S
Shell Xu Hu
Researcher at École des ponts ParisTech
Publications - 20
Citations - 781
Shell Xu Hu is an academic researcher from École des ponts ParisTech. The author has contributed to research in topics: Computer science & Artificial neural network. The author has an hindex of 5, co-authored 9 publications receiving 273 citations.
Papers
More filters
Proceedings ArticleDOI
Variational Information Distillation for Knowledge Transfer
TL;DR: In this article, the authors propose an information-theoretic framework for knowledge transfer which formulates knowledge transfer as maximizing the mutual information between the teacher and the student networks, and compare their method with existing knowledge transfer methods on both knowledge distillation and transfer learning tasks and show that their method consistently outperforms existing methods.
Posted Content
Variational Information Distillation for Knowledge Transfer.
TL;DR: An information-theoretic framework for knowledge transfer is proposed which formulates knowledge transfer as maximizing the mutual information between the teacher and the student networks and which consistently outperforms existing methods.
Posted Content
Empirical Bayes Transductive Meta-Learning with Synthetic Gradients
Shell Xu Hu,Pablo Garcia Moreno,Yang Xiao,Xi Shen,Guillaume Obozinski,Neil D. Lawrence,Andreas Damianou +6 more
TL;DR: A novel amortized variational inference that couples all the variational posteriors into a meta-model, which consists of a synthetic gradient network and an initialization network that allows for backpropagating information from unlabeled data, thereby enabling transduction.
Proceedings ArticleDOI
Self-Supervised Transformers for Unsupervised Object Discovery using Normalized Cut
TL;DR: A graph-based method that uses the selfsupervised transformer features to discover an object from an image using spectral clustering with generalized eigen-decomposition and showing that the second smallest eigenvector provides a cutting solution since its absolute value indicates the likelihood that a token belongs to a foreground object.
Proceedings ArticleDOI
Pushing the Limits of Simple Pipelines for Few-Shot Learning: External Data and Fine-Tuning Make a Difference
TL;DR: It is shown that a simple transformer-based pipeline yields surprisingly good performance on standard benchmarks such as Mini-ImageNet, CIFAR-FS, CDFSL and Meta-Dataset.