Illustration for the KIAS Webzine HORIZON:
“Information Geometry and Machine Learning [1]: Distances Between Probabilistic Models”
This piece visualizes a conceptual space where a right triangle—formed by coordinates derived from dual geometric perspectives—does not exist in reality, but rather emerges from inference based on the horizontal axis of the primal space and the vertical axis of the dual space.
To emphasize this duality, I used a GIF animation, highlighting the inferred flat space with the x-axis from the primal space and the y-axis from the dual space. The visual intentionally ignores perspective, keeping horizontal intervals uniform to reinforce the continuity of the x-axis in the primal space.
Rather than illustrating multiple cases, the focus was on how a right triangle can be derived using these horizontal and vertical axes. To enhance clarity and visual emphasis:
Sigma² was explicitly written as a squared term.
The curves were exaggerated to make the geometric relationship more pronounced.
Category
Visual Graphic, Illustration
Belong
Korea Institute for Advanced Study
Date
2022



