Transfer learning is a widely used method to build high performing computer vision models. In this paper, we study the efficacy of transfer learning by examining how the choice of data impacts performance. We find that more pre-training data does not always help, and transfer performance depends on a judicious choice of pre-training data. These findings are important given the continued increase in dataset sizes. We further propose domain adaptive transfer learning, a simple and effective pre-training method using importance weights computed based on the target dataset. Our method to compute importance weights follow from ideas in domain adaptation, and we show a novel application to transfer learning. Our methods achieve state-of-the-art results on multiple fine-grained classification datasets and are well-suited for use in practice.
Deep neural networks often work well when they are over-parameterized and trained with a massive amount of noise and regularization, such as weight decay and dropout. Although dropout is widely used as a regularization technique for fully connected layers, it is often less effective for convolutional layers. This lack of success of dropout for convolutional layers is perhaps due to the fact that activation units in convolutional layers are spatially correlated so information can still flow through convolutional networks despite dropout. Thus a structured form of dropout is needed to regularize convolutional networks. In this paper, we introduce DropBlock, a form of structured dropout, where units in a contiguous region of a feature map are dropped together. We found that applying DropbBlock in skip connections in addition to the convolution layers increases the accuracy. Also, gradually increasing number of dropped units during training leads to better accuracy and more robust to hyperparameter choices. Extensive experiments show that DropBlock works better than dropout in regularizing convolutional networks. On ImageNet classification, ResNet-50 architecture with DropBlock achieves $78.13\%$ accuracy, which is more than $1.6\%$ improvement on the baseline. On COCO detection, DropBlock improves Average Precision of RetinaNet from $36.8\%$ to $38.4\%$.
We prove that as the model predictions on the training set approach the true conditional distribution of labels given inputs, the noise inherent in minibatch gradients causes the stationary distribution of natural gradient descent to approach a Bayesian posterior near local minima as the learning rate $\epsilon \rightarrow 0$. The temperature $T \approx \epsilon N/(2B)$ of this posterior is controlled by the learning rate, training set size $N$ and batch size $B$. However minibatch NGD is not parameterisation invariant, and we therefore introduce "stochastic natural gradient descent", which preserves parameterisation invariance by introducing a multiplicative bias to the stationary distribution. We identify this bias as the well known Jeffreys prior. To support our claims, we show that the distribution of samples from NGD is close to the Laplace approximation to the posterior when $T = 1$. Furthermore, the test loss of ensembles drawn using NGD falls rapidly as we increase the batch size until $B \approx \epsilon N/2$, while above this point the test loss is constant or rises slowly.
In this paper, we take a closer look at data augmentation for images, and describe a simple procedure called AutoAugment to search for improved data augmentation policies. Our key insight is to create a search space of data augmentation policies, evaluating the quality of a particular policy directly on the dataset of interest. 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 83.54%. On CIFAR-10, we achieve an error rate of 1.48%, which is 0.65% better than the previous state-of-the-art. Finally, policies learned from one dataset can be transferred to work well on other similar datasets. For example, the policy learned on ImageNet allows us to achieve state-of-the-art accuracy on the fine grained visual classification dataset Stanford Cars, without fine-tuning weights pre-trained on additional data. Code to train Wide-ResNet, Shake-Shake and ShakeDrop models with AutoAugment policies can be found at https://github.com/tensorflow/models/tree/master/research/autoaugment
Unsupervised representation learning algorithms such as word2vec and ELMo improve the accuracy of many supervised NLP models, mainly because they can take advantage of large amounts of unlabeled text. However, the supervised models only learn from task-specific labeled data during the main training phase. We therefore propose Cross-View Training (CVT), a semi-supervised learning algorithm that improves the representations of a Bi-LSTM sentence encoder using a mix of labeled and unlabeled data. On labeled examples, standard supervised learning is used. On unlabeled examples, CVT teaches auxiliary prediction modules that see restricted views of the input (e.g., only part of a sentence) to match the predictions of the full model seeing the whole input. Since the auxiliary modules and the full model share intermediate representations, this in turn improves the full model. Moreover, we show that CVT is particularly effective when combined with multi-task learning. We evaluate CVT on five sequence tagging tasks, machine translation, and dependency parsing, achieving state-of-the-art results.
Designing convolutional neural networks (CNN) models for mobile devices is challenging because mobile models need to be small and fast, yet still accurate. Although significant effort has been dedicated to design and improve mobile models on all three dimensions, it is challenging to manually balance these trade-offs when there are so many architectural possibilities to consider. In this paper, we propose an automated neural architecture search approach for designing resource-constrained mobile CNN models. We propose to explicitly incorporate latency information into the main objective so that the search can identify a model that achieves a good trade-off between accuracy and latency. Unlike in previous work, where mobile latency is considered via another, often inaccurate proxy (e.g., FLOPS), in our experiments, we directly measure real-world inference latency by executing the model on a particular platform, e.g., Pixel phones. To further strike the right balance between flexibility and search space size, we propose a novel factorized hierarchical search space that permits layer diversity throughout the network. Experimental results show that our approach consistently outperforms state-of-the-art mobile CNN models across multiple vision tasks. On the ImageNet classification task, our model achieves 74.0% top-1 accuracy with 76ms latency on a Pixel phone, which is 1.5x faster than MobileNetV2 (Sandler et al. 2018) and 2.4x faster than NASNet (Zoph et al. 2018) with the same top-1 accuracy. On the COCO object detection task, our model family achieves both higher mAP quality and lower latency than MobileNets.
Deep reinforcement learning has achieved many recent successes, but our understanding of its strengths and limitations is hampered by the lack of rich environments in which we can fully characterize optimal behavior, and correspondingly diagnose individual actions against such a characterization. Here we consider a family of combinatorial games, arising from work of Erdos, Selfridge, and Spencer, and we propose their use as environments for evaluating and comparing different approaches to reinforcement learning. These games have a number of appealing features: they are challenging for current learning approaches, but they form (i) a low-dimensional, simply parametrized environment where (ii) there is a linear closed form solution for optimal behavior from any state, and (iii) the difficulty of the game can be tuned by changing environment parameters in an interpretable way. We use these Erdos-Selfridge-Spencer games not only to compare different algorithms, but test for generalization, make comparisons to supervised learning, analyse multiagent play, and even develop a self play algorithm. Code can be found at: https://github.com/rubai5/ESS_Game
Despite recent advances in training recurrent neural networks (RNNs), capturing long-term dependencies in sequences remains a fundamental challenge. Most approaches use backpropagation through time (BPTT), which is difficult to scale to very long sequences. This paper proposes a simple method that improves the ability to capture long term dependencies in RNNs by adding an unsupervised auxiliary loss to the original objective. This auxiliary loss forces RNNs to either reconstruct previous events or predict next events in a sequence, making truncated backpropagation feasible for long sequences and also improving full BPTT. We evaluate our method on a variety of settings, including pixel-by-pixel image classification with sequence lengths up to 16\,000, and a real document classification benchmark. Our results highlight good performance and resource efficiency of this approach over competitive baselines, including other recurrent models and a comparable sized Transformer. Further analyses reveal beneficial effects of the auxiliary loss on optimization and regularization, as well as extreme cases where there is little to no backpropagation.
Commonsense reasoning is a long-standing challenge for deep learning. For example, it is difficult to use neural networks to tackle the Winograd Schema dataset~\cite{levesque2011winograd}. In this paper, we present a simple method for commonsense reasoning with neural networks, using unsupervised learning. Key to our method is the use of language models, trained on a massive amount of unlabled data, to score multiple choice questions posed by commonsense reasoning tests. On both Pronoun Disambiguation and Winograd Schema challenges, our models outperform previous state-of-the-art methods by a large margin, without using expensive annotated knowledge bases or hand-engineered features. We train an array of large RNN language models that operate at word or character level on LM-1-Billion, CommonCrawl, SQuAD, Gutenberg Books, and a customized corpus for this task and show that diversity of training data plays an important role in test performance. Further analysis also shows that our system successfully discovers important features of the context that decide the correct answer, indicating a good grasp of commonsense knowledge.
Transfer learning has become a cornerstone of computer vision with the advent of ImageNet features, yet little work has been done to evaluate the performance of ImageNet architectures across different datasets. An implicit hypothesis in modern computer vision research is that models that perform better on ImageNet necessarily perform better on other vision tasks. However, this hypothesis has never been systematically tested. Here, we compare the performance of 13 classification models on 12 image classification tasks in three settings: as fixed feature extractors, fine-tuned, and trained from random initialization. We find that, when networks are used as fixed feature extractors, ImageNet accuracy is only weakly predictive of accuracy on other tasks ($r^2=0.24$). In this setting, ResNets consistently outperform networks that achieve higher accuracy on ImageNet. When networks are fine-tuned, we observe a substantially stronger correlation ($r^2 = 0.86$). We achieve state-of-the-art performance on eight image classification tasks simply by fine-tuning state-of-the-art ImageNet architectures, outperforming previous results based on specialized methods for transfer learning. Finally, we observe that, on three small fine-grained image classification datasets, networks trained from random initialization perform similarly to ImageNet-pretrained networks. Together, our results show that ImageNet architectures generalize well across datasets, with small improvements in ImageNet accuracy producing improvements across other tasks, but ImageNet features are less general than previously suggested.