought process aims to be effective rather than mathematically orthodox. Still, an approach inspired by mathematical axioms helped me to shape some clear thinking.

An iterative path towards an axiom-based approach could be to Formulate assumptions and hypotheses to prove Agree on the assumptions Prove the hypotheses with the help of data