Gan & Isola
A pair of AI researchers who proposed the neural thicket hypothesis and the RandOpt method, and who study the discovery of task experts within the weight space of pretrained models.
Timeline (last 90 days)
Gan & Isola (2026) propose the neural thicket hypothesis: a pretrained model lies in a dense set of task experts, and simple Gaussian weight perturbations can reveal these experts.
The RandOpt(N, K) method proposed by Gan & Isola takes a majority vote over multiple perturbed versions to improve performance.
The main disagreement between Gan & Isola's original work and the re-examination study lies in whether the improvement from perturbations generalizes and in the effectiveness of the majority-voting strategy.
All Facts (3)
Gan & Isola (2026) propose the neural thicket hypothesis: a pretrained model lies in a dense set of task experts, and simple Gaussian weight perturbations can reveal these experts.
50%UnverifiedThe RandOpt(N, K) method proposed by Gan & Isola takes a majority vote over multiple perturbed versions to improve performance.
50%UnverifiedThe main disagreement between Gan & Isola's original work and the re-examination study lies in whether the improvement from perturbations generalizes and in the effectiveness of the majority-voting strategy.
50%