Potential Function
In SMC, a function used to score partial sequences that measures how close the current sequence is to the target constraints. It can participate in particle reweighting without being differentiable or trained.
Timeline (last 90 days)
Near-perfect grammar coverage can be achieved with only a small number of handcrafted potential functions.
A potential function is a scoring function in SMC that defines "how close the current partial sequence is to the goal"; it need not be differentiable or trained, and only needs to be computable in code.
The research team used frontier models for code synthesis to automatically generate new potential functions covering hundreds of target grammar skills, without additional training or supervised data.
The generated potential functions only need to pass simple tests to be plugged into the SMC pipeline, and the whole process involves no retraining of the small model or additional labeled data.
All Facts (4)
Near-perfect grammar coverage can be achieved with only a small number of handcrafted potential functions.
50%UnverifiedA potential function is a scoring function in SMC that defines "how close the current partial sequence is to the goal"; it need not be differentiable or trained, and only needs to be computable in code.
50%UnverifiedThe research team used frontier models for code synthesis to automatically generate new potential functions covering hundreds of target grammar skills, without additional training or supervised data.
50%UnverifiedThe generated potential functions only need to pass simple tests to be plugged into the SMC pipeline, and the whole process involves no retraining of the small model or additional labeled data.
50%