AutoAugment: Learning Augmentation Policies from Data

Abstract

Data augmentation is an effective technique for improving the accuracy of modern image classifiers. However, current data augmentation implementations are manually designed. In this paper, we describe a simple procedure called AutoAugment to automatically search for improved data augmentation policies. In our implementation, we have designed a search space where a policy consists of many sub-policies, one of which is randomly chosen for each image in each mini-batch. A sub-policy consists of two operations, each operation being an image processing function such as translation, rotation, or shearing, and the probabilities and magnitudes with which the functions are applied. We use a search algorithm to find the best policy such that the neural network yields the highest validation accuracy on a target dataset. Our method achieves state-of-the-art accuracy on CIFAR-10, CIFAR-100, SVHN, and ImageNet (without additional data). On ImageNet, we attain a Top-1 accuracy of 83.5% which is 0.4% better than the previous record of 83.1%. On CIFAR-10, we achieve an error rate of 1.5%, which is 0.6% better than the previous state-of-the-art. Augmentation policies we find are transferable between datasets. The policy learned on ImageNet transfers well to achieve significant improvements on other datasets, such as Oxford Flowers, Caltech-101, Oxford-IIT Pets, FGVC Aircraft, and Stanford Cars.

CVPR 2019

References in corpus (20)

• Neural Architecture Search with Reinforcement Learning
• Improved Regularization of Convolutional Neural Networks with Cutout
• The Effectiveness of Data Augmentation in Image Classification using Deep Learning
• Temporal Ensembling for Semi-Supervised Learning
• Going Deeper with Convolutions
• Random Erasing Data Augmentation
• SMASH: One-Shot Model Architecture Search through HyperNetworks
• Shake-Shake regularization
• Realistic Evaluation of Deep Semi-Supervised Learning Algorithms
• Neural Optimizer Search with Reinforcement Learning
• ShakeDrop Regularization for Deep Residual Learning
• Simple And Efficient Architecture Search for Convolutional Neural Networks
• Simple random search provides a competitive approach to reinforcement learning
• The Relationship Between Local Structure and Relaxation in Out-of-Equilibrium Glassy Systems
• Do CIFAR-10 Classifiers Generalize to CIFAR-10?
• A Bayesian Data Augmentation Approach for Learning Deep Models
• APAC: Augmented PAttern Classification with Neural Networks
• Data Augmentation in Emotion Classification Using Generative Adversarial Networks
• Intriguing Properties of Adversarial Examples
• Parallel Architecture and Hyperparameter Search via Successive Halving and Classification

Cited by in corpus (333)

• EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks
• SpecAugment: A Simple Data Augmentation Method for Automatic Speech Recognition
• Generative Adversarial Network in Medical Imaging: A Review
• Neural Architecture Search: A Survey
• Unsupervised Data Augmentation for Consistency Training
• MLP-Mixer: An all-MLP Architecture for Vision
• Training Generative Adversarial Networks with Limited Data
• Data-Efficient Image Recognition with Contrastive Predictive Coding
• Generalizing from a Few Examples: A Survey on Few-Shot Learning
• Solving Rubik's Cube with a Robot Hand
• MixMatch: A Holistic Approach to Semi-Supervised Learning
• AugMix: A Simple Data Processing Method to Improve Robustness and Uncertainty
• Data Augmentation using Random Image Cropping and Patching for Deep CNNs
• Avoiding Overfitting: A Survey on Regularization Methods for Convolutional Neural Networks
• Do ImageNet Classifiers Generalize to ImageNet?
• Rethinking Pre-training and Self-training
• Early Convolutions Help Transformers See Better
• Billion-scale semi-supervised learning for image classification
• DARTS+: Improved Differentiable Architecture Search with Early Stopping
• Out-of-Distribution Generalization via Risk Extrapolation (REx)
• GPipe: Efficient Training of Giant Neural Networks using Pipeline Parallelism
• Machine-Learning-Based Diagnostics of EEG Pathology
• Revisiting ResNets: Improved Training and Scaling Strategies
• A Survey on Neural Architecture Search
• High Quality Monocular Depth Estimation via Transfer Learning
• Quantifying Generalization in Reinforcement Learning
• On Empirical Comparisons of Optimizers for Deep Learning
• Measuring Robustness to Natural Distribution Shifts in Image Classification
• ReMixMatch: Semi-Supervised Learning with Distribution Alignment and Augmentation Anchoring
• Improving robustness against common corruptions by covariate shift adaptation
• A Large-scale Study of Representation Learning with the Visual Task Adaptation Benchmark
• A Survey on Data Collection for Machine Learning: a Big Data -- AI Integration Perspective
• Uncovering the Limits of Adversarial Training against Norm-Bounded Adversarial Examples
• Adversarial Examples Are a Natural Consequence of Test Error in Noise
• AutoML-Zero: Evolving Machine Learning Algorithms From Scratch
• Cross-Domain Few-Shot Classification via Learned Feature-Wise Transformation
• Recent Advances in Object Detection in the Age of Deep Convolutional Neural Networks
• NAS evaluation is frustratingly hard
• Fixing Data Augmentation to Improve Adversarial Robustness
• Sharpness-Aware Minimization for Efficiently Improving Generalization
• See Better Before Looking Closer: Weakly Supervised Data Augmentation Network for Fine-Grained Visual Classification
• The Conditional Entropy Bottleneck
• Differentiable Augmentation for Data-Efficient GAN Training
• Label-Only Membership Inference Attacks
• Smooth Adversarial Training
• Improving Robustness Without Sacrificing Accuracy with Patch Gaussian Augmentation
• Best Practices for Scientific Research on Neural Architecture Search
• A Fourier Perspective on Model Robustness in Computer Vision
• AdamP: Slowing Down the Slowdown for Momentum Optimizers on Scale-invariant Weights
• Selfie: Self-supervised Pretraining for Image Embedding
• Unlabeled Data Improves Adversarial Robustness
• Fixing the train-test resolution discrepancy: FixEfficientNet
• ASAP: Architecture Search, Anneal and Prune
• Rethinking the Hyperparameters for Fine-tuning
• ConTNet: Why not use convolution and transformer at the same time?
• Soft Contextual Data Augmentation for Neural Machine Translation
• Data augmentation using learned transformations for one-shot medical image segmentation
• Small Sample Learning in Big Data Era
• i-Mix: A Domain-Agnostic Strategy for Contrastive Representation Learning
• Affinity and Diversity: Quantifying Mechanisms of Data Augmentation
• Dataset Condensation with Differentiable Siamese Augmentation
• Multi-Sample Dropout for Accelerated Training and Better Generalization
• Augment your batch: better training with larger batches
• MEAL V2: Boosting Vanilla ResNet-50 to 80%+ Top-1 Accuracy on ImageNet without Tricks
• Adversarial Examples Improve Image Recognition
• An Ensemble of Simple Convolutional Neural Network Models for MNIST Digit Recognition
• Network Randomization: A Simple Technique for Generalization in Deep Reinforcement Learning
• Skin Lesions Classification Using Convolutional Neural Networks in Clinical Images
• Reinforcement Learning Applications
• Stabilizing DARTS with Amended Gradient Estimation on Architectural Parameters
• Subspace Attack: Exploiting Promising Subspaces for Query-Efficient Black-box Attacks
• Optimizing Millions of Hyperparameters by Implicit Differentiation
• AtomNAS: Fine-Grained End-to-End Neural Architecture Search
• FedCV: A Federated Learning Framework for Diverse Computer Vision Tasks
• ReSSL: Relational Self-Supervised Learning with Weak Augmentation
• Sampling Strategies for GAN Synthetic Data
• ResizeMix: Mixing Data with Preserved Object Information and True Labels
• BigNAS: Scaling Up Neural Architecture Search with Big Single-Stage Models
• Ensemble learning in CNN augmented with fully connected subnetworks
• Auto-FedAvg: Learnable Federated Averaging for Multi-Institutional Medical Image Segmentation
• Noisy Differentiable Architecture Search
• Compounding the Performance Improvements of Assembled Techniques in a Convolutional Neural Network
• Contrastive Learning with Stronger Augmentations
• An Empirical Evaluation on Robustness and Uncertainty of Regularization Methods
• sharpDARTS: Faster and More Accurate Differentiable Architecture Search
• Assume, Augment and Learn: Unsupervised Few-Shot Meta-Learning via Random Labels and Data Augmentation
• MultiGrain: a unified image embedding for classes and instances
• DeepLab2: A TensorFlow Library for Deep Labeling
• MaxUp: A Simple Way to Improve Generalization of Neural Network Training
• ResNet strikes back: An improved training procedure in timm
• Adversarial Examples Make Strong Poisons
• An embedded system for the automated generation of labeled plant images to enable machine learning applications in agriculture
• Data-Efficient Instance Generation from Instance Discrimination
• Regularizing Class-wise Predictions via Self-knowledge Distillation
• DARTS-: Robustly Stepping out of Performance Collapse Without Indicators
• Fair DARTS: Eliminating Unfair Advantages in Differentiable Architecture Search
• GOLD-NAS: Gradual, One-Level, Differentiable
• Using Videos to Evaluate Image Model Robustness
• Large-Scale Generative Data-Free Distillation
• Scaling Wide Residual Networks for Panoptic Segmentation
• Weight-Sharing Neural Architecture Search: A Battle to Shrink the Optimization Gap
• CoDA: Contrast-enhanced and Diversity-promoting Data Augmentation for Natural Language Understanding
• Data Augmentation for Meta-Learning
• Dynamic Scale Training for Object Detection
• Overton: A Data System for Monitoring and Improving Machine-Learned Products
• Snorkel DryBell: A Case Study in Deploying Weak Supervision at Industrial Scale
• Meta-Learning Symmetries by Reparameterization
• Regularizing Deep Networks with Semantic Data Augmentation
• Fixing the train-test resolution discrepancy
• SemiFed: Semi-supervised Federated Learning with Consistency and Pseudo-Labeling
• Adapting Semantic Segmentation Models for Changes in Illumination and Camera Perspective
• An Effective Anti-Aliasing Approach for Residual Networks
• Neural Transformation Learning for Deep Anomaly Detection Beyond Images
• XNAS: Neural Architecture Search with Expert Advice
• Viewmaker Networks: Learning Views for Unsupervised Representation Learning
• Data Augmentation for Electrocardiogram Classification with Deep Neural Network
• Domain Randomization and Pyramid Consistency: Simulation-to-Real Generalization without Accessing Target Domain Data
• Semi-Supervised Visual Representation Learning for Fashion Compatibility
• CIFAR-10 Image Classification Using Feature Ensembles
• Role-Wise Data Augmentation for Knowledge Distillation
• SystemDS: A Declarative Machine Learning System for the End-to-End Data Science Lifecycle
• HARK Side of Deep Learning -- From Grad Student Descent to Automated Machine Learning
• MANAS: Multi-Agent Neural Architecture Search
• HardCoRe-NAS: Hard Constrained diffeRentiable Neural Architecture Search
• MoViNets: Mobile Video Networks for Efficient Video Recognition
• Neural Architecture Generator Optimization
• On the effectiveness of adversarial training against common corruptions
• AdaBits: Neural Network Quantization with Adaptive Bit-Widths
• Phase Transitions for the Information Bottleneck in Representation Learning
• On Data-Augmentation and Consistency-Based Semi-Supervised Learning
• Improving Semi-supervised Federated Learning by Reducing the Gradient Diversity of Models
• EnAET: A Self-Trained framework for Semi-Supervised and Supervised Learning with Ensemble Transformations
• What do AI algorithms actually learn? - On false structures in deep learning
• MixPath: A Unified Approach for One-shot Neural Architecture Search
• Evolutionary-Neural Hybrid Agents for Architecture Search
• BraggNN: Fast X-ray Bragg Peak Analysis Using Deep Learning
• A Large Multi-Target Dataset of Common Bengali Handwritten Graphemes
• Efficient Differentiable Neural Architecture Search with Meta Kernels
• Data Augmentation for Deep Learning-based Radio Modulation Classification
• LibFewShot: A Comprehensive Library for Few-shot Learning
• KeepAugment: A Simple Information-Preserving Data Augmentation Approach
• Energy Models for Better Pseudo-Labels: Improving Semi-Supervised Classification with the 1-Laplacian Graph Energy
• Using learned optimizers to make models robust to input noise
• Novelty Detection Via Blurring
• On the Generalization Effects of Linear Transformations in Data Augmentation
• AttentiveNAS: Improving Neural Architecture Search via Attentive Sampling
• SGAS: Sequential Greedy Architecture Search
• BETANAS: BalancEd TrAining and selective drop for Neural Architecture Search
• Improving 3D Object Detection through Progressive Population Based Augmentation
• Inspector Gadget: A Data Programming-based Labeling System for Industrial Images
• Data Augmentation with Manifold Exploring Geometric Transformations for Increased Performance and Robustness
• Zen-NAS: A Zero-Shot NAS for High-Performance Deep Image Recognition
• VIPriors 1: Visual Inductive Priors for Data-Efficient Deep Learning Challenges
• Neuroevolution of Neural Network Architectures Using CoDeepNEAT and Keras
• Self-supervised Neural Architecture Search
• Learning Optimal Conformal Classifiers
• Robust Nucleus Detection with Partially Labeled Exemplars
• Neural Architecture Design for GPU-Efficient Networks
• Learning by Analogy: Reliable Supervision from Transformations for Unsupervised Optical Flow Estimation
• Strategy to Increase the Safety of a DNN-based Perception for HAD Systems
• Neural Semi-supervised Learning for Text Classification Under Large-Scale Pretraining
• Image quality assessment by overlapping task-specific and task-agnostic measures: application to prostate multiparametric MR images for cancer segmentation
• Discrete Representations Strengthen Vision Transformer Robustness
• Efficient Neural Architecture Search via Proximal Iterations
• FeatMatch: Feature-Based Augmentation for Semi-Supervised Learning
• Improving Neural Architecture Search Image Classifiers via Ensemble Learning
• Hierarchical Neural Architecture Search via Operator Clustering
• Adversarial Feature Augmentation and Normalization for Visual Recognition
• Few-shot Neural Architecture Search
• ALBA : Reinforcement Learning for Video Object Segmentation
• Online Hyper-parameter Learning for Auto-Augmentation Strategy
• It Takes Two to Tango: Mixup for Deep Metric Learning
• SSFN -- Self Size-estimating Feed-forward Network with Low Complexity, Limited Need for Human Intervention, and Consistent Behaviour across Trials
• Data Augmentation for Object Detection via Differentiable Neural Rendering
• Bit Error Robustness for Energy-Efficient DNN Accelerators
• TF-NAS: Rethinking Three Search Freedoms of Latency-Constrained Differentiable Neural Architecture Search
• Causally motivated Shortcut Removal Using Auxiliary Labels
• DHA: End-to-End Joint Optimization of Data Augmentation Policy, Hyper-parameter and Architecture
• Automated Learning Rate Scheduler for Large-batch Training
• Making EfficientNet More Efficient: Exploring Batch-Independent Normalization, Group Convolutions and Reduced Resolution Training
• A Framework for Studying Reinforcement Learning and Sim-to-Real in Robot Soccer
• When Human Pose Estimation Meets Robustness: Adversarial Algorithms and Benchmarks
• Escaping Saddle Points Faster with Stochastic Momentum
• PV-NAS: Practical Neural Architecture Search for Video Recognition
• A Simple Baseline for Semi-supervised Semantic Segmentation with Strong Data Augmentation
• ViP-DeepLab: Learning Visual Perception with Depth-aware Video Panoptic Segmentation
• Powering One-shot Topological NAS with Stabilized Share-parameter Proxy
• Improved Adversarial Robustness via Logit Regularization Methods
• Dialog State Tracking with Reinforced Data Augmentation
• Improved Mutual Mean-Teaching for Unsupervised Domain Adaptive Re-ID
• CoKe: Localized Contrastive Learning for Robust Keypoint Detection
• E-Stitchup: Data Augmentation for Pre-Trained Embeddings
• Meta Approach to Data Augmentation Optimization
• Learning to Generate Synthetic Data via Compositing
• Multi-Task Curriculum Framework for Open-Set Semi-Supervised Learning
• VEGA: Towards an End-to-End Configurable AutoML Pipeline
• MetaMixUp: Learning Adaptive Interpolation Policy of MixUp with Meta-Learning
• Series Saliency: Temporal Interpretation for Multivariate Time Series Forecasting
• AdaSGD: Bridging the gap between SGD and Adam
• A Preliminary Study on Data Augmentation of Deep Learning for Image Classification
• Neural Packet Classification
• How can AI Automate End-to-End Data Science?
• Improving Robustness of Learning-based Autonomous Steering Using Adversarial Images
• ObjectAug: Object-level Data Augmentation for Semantic Image Segmentation
• Machine learning with limited data
• 1st Place Solution for the UVO Challenge on Image-based Open-World Segmentation 2021
• EvoGrad: Efficient Gradient-Based Meta-Learning and Hyperparameter Optimization
• No One Representation to Rule Them All: Overlapping Features of Training Methods
• An Asymptotically Optimal Multi-Armed Bandit Algorithm and Hyperparameter Optimization
• Improving Robustness using Generated Data
• Learning Optimal Data Augmentation Policies via Bayesian Optimization for Image Classification Tasks
• Neural Networks Are More Productive Teachers Than Human Raters: Active Mixup for Data-Efficient Knowledge Distillation from a Blackbox Model
• Privacy-preserving Collaborative Learning with Automatic Transformation Search
• AutoFlow: Learning a Better Training Set for Optical Flow
• Disentanglement-based Cross-Domain Feature Augmentation for Effective Unsupervised Domain Adaptive Person Re-identification
• GuidedMix-Net: Learning to Improve Pseudo Masks Using Labeled Images as Reference
• Inner-Imaging Networks: Put Lenses into Convolutional Structure
• An Empirical Study and Analysis on Open-Set Semi-Supervised Learning
• AlphaNet: Improved Training of Supernets with Alpha-Divergence
• Augmented Parallel-Pyramid Net for Attention Guided Pose-Estimation
• SCARLET-NAS: Bridging the Gap between Stability and Scalability in Weight-sharing Neural Architecture Search
• Imponderous Net for Facial Expression Recognition in the Wild
• Random Shadows and Highlights: A new data augmentation method for extreme lighting conditions
• Data augmentation and image understanding
• Learning from Data-Rich Problems: A Case Study on Genetic Variant Calling
• Automatic Data Augmentation by Learning the Deterministic Policy
• Intelligence, physics and information -- the tradeoff between accuracy and simplicity in machine learning
• A Survey of Techniques All Classifiers Can Learn from Deep Networks: Models, Optimizations, and Regularization
• Defending Against Image Corruptions Through Adversarial Augmentations
• A Simple yet Effective Baseline for Robust Deep Learning with Noisy Labels
• A Robotic Approach towards Quantifying Epipelagic Bound Plastic Using Deep Visual Models
• Optimizing Neural Architecture Search using Limited GPU Time in a Dynamic Search Space: A Gene Expression Programming Approach
• Stronger Baseline for Person Re-Identification
• Effective Evaluation of Deep Active Learning on Image Classification Tasks
• Learning Visual Representations for Transfer Learning by Suppressing Texture
• AutoEG: Automated Experience Grafting for Off-Policy Deep Reinforcement Learning
• HR-NAS: Searching Efficient High-Resolution Neural Architectures with Lightweight Transformers
• Reweighting Augmented Samples by Minimizing the Maximal Expected Loss
• Pretraining Image Encoders without Reconstruction via Feature Prediction Loss
• Practical Assessment of Generalization Performance Robustness for Deep Networks via Contrastive Examples
• BERT for Large-scale Video Segment Classification with Test-time Augmentation
• Exploiting Cross-Modal Prediction and Relation Consistency for Semi-Supervised Image Captioning
• Patch augmentation: Towards efficient decision boundaries for neural networks
• Deep Neural Network Ensembles
• Regularized Evolutionary Population-Based Training
• Rethinking ResNets: Improved Stacking Strategies With High Order Schemes
• Selective sampling for accelerating training of deep neural networks
• Beyond Dropout: Feature Map Distortion to Regularize Deep Neural Networks
• Interpret-able feedback for AutoML systems
• Margin-Based Regularization and Selective Sampling in Deep Neural Networks
• Efficient Neural Architecture Search with Performance Prediction
• Automatically Learning Data Augmentation Policies for Dialogue Tasks
• On the Orthogonality of Knowledge Distillation with Other Techniques: From an Ensemble Perspective
• Learning Purified Feature Representations from Task-irrelevant Labels
• Local Patch AutoAugment with Multi-Agent Collaboration
• Ferrograph image classification
• Discriminative Cross-Modal Data Augmentation for Medical Imaging Applications
• DeepMix: Online Auto Data Augmentation for Robust Visual Object Tracking
• Robust Training Using Natural Transformation
• Dialogue Distillation: Open-Domain Dialogue Augmentation Using Unpaired Data
• ExCon: Explanation-driven Supervised Contrastive Learning for Image Classification
• Confounder Identification-free Causal Visual Feature Learning
• Reliable Label Bootstrapping for Semi-Supervised Learning
• What augmentations are sensitive to hyper-parameters and why?
• HERO: Hessian-Enhanced Robust Optimization for Unifying and Improving Generalization and Quantization Performance
• Object-Aware Cropping for Self-Supervised Learning
• Temporally Resolution Decrement: Utilizing the Shape Consistency for Higher Computational Efficiency
• Gradient-based Data Augmentation for Semi-Supervised Learning
• Distribution Mismatch Correction for Improved Robustness in Deep Neural Networks
• LogAvgExp Provides a Principled and Performant Global Pooling Operator
• PanDA: Panoptic Data Augmentation
• Learning Non-Parametric Invariances from Data with Permanent Random Connectomes
• Learning image quality assessment by reinforcing task amenable data selection
• Efficient Model Performance Estimation via Feature Histories
• Similarity Transfer for Knowledge Distillation
• UniMoCo: Unsupervised, Semi-Supervised and Full-Supervised Visual Representation Learning
• Enabling Data Diversity: Efficient Automatic Augmentation via Regularized Adversarial Training
• Hyperparameter Optimization in Neural Networks via Structured Sparse Recovery
• A Technical Report for VIPriors Image Classification Challenge
• On the Accuracy of CRNNs for Line-Based OCR: A Multi-Parameter Evaluation
• Dynamic Routing Networks
• EENA: Efficient Evolution of Neural Architecture
• Dynamic Mode Decomposition based feature for Image Classification
• Exploring Frequency Domain Interpretation of Convolutional Neural Networks
• ADWPNAS: Architecture-Driven Weight Prediction for Neural Architecture Search
• FlipReID: Closing the Gap between Training and Inference in Person Re-Identification
• Signed Input Regularization
• A Reinforcement Learning Approach for Sequential Spatial Transformer Networks
• Exploiting Class Similarity for Machine Learning with Confidence Labels and Projective Loss Functions
• Understanding Modern Techniques in Optimization: Frank-Wolfe, Nesterov's Momentum, and Polyak's Momentum
• Semi-supervised Learning of Fetal Anatomy from Ultrasound
• Differentiable Network Adaption with Elastic Search Space
• What Else Can Fool Deep Learning? Addressing Color Constancy Errors on Deep Neural Network Performance
• Go Small and Similar: A Simple Output Decay Brings Better Performance
• Learning to Transfer Learn: Reinforcement Learning-Based Selection for Adaptive Transfer Learning
• Research on Optimization Method of Multi-scale Fish Target Fast Detection Network
• Safe Augmentation: Learning Task-Specific Transformations from Data
• Rotating spiders and reflecting dogs: a class conditional approach to learning data augmentation distributions
• Searching Learning Strategy with Reinforcement Learning for 3D Medical Image Segmentation
• Efficient data augmentation using graph imputation neural networks
• Robust Learning with Frequency Domain Regularization
• One-Shot Neural Ensemble Architecture Search by Diversity-Guided Search Space Shrinking
• Selective Output Smoothing Regularization: Regularize Neural Networks by Softening Output Distributions
• Towards All-around Knowledge Transferring: Learning From Task-irrelevant Labels
• On the potential for open-endedness in neural networks
• Pathology-Aware Generative Adversarial Networks for Medical Image Augmentation
• Stochastic Contrastive Learning
• Doctor of Crosswise: Reducing Over-parametrization in Neural Networks
• MC-SSL0.0: Towards Multi-Concept Self-Supervised Learning
• EffCNet: An Efficient CondenseNet for Image Classification on NXP BlueBox
• DIVA: Dataset Derivative of a Learning Task
• Large-Scale Multi-Agent Deep FBSDEs
• Implicit Label Augmentation on Partially Annotated Clips via Temporally-Adaptive Features Learning
• Contrastive Representation Learning with Trainable Augmentation Channel
• CoopSubNet: Cooperating Subnetwork for Data-Driven Regularization of Deep Networks under Limited Training Budgets
• Hierarchical Auxiliary Learning
• Multi-Glimpse Network: A Robust and Efficient Classification Architecture based on Recurrent Downsampled Attention
• FROB: Few-shot ROBust Model for Classification and Out-of-Distribution Detection
• $P^2$ Net: Augmented Parallel-Pyramid Net for Attention Guided Pose Estimation
• Communication-Efficient Separable Neural Network for Distributed Inference on Edge Devices
• Unlabeled Data Guided Semi-supervised Histopathology Image Segmentation
• Mining Domain Knowledge: Improved Framework towards Automatically Standardizing Anatomical Structure Nomenclature in Radiotherapy
• Mixed precision in Graphics Processing Unit
• Unchain the Search Space with Hierarchical Differentiable Architecture Search
• Inspect Transfer Learning Architecture with Dilated Convolution
• Revisiting Self-Training for Few-Shot Learning of Language Model
• ISyNet: Convolutional Neural Networks design for AI accelerator
• Boosting Network Weight Separability via Feed-Backward Reconstruction
• Deep Epidemiological Modeling by Black-box Knowledge Distillation: An Accurate Deep Learning Model for COVID-19
• Sparsity-Probe: Analysis tool for Deep Learning Models
• Exploring and Improving Mobile Level Vision Transformers
• A Generalization Theory based on Independent and Task-Identically Distributed Assumption
• Geometric Data Augmentation Based on Feature Map Ensemble