Evolving Losses for Unlabeled Video Representation Learning
This work addresses the problem of leveraging large-scale unlabeled video data for transferable representations, which is incremental as it builds on existing self-supervised and multi-modal learning approaches.
The paper tackles learning video representations from unlabeled data by formulating it as a multi-modal, multi-task problem and using an evolutionary algorithm to search for better loss functions, resulting in a method that distills audio, optical flow, and temporal information into a single RGB-based CNN and shows benefits on standard datasets.
We present a new method to learn video representations from unlabeled data. Given large-scale unlabeled video data, the objective is to benefit from such data by learning a generic and transferable representation space that can be directly used for a new task such as zero/few-shot learning. We formulate our unsupervised representation learning as a multi-modal, multi-task learning problem, where the representations are also shared across different modalities via distillation. Further, we also introduce the concept of finding a better loss function to train such multi-task multi-modal representation space using an evolutionary algorithm; our method automatically searches over different combinations of loss functions capturing multiple (self-supervised) tasks and modalities. Our formulation allows for the distillation of audio, optical flow and temporal information into a single, RGB-based convolutional neural network. We also compare the effects of using additional unlabeled video data and evaluate our representation learning on standard public video datasets.