Home /
Research
Showing 13 - 18 / 904
MixSize: Training Convnets With Mixed Image Sizes for Improved Accuracy, Speed and Scale Resiliency
Anonymous....
Published date-01/01/2021
Convolutional neural networks (CNNs) are commonly trained using a fixed spatial image size predetermined for a given model. Although trained on images of a specific size, it is well established …
Colorization Transformer
Anonymous....
Published date-01/01/2021
Colorization
We present the Colorization Transformer, a novel approach for diverse high fidelity image colorization based on self-attention. Given a grayscale image, the colorization proceeds in three steps. We first use …
Ruminating Word Representations with Random Noise Masking
Anonymous....
Published date-01/01/2021
TextClassification, WordEmbeddings
We introduce a training method for better word representation and performance, which we call \textbf{GraVeR} (\textbf{Gra}dual \textbf{Ve}ctor \textbf{R}umination). The method is to gradually and iteratively add random noises and bias …
Contrast to Divide: self-supervised pre-training for learning with noisy labels
Anonymous....
Published date-01/01/2021
ImageClassification, Learningwithnoisylabels
Advances in semi-supervised methods for image classification significantly boosted performance in the learning with noisy labels (LNL) task. Specifically, by discarding the erroneous labels (and keeping the samples), the LNL …
Trust, but verify: model-based exploration in sparse reward environments
Anonymous....
Published date-01/01/2021
We propose $\textit{trust-but-verify}$ (TBV) mechanism, a new method which uses model uncertainty estimates to guide exploration. The mechanism augments graph search planning algorithms by the capacity to deal with learned …
Disentangled cyclic reconstruction for domain adaptation
Anonymous....
Published date-01/01/2021
DomainAdaptation, UnsupervisedDomainAdaptation
The domain adaptation problem involves learning a unique classification or regres-sion model capable of performing on both a source and a target domain. Althoughthe labels for the source data are …