AIScientific equationsStory 07
RISR lets residual errors steer the next equation
What changedRISR gives an equation-searching language model the shape of its remaining numerical error, not just a single fit score. The residual signal guides new formula proposals and corrections on a symbolic-regression benchmark; this is benchmark recovery, not laboratory discovery.
The useful part
Why it matters
Instead of only saying a fit is bad, the system shows where the error lives and uses that shape to propose a correction.
Keep in mind
Good to know
This is benchmark equation recovery, not laboratory discovery.
Evidence