Evolving Losses for Unsupervised Video Representation Learning
This work addresses the problem of learning generic video representations without labels for tasks like action recognition, though it is incremental in combining existing techniques like distillation and evolutionary search.
The authors tackled unsupervised video representation learning by formulating it as a multi-modal, multi-task problem with loss function evolution and an unsupervised evaluation metric, resulting in a single RGB network that outperforms previous unsupervised methods and is more effective than several label-based methods except for large fully labeled video datasets.
We present a new method to learn video representations from large-scale unlabeled video data. Ideally, this representation will be generic and transferable, directly usable for new tasks such as action recognition and zero or few-shot learning. We formulate unsupervised representation learning as a multi-modal, multi-task learning problem, where the representations are shared across different modalities via distillation. Further, we introduce the concept of loss function evolution by using an evolutionary search algorithm to automatically find optimal combination of loss functions capturing many (self-supervised) tasks and modalities. Thirdly, we propose an unsupervised representation evaluation metric using distribution matching to a large unlabeled dataset as a prior constraint, based on Zipf's law. This unsupervised constraint, which is not guided by any labeling, produces similar results to weakly-supervised, task-specific ones. The proposed unsupervised representation learning results in a single RGB network and outperforms previous methods. Notably, it is also more effective than several label-based methods (e.g., ImageNet), with the exception of large, fully labeled video datasets.