Improved uncertainty estimates for classification models reduce calibration error.
problem Lack of calibrated uncertainty estimates in modern deep learning models.
method Restricting predictions to Top-1 error probabilities to improve calibration.
result Calibration error decreased to less than 1%.
We explore efficient neural architecture search methods and show that a simple yet powerful evolutionary algorithm can discover new architectures with excellent performance. Our approach combines a novel hierarchical genetic representation scheme that imitates the modularized design pattern commonly adopted by human ex…
This paper improves auto-augment efficiency by sharing augmentation weights.
problem Efficient evaluation of augmentation policies for model training.
method Augmentation-Wise Weight Sharing (AWS) to create a fast yet accurate proxy task.
result Augmentation policies found achieve superior accuracies compared to existing methods.
IC-Network improves CNNs by integrating elastic collision units.
problem Designing more effective basic units in neural networks.
method Developed IC layer and IC block units combining the IC structure with convolution operations.
result Significant performance improvements in existing CNNs, reducing top-1 error from 22.85% to 21.49% on imagenet.
Extracurricular learning closes the accuracy gap in knowledge distillation.
problem Accuracy gap between teacher and student models after knowledge distillation.
method Modeling student and teacher output distributions, sampling from an extended data distribution, and matching over this set.
result Extracurricular learning reduces the accuracy gap by 46% to 68%.
Developing neural network image classification models often requires significant architecture engineering. In this paper, we study a method to learn the model architectures directly on the dataset of interest. As this approach is expensive when the dataset is large, we propose to search for an architectural building bl…
In order to push the performance on realistic computer vision tasks, the number of classes in modern benchmark datasets has significantly increased in recent years. This increase in the number of classes comes along with increased ambiguity between the class labels, raising the question if top-1 error is the right perf…
Predict accuracy of neural architectures using non-neural models.
problem Improving efficiency and effectiveness of neural architecture search.
method Used gradient boosting decision tree (GBDT) for accuracy prediction and pruned the search space based on features learned from GBDT.
result GBDT-based predictor achieves comparable or better accuracy than neural network-based predictors and is 22x more sample efficient on NASBench-101.
Develops confidence intervals for ECE, a measure of model calibration.
problem Ensuring the calibration of probabilistic predictions in machine learning models.
method Develops confidence intervals for the ℓ2 Expected Calibration Error (ECE), considering top-1-to-k calibration. result Shows asymptotic normality and different convergence rates for calibrated and miscalibrated models, developing methods to construct valid confidence intervals.
Improved ImageNet classification with semi-supervised learning.
problem Image classification with limited labeled data.
method Noisy Student Training: semi-supervised learning with noisy student models.
result 88.4% top-1 accuracy on ImageNet, 2.0% better than state-of-the-art.
Paper analyzes trade-offs in top-k classification accuracies and proposes a new loss function.
problem CE loss does not always optimize top-k prediction, especially with complex data.
method Introduces a novel top-k transition loss to improve top-k accuracy.
result Our loss function improves top-k accuracy, especially for k > 10.
Adversarial AutoAugment improves image classification with less computation.
problem Improving deep neural network generalization through data augmentation.
method Adversarial method to optimize augmentation policies without retraining the target network.
result Significant performance improvements on ImageNet and CIFAR-10.
Minimalistic unsupervised learning with sparse manifold transform achieves SOTA performance.
problem Achieving state-of-the-art unsupervised learning performance without complex engineering.
method Sparse manifold transform, leveraging sparse coding, manifold learning, and slow feature analysis.
result 99.3% KNN top-1 accuracy on MNIST, 81.1% on CIFAR-10, and 53.2% on CIFAR-100.
Top-k error is currently a popular performance measure on large scale image classification benchmarks such as ImageNet and Places. Despite its wide acceptance, our understanding of this metric is limited as most of the previous research is focused on its special case, the top-1 error. In this work, we explore two direc…
Synchronized stochastic gradient descent (SGD) optimizers with data parallelism are widely used in training large-scale deep neural networks. Although using larger mini-batch sizes can improve the system scalability by reducing the communication-to-computation ratio, it may hurt the generalization ability of the models…
Proposes learning ordered Top-k attacks for image classification.
problem Vulnerability of DNNs to adversarial attacks, especially white-box targeted attacks.
method Adversarial distillation framework to compute adversarial probability distributions and minimize KL divergence.
result Outperforms C&W method in Top-1 and Top-5 settings for image classification.
BASE framework learns task-agnostic architectures for faster NAS.
problem High computational cost in Neural Architecture Search.
method Bayesian Meta Architecture Search (BASE) framework.
result Found models achieving 25.7% top-1 error and 8.1% top-5 error in less than an hour.
For fast and energy-efficient deployment of trained deep neural networks on resource-constrained embedded hardware, each learned weight parameter should ideally be represented and stored using a single bit. Error-rates usually increase when this requirement is imposed. Here, we report large improvements in error rates …
Ensemble methods are arguably the most trustworthy techniques for boosting the performance of machine learning models. Popular independent ensembles (IE) relying on naive averaging/voting scheme have been of typical choice for most applications involving deep neural networks, but they do not consider advanced collabora…
Photo-identification technique improved for new dolphin individuals.
problem Traditional photo-identification of dolphins is laborious and manual.
method Metric embedding learning using triplet loss function in Euclidean space.
result Compact representation of fin images generalizes well to new identities.
This work aims to reduce inexplicable errors in deep neural networks by obtaining class-level semantics and penalizing misclassifications.
problem Deep neural networks misclassify images, leading to inexplicable errors that can harm trust and societal impact.
method Obtain class-level semantics, propose Weighted Loss Functions (WLFs), and train classifiers with these methods.
result Trained networks have more explicable failure modes and comparable accuracy to existing methods.
In this paper, we propose a novel learning method for image classification called Between-Class learning (BC learning). We generate between-class images by mixing two images belonging to different classes with a random ratio. We then input the mixed image to the model and train the model to output the mixing ratio. BC …
Language models predict inorganic synthesis conditions and temperatures.
problem Limited data and heuristic approaches constrain inorganic synthesis planning.
method Language models without fine-tuning predict precursor conditions and temperatures.
result Language models achieve high accuracy in predicting synthesis conditions and temperatures.
New method improves accuracy of quantized neural networks.
problem Accuracy drop in quantized neural networks, especially MobileNet family.
method Weight equalizing shift scaler, binary shifting to recover output range.
result Top-1 accuracy improved from 0.1% to 69.78% ~ 70.96% in MobileNets.
In image classification, visual separability between different object categories is highly uneven, and some categories are more difficult to distinguish than others. Such difficult categories demand more dedicated classifiers. However, existing deep convolutional neural networks (CNN) are trained as flat N-way classifi…
Paper finds mislabeled instances in various datasets.
problem Mislabeled instances in labeled datasets.
method Non-parametric end-to-end pipeline for finding mislabeled instances.
result Average precision of more than 0.84 for finding top 1% mislabeled instances.
Unified model for prediction and deferral selects top-k entities efficiently.
problem Efficiently selecting top-k entities for deferral in machine learning.
method One-stage Top-k Learning-to-Defer framework with a convex surrogate. result Unified model achieves superior accuracy-cost trade-offs.
BNAS improves neural architecture search with a scalable, fast, and efficient approach.
problem Efficiently searching for optimal neural architectures with high performance and low training time.
method Designing a broad scalable architecture (BCNN) with reinforcement learning and parameter sharing, and developing two variants.
result Significantly reduces training time and achieves state-of-the-art performance on CIFAR-10 and ImageNet.
Work on making classifiers robust against adversarial attacks for top-k predictions.
problem Vulnerability of classifiers to adversarial perturbations, especially for top-k predictions.
method Randomized smoothing to turn any classifier into a robust one, using Gaussian noise.
result Derives a tight robustness in ℓ2 norm for top-k predictions, achieving 62.8% certified top-5 accuracy on ImageNet.
New BNN architecture achieves high accuracy without complex tricks.
problem Training accurate Binary Neural Networks (BNNs) from scratch is challenging.
method Revisited design principles and a simple training strategy.
result BinaryDenseNet achieves high accuracy on ImageNet.
NLCG optimizes DNN training, especially with large mini-batches.
problem Improving convergence speed in large-scale DNN training.
method Stochastic Preconditioned Nonlinear Conjugate Gradient (SP-NLCG) algorithm.
result NLCG improves DNN training accuracy by over 10 percentage points at large mini-batch sizes.
Neural network pruning is an important step in design process of efficient neural networks for edge devices with limited computational power. Pruning is a form of knowledge transfer from the weights of the original network to a smaller target subnetwork. We propose a new method for compute-constrained structured channe…
Neural Architecture Search (NAS) has shown great success in automating the design of neural networks, but the prohibitive amount of computations behind current NAS methods requires further investigations in improving the sample efficiency and the network evaluation cost to get better results in a shorter time. In this …
Improved self-supervised learning on ImageNet achieves top-1 accuracy of 77.1%.
problem Self-supervised ResNets underperform supervised learning on ImageNet.
method ReLICv2 combines explicit invariance loss with contrastive objective over varied data views.
result ReLICv2 achieves 77.1% top-1 accuracy on ImageNet, improving over previous state-of-the-art by 1.5%.
A statistical description and model of individual healthcare expenditures in the US has been developed for measuring value in healthcare. We find evidence that healthcare expenditures are quantifiable as an infusion-diffusion process, which can be thought of intuitively as a steady change in the intensity of treatment …
Improved chemical reaction prediction using augmented NLP models.
problem Predicting chemical reactions from text representations.
method Data augmentation and Transformer architecture for SMILES representation.
result Significantly improved accuracy in predicting chemical reactions.
SCR improves GNN training with consistency regularization.
problem Balancing labeled and unlabeled data in GNNs.
method Introduces two consistency regularization strategies: perturbed predictions and Mean Teacher.
result Enhances various GNNs to achieve better performance.
SemiNAS reduces NAS cost by predicting accuracy of unlabeled architectures.
problem Costly evaluation of architectures limits NAS efficiency.
method SemiNAS uses unlabeled architectures to train an accuracy predictor.
result SemiNAS achieves comparable accuracy with less data.
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 implement…
Data augmentation is a widely used technique in many machine learning tasks, such as image classification, to virtually enlarge the training dataset size and avoid overfitting. Traditional data augmentation techniques for image classification tasks create new samples from the original training data by, for example, fli…
Improved DP-SGD on large models achieves high accuracy on image classification tasks.
problem Differentially private image classification often degrades performance.
method Careful hyper-parameter tuning and signal propagation techniques.
result Achieved 81.4% top-1 accuracy on CIFAR-10 under (8, 10^{-5})-DP.
Big models pretrain and fine-tune for semi-supervised learning on ImageNet.
problem Learning from few labeled examples with a large amount of unlabeled data.
method Unsupervised pretraining of a big ResNet model followed by supervised fine-tuning and distillation.
result 73.9% ImageNet top-1 accuracy with just 1% of labels (≤13 labeled images per class). A new distillation method transfers channel information from teacher to student.
problem Transfer knowledge from teacher to student with fewer parameters and calculations.
method Channel Distillation (CD) and Guided Knowledge Distillation (GKD) with loss decay.
result Achieved 27.68% top-1 error on ImageNet with ResNet18, outperforming state-of-the-art methods.
For the past 5 years, the ILSVRC competition and the ImageNet dataset have attracted a lot of interest from the Computer Vision community, allowing for state-of-the-art accuracy to grow tremendously. This should be credited to the use of deep artificial neural network designs. As these became more complex, the storage,…
Improved neural architecture search through balanced training.
problem Inconsistent ranking of architectures under one-shot training.
method Balanced NAO: balanced training of supernet during search.
result Significant improvements in architecture discovery and performance.
Optimizes identifying top-k items from comparisons with minimal comparisons.
problem Finding the top-k items from pairwise comparisons with a fixed error rate.
method Developed an asymptotically optimal algorithm using primal-dual procedure and adaptive comparison allocation.
result Proves the algorithm is asymptotically optimal for top-k identification.
SaliencyMix augments images with salient patches to improve model generalization.
problem Improving deep learning model generalization through better data augmentation.
method Carefully selects salient patches from images and mixes them with the target image.
result Achieves state-of-the-art top-1 error rates and robustness against adversarial attacks.
Decomposable-Net compresses neural networks without retraining for various sizes.
problem Performance degradation when changing model size after training.
method Decomposes weight matrices via SVD and adjusts ranks for different sizes.
result Maintains and improves performance across multiple model sizes.