ELECTRA: Pre-training Text Encoders as Discriminators Rather Than Generators
Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context.
MnasNet: Platform-Aware Neural Architecture Search for Mobile.
SpecAugment: A Simple Data Augmentation Method for Automatic Speech Recognition.
Using Videos to Evaluate Image Model Robustness.
Attention Augmented Convolutional Networks.
NAS-FPN: Learning Scalable Feature Pyramid Architecture for Object Detection.
Aging Evolution for Image Classifier Architecture Search
EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks.
Rethinking Model Scaling for Deep Convolutional Neural Networks
The Effect of Network Width on Stochastic Gradient Descent: an Empirical Study
Do better ImageNet models transfer better
BAM! Born-Again Multi-Task Networks for Natural Language Understanding
Regularized Evolution for Image Classifier Architecture Search
CondConv: Conditionally Parameterized Convolutions for Efficient Inference
Mixtape: Breaking the Softmax Bottleneck Efficiently
Saccader: Accurate, Interpretable Image Classification with Hard Attention
High Fidelity Video Prediction with Large Neural Nets
TensorPipe: Easy Scaling with Micro-Batch Pipeline Parallelism
High Fidelity Video Prediction with Large Stochastic Recurrent Neural Networks.
Saccader: Improving Accuracy of Hard Attention Models for Vision
GPipe: Efficient Training of Giant Neural Networks using Pipeline Parallelism
XLNet - Generalized Autoregressive Pretraining for Language Understanding.
Neural Program Synthesis with Priority Queue Training.
Efficient Neural Architecture Search via Parameter Sharing.
Learning Longer-term Dependencies in RNNs with Auxiliary Losses.
AutoAugment: Learning Augmentation Policies from Data.
Evolving modular neural sequence architectures with genetic programming.
Understanding and Simplifying One-Shot Architecture Search.
Stochastic natural gradient descent draws posterior samples in function space.
Memory Augmented Policy Optimization for Program Synthesis with Generalization.
A Simple Method for Commonsense Reasoning.
AirDialogue: An Environment for Goal-Oriented Dialogue Research.
Semi-Supervised Sequence Modeling with Cross-View Training.
Domain Adaptive Transfer Learning with Specialist Models.
Memory Augmented Policy Optimization for Program Synthesis and Semantic Parsing.
Scalable and accurate deep learning with electronic health records
Cross-View Training for Semi-Supervised Learning
A Hierarchical Model for Device Placement
Fast and Accurate Reading Comprehension by Combining Self-Attention and Convolution
Faster Discovery of Neural Architectures by Searching for Paths in a Large Model
A Hierarchical Model for Device Placement
Searching for Activation Functions
A Bayesian Perspective on Generalization and Stochastic Gradient Descent
Intriguing Properties of Adversarial Examples
Don't Decay the Learning Rate, Increase the Batch Size
Learning Transferable Architectures for Scalable Image Recognition
Adding Gradient Noise Improves Learning for Very Deep Networks.
Massive Exploration of Neural Machine Translation Architectures.
Effective Domain Mixing for Neural Machine Translation.
Tacotron: Towards End-to-End Speech Synthesis.
Neural Architecture Search with Reinforcement Learning
Listen, attend and spell: A neural network for large vocabulary conversational speech recognition.