… Self-supervision provides effective … fully supervised training and are often not thought
beneficial beyond obviating or reducing the need for annotations. We find that self-supervision can …

… In the semi-supervised setting, we demonstrate results equal … (iii) We show strong
generalization capabilities and outper… models: Weight-averaged consistency targets improve semi-…

… of the loss landscape and generalization, we … improves model generalization across a
variety of benchmark datasets (eg, CIFAR-10, CIFAR-100, ImageNet, finetuning tasks) and models…

… Our loss can be seen as a generalization of both the triplet [… improve upon [41] by normalizing
the embeddings and replacing euclidean distance with inner products. We further improve …

… was able to improve the generalization performance. Random … to improve generalization
performance for semi-supervised … of techniques that smooth the model around input data points …

… ) neural network for self-supervised pretraining and fine-tuning greatly improves accuracy. In
… , which could easily overfit with few labeled examples, can generalize much better. With task…

… the base model, emphasizing the need for domain generalization. Our … model to provide
better performance. Figure 4(c) shows the improvement of our DG model over the base model …

… Since we are interested in selecting examples that improve the generalization behavior of
some given neural network architecture N, we define the concept class in question to be the …

… We show how the S3VM model for I-norm linear support vector … data improves generalization
when insufficient training information is available. In every case, S3VM either improved or …

… Thus DG is to achieve an agent with improved generalization ability in the sense of … ) –
without being allowed any rewarded (≈ supervised domain adaptation (Finn, Abbeel, and Levine …