Machine learning

Recently I came across a few new areas of study that I found interesting at a first glance. Whether they truly are is something I will be assessing with a preliminary study.

Here’s the list:

  • Conformal prediction - this is an interesting mechanism in machine learning that allows to asses and control the uncertainty of a prediction apriori,
  • Relevance based prediction - an approach to structuring regression incorporating the knowledge of and the information content of the outliers for informing fit of the prediction apriori.

The goals of these approaches are similar in nature and have some overlap with the flexibile probabilities approach and that’s what attracted my attention as it can be potentially used in risk control and portfolio position sizing or even as a predictive feature. I will have to investigate it further.

Engineering

On the engineering front, I am learning more about the SwiftUI’s View protocol because I did come across its limitations in a form of inability to demultiplex view type at runtime based on a protocol only. I am not discounting a possible design flaw in my code, but for now, it does feel like a limitation of the framework or in my design.