How much meta-learning is in image-to-image translation?

Maximilian Eißler, Thomas Goerttler, Klaus Obermayer

Keywords: transfer learning, image-to-image translation, generative invariance transfer, meta-learning

Abstract:

At the last ICLR conference, Zhou et al. [2022] presented work showing that CNNs do not transfer information between classes of a classification task.

Allan Zhou, Fahim Tajwar, Alexander Robey, Tom Knowles, George J. Pappas, Hamed Hassani, Chelsea Finn [ICLR, 2022] Do Deep Networks Transfer Invariances Across Classes?

Here is a quick summary of their findings: If we train a Convolutional Neural Net (CNN) to classify fruit on a set of randomly brightened and darkened images of apples and oranges, it will learn to ignore the scene’s brightness. We say that the CNN learned that classification is invariant to the nuisance transformation of randomly changing the brightness of an image. We now add a set of plums to the training data, but fewer examples of them than we have apples and oranges. However, we keep using the same random transformations. The training set thus becomes class-imbalanced.
We might expect a sophisticated learner to look at the entire dataset, recognize the random brightness modifications across all types of fruit and henceforth ignore brightness when making predictions. If this applied to our fruit experiment, the CNN would be similarly good at ignoring lighting variations on all types of fruit. Furthermore, we would expect the CNN to become more competent at ignoring lighting variations in proportion to the total amount of images, irrespective of which fruit they depict.
Zhou et al. [2022] show that a CNN does not behave like this: When using a CNN on a class-imbalanced classification task with random nuisance transformations, the CNNs invariance to the transformation is proportional to the size of the training set for each class. This finding suggests CNNs don’t transfer invariance between classes when learning such a classification task.
However, there is a solution: Zhou et al. [2022] use an Image to Image translation architecture called MUNIT to learn the transformations and generate additional data from which the CNN can learn the invariance separately for each class. Thus, the invariance to nuisance transformations is transferred generatively. They call this method Generative Invariance Transfer (GIT).
In this blog post, we are going to argue that the experiment described above is a meta-learning experiment and that MUNIT is related to meta-learning methods.

Blogpost Url: https://iclr-blogposts.github.io/staging/blog/2022/how-much-meta-learning-is-in-image-to-image-translation/

ICLR Papers: https://openreview.net/forum?id=Fn7i_r5rR0q

ID Of The Authors Of The ICLR Paper: ~Allan_Zhou1

Conflict Of Interest: No