Skip to search formSkip to main contentSkip to account menu
  • Publications
  • Influence
GradNorm: Gradient Normalization for Adaptive Loss Balancing in Deep Multitask Networks
TLDR
A gradient normalization (GradNorm) algorithm that automatically balances training in deep multitask models by dynamically tuning gradient magnitudes is presented, showing that for various network architectures, for both regression and classification tasks, and on both synthetic and real datasets, GradNorm improves accuracy and reduces overfitting across multiple tasks.
View PDF on arXiv
Deeply-Supervised Nets
TLDR
The proposed deeply-supervised nets (DSN) method simultaneously minimizes classification error while making the learning process of hidden layers direct and transparent, and extends techniques from stochastic gradient methods to analyze the algorithm.
View PDF on arXiv
Recursive Recurrent Nets with Attention Modeling for OCR in the Wild
TLDR
This work presents recursive recurrent neural networks with attention modeling (R2AM) for lexicon-free optical character recognition in natural scene images and validates the method with state-of-the-art performance on challenging benchmark datasets.
View on IEEE
A Simple Semi-Supervised Learning Framework for Object Detection
TLDR
STAC is proposed, a simple yet effective SSL framework for visual object detection along with a data augmentation strategy that deploys highly confident pseudo labels of localized objects from an unlabeled image and updates the model by enforcing consistency via strong augmentations.
View PDF on arXiv
Sliced Wasserstein Discrepancy for Unsupervised Domain Adaptation
TLDR
The proposed sliced Wasserstein discrepancy (SWD) is designed to capture the natural notion of dissimilarity between the outputs of task-specific classifiers and enables efficient distribution alignment in an end-to-end trainable fashion.
View on IEEE
Generalizing Pooling Functions in Convolutional Neural Networks: Mixed, Gated, and Tree
TLDR
The proposed pooling operations provide a boost in invariance properties relative to conventional pooling and set the state of the art on several widely adopted benchmark datasets; they are also easy to implement, and can be applied within various deep neural network architectures.
View PDF on arXiv
RoomNet: End-to-End Room Layout Estimation
TLDR
This paper predicts the locations of the room layout keypoints using RoomNet, an end-to-end trainable encoder-decoder network and presents optional extensions to the RoomNet architecture such as including recurrent computations and memory units to refine the keypoint locations under the same parametric capacity.
View on IEEE
Training Deeper Convolutional Networks with Deep Supervision
TLDR
A simple rule of thumb is formulated to determine where auxiliary supervision branches after certain intermediate layers during training are added in order to train deeper networks.
View PDF on arXiv
Learning to Branch for Multi-Task Learning
TLDR
This work proposes a novel tree-structured design space that casts a tree branching operation as a gumbel-softmax sampling procedure that enables differentiable network splitting that is end-to-end trainable.
View PDF on arXiv
Region-Based Discriminative Feature Pooling for Scene Text Recognition
TLDR
This work proposes a discriminative feature pooling method that automatically learns the most informative sub-regions of each scene character within a multi-class classification framework, whereas each sub-region seamlessly integrates a set of low-level image features through integral images.
View on IEEE
...
...
By clicking accept or continuing to use the site, you agree to the terms outlined in our Privacy Policy, Terms of Service, and Dataset License