… the author
Samuel Loomis currently works at Syngenta Seeds as a complex systems modeler and holds a PhD. in physics with his expertise in the physics of information. He applies information-theory-based data science combined with physics-inspired models to better understand the often messy, nonlinear and downright weird data collected from complex systems of all shapes and sizes, including genomics, supply chains, and agricultural systems. On the side he and his wife manage their own complex system: Shangri-Haw Farm, a duck and goose farm oriented around regenerative grazing and conservation breeding (for more details check out his farm blog and farm site).
… the blog
Sometimes when I’m deep in model-world I feel like Luke Skywalker, who learned to wield a lightsaber only to lose his hand. Oftentimes we approach data with a mindset fixated on finding the “signal” and cutting away the “noise”. But when we make too-harsh modeling assumptions, hacking away at the data without regard for its complexity, we handicap our own ability to understand what’s going on.
This is why I love information theory: its simplicity. It needs to be simple: every complex system is unique and domain-specific, so only a simple, unassuming framework could prove useful across the board. Information-theoretic tools help us to quantify what can and cannot be understood from the data in a model-agnostic way. It inverts our normal perspective of only extracting the signal and tossing the noise, instead modeling the noise and signal together as a whole. This allows information theory to identify noisy, nonlinear, and multivariate relationships where parametric methods fail.
So this blog is, by and large, about information theory, the way of thinking it inspires and all the things you might do with it. I’ll generally constrain myself to three kinds of posts:
- Diving into information theory and all its characters, as well as related areas and tools;
- Deep-dive demos showing, with toy models and open datasets, how information theory can be applied to practical problems;
- Just-for-fun explorations of creative and unorthodox ways to get the data to talk to you.