This paper deals with deep transductive learning, and proposes TransBoost as a procedure for fine-tuning any deep neural model to improve its performance on any (unlabeled) test set provided at training time. TransBoost is inspired by a large margin principle and is efficient and simple to use. The ImageNet classification performance is consistently and significantly improved with TransBoost on many architectures such as ResNets, MobileNetV3-L, EfficientNetB0, ViT-S, and ConvNext-T. Additionally we show that TransBoost is effective on a wide variety of image classification datasets.
翻译:本文论述深层感应学习,并提议将TransBoost作为微调任何深神经模型的程序,以提高其在培训时提供的任何(未贴标签的)测试集的性能。TransBoost受大边距原则的启发,并且高效和简单使用。图像网络分类性能与TransBoost在许多结构上如ResNets、MobilNetV3-L、高效NetB0、ViT-S和ConvNext-T, 一致和显著改进。此外,我们还表明,TransBoost对各种图像分类数据集有效。