Comparing expert systems through similarity
A short, informal look at using representational similarity to reason about explainability.
- xai
- decision-support
- research
One thread in my work asks a deceptively simple question: when two models make the same decision, are they reasoning the same way? Similarity-based analysis gives us a handle on that — comparing internal representations rather than just outputs.
The idea in one paragraph
Representational Similarity Analysis (RSA) compares the geometry of how different systems represent the same inputs. Instead of asking “did they agree?”, we ask “did they organize the problem the same way?”. Two models can reach identical predictions through very different internal structure — and that difference is exactly what an explanation should surface.
More details
Want to Know more about RSA? jump into the Visiual Scorolly telling experience I created.