bothconditionalandmargi...
RealisticEvaluationofTransductive Few-ShotLearning-SupplementaryMaterial
In the main tables of the paper, we did not include the performances ofα-TIM in the standard balanced setting. Now,wewanttoemphasize thatthemodel differences mentioned abovecanbestraightforwardly applied to ourα-TIM (and likely the other methods) in order to boost the results at the cost of a significant increase of compute requirement. We provide the derivation of Eq. (4) in the main paper, which linksα-entropyHα(p) to the αdivergence: The study in [4] examined the effect of class imbalance on the support set after defining several processes togenerate class-imbalanced support sets.