RandOpt
A model adaptation method based on random weight perturbations. It discovers task experts by adding Gaussian noise to pretrained weights and applying majority voting.
Timeline (last 90 days)
The RandOpt(N, K) method proposed by Gan & Isola takes a majority vote over multiple perturbed versions to improve performance.
The majority-voting strategy of RandOpt(N, K) was shown to often be matched or surpassed by a simple baseline: sampling K samples directly from the base model and taking a majority vote.
All Facts (2)
The RandOpt(N, K) method proposed by Gan & Isola takes a majority vote over multiple perturbed versions to improve performance.
50%UnverifiedThe majority-voting strategy of RandOpt(N, K) was shown to often be matched or surpassed by a simple baseline: sampling K samples directly from the base model and taking a majority vote.
50%