Subscribe to the gold package and get unlimited access to Shamra Academy
Register a new userThe Evolution of Out-of-Distribution Robustness Throughout Fine-Tuning
133
0
0.0
(
0
)
Added by
Anders Johan Andreassen
Publication date
2021
fields
Informatics Engineering
and research's language is
English
Ask ChatGPT about the research
No Arabic abstract
Although machine learning models typically experience a drop in performance on out-of-distribution data, accuracies on in- versus out-of-distribution data are widely observed to follow a single linear trend when evaluated across a testbed of models. Models that are more accurate on the out-of-distribution data relative to this baseline exhibit effective robustness and are exceedingly rare. Identifying such models, and understanding their properties, is key to improving out-of-distribution performance. We conduct a thorough empirical investigation of effective robustness during fine-tuning and surprisingly find that models pre-trained on larger datasets exhibit effective robustness during training that vanishes at convergence. We study how properties of the data influence effective robustness, and we show that it increases with the larger size, more diversity, and higher example difficulty of the dataset. We also find that models that display effective robustness are able to correctly classify 10% of the examples that no other current testbed model gets correct. Finally, we discuss several strategies for scaling effective robustness to the high-accuracy regime to improve the out-of-distribution accuracy of state-of-the-art models.
rate research
Read More
Approaches based on deep neural networks have achieved striking performance when testing data and training data share similar distribution, but can significantly fail otherwise. Therefore, eliminating the impact of distribution shifts between training and testing data is crucial for building performance-promising deep models. Conventional methods assume either the known heterogeneity of training data (e.g. domain labels) or the approximately equal capacities of different domains. In this paper, we consider a more challenging case where neither of the above assumptions holds. We propose to address this problem by removing the dependencies between features via learning weights for training samples, which helps deep models get rid of spurious correlations and, in turn, concentrate more on the true connection between discriminative features and labels. Extensive experiments clearly demonstrate the effectiveness of our method on multiple distribution generalization benchmarks compared with state-of-the-art counterparts. Through extensive experiments on distribution generalization benchmarks including PACS, VLCS, MNIST-M, and NICO, we show the effectiveness of our method compared with state-of-the-art counterparts.
With the recently rapid development in deep learning, deep neural networks have been widely adopted in many real-life applications. However, deep neural networks are also known to have very little control over its uncertainty for unseen examples, which potentially causes very harmful and annoying consequences in practical scenarios. In this paper, we are particularly interested in designing a higher-order uncertainty metric for deep neural networks and investigate its effectiveness under the out-of-distribution detection task proposed by~cite{hendrycks2016baseline}. Our method first assumes there exists an underlying higher-order distribution $mathbb{P}(z)$, which controls label-wise categorical distribution $mathbb{P}(y)$ over classes on the K-dimension simplex, and then approximate such higher-order distribution via parameterized posterior function $p_{theta}(z|x)$ under variational inference framework, finally we use the entropy of learned posterior distribution $p_{theta}(z|x)$ as uncertainty measure to detect out-of-distribution examples. Further, we propose an auxiliary objective function to discriminate against synthesized adversarial examples to further increase the robustness of the proposed uncertainty measure. Through comprehensive experiments on various datasets, our proposed framework is demonstrated to consistently outperform competing algorithms.
Enabling out-of-distribution (OOD) detection for DNNs is critical for their safe and reliable operation in the open world. Unfortunately, current works in both methodology and evaluation focus on rather contrived detection problems, and only consider a coarse level of granularity w.r.t.: 1) the in-distribution (ID) classes, and 2) the OOD datas closeness to the ID data. We posit that such settings may be poor approximations of many real-world tasks that are naturally fine-grained (e.g., bird species classification), and thus the reported detection abilities may be over-estimates. Differently, in this work we make granularity a top priority and focus on fine-grained OOD detection. We start by carefully constructing five novel fine-grained test environments in which existing methods are shown to have difficulties. We then propose a new DNN training algorithm, Mixup Outlier Exposure (MixupOE), which leverages an outlier distribution and principles from vicinal risk minimization. Finally, we perform extensive experiments and analyses in our custom test environments and demonstrate that MixupOE can consistently improve fine-grained detection performance, establishing a strong baseline in these more realistic and challenging OOD detection settings.
Determining whether inputs are out-of-distribution (OOD) is an essential building block for safely deploying machine learning models in the open world. However, previous methods relying on the softmax confidence score suffer from overconfident posterior distributions for OOD data. We propose a unified framework for OOD detection that uses an energy score. We show that energy scores better distinguish in- and out-of-distribution samples than the traditional approach using the softmax scores. Unlike softmax confidence scores, energy scores are theoretically aligned with the probability density of the inputs and are less susceptible to the overconfidence issue. Within this framework, energy can be flexibly used as a scoring function for any pre-trained neural classifier as well as a trainable cost function to shape the energy surface explicitly for OOD detection. On a CIFAR-10 pre-trained WideResNet, using the energy score reduces the average FPR (at TPR 95%) by 18.03% compared to the softmax confidence score. With energy-based training, our method outperforms the state-of-the-art on common benchmarks.
Two crucial requirements for a successful adoption of deep learning (DL) in the wild are: (1) robustness to distributional shifts, and (2) model compactness for achieving efficiency. Unfortunately, efforts towards simultaneously achieving Out-of-Distribution (OOD) robustness and extreme model compactness without sacrificing accuracy have mostly been unsuccessful. This raises an important question: Is the inability to create compact, accurate, and robust deep neural networks (CARDs) fundamental? To answer this question, we perform a large-scale analysis for a range of popular model compression techniques which uncovers several intriguing patterns. Notably, in contrast to traditional pruning approaches (e.g., fine tuning and gradual magnitude pruning), we find that lottery ticket-style pruning approaches can surprisingly be used to create high performing CARDs. Specifically, we are able to create extremely compact CARDs that are dramatically more robust than their significantly larger and full-precision counterparts while matching (or beating) their test accuracy, simply by pruning and/or quantizing. To better understand these differences, we perform sensitivity analysis in the Fourier domain for CARDs trained using different data augmentation methods. Motivated by our analysis, we develop a simple domain-adaptive test-time ensembling approach (CARD-Deck) that uses a gating module to dynamically select an appropriate CARD from the CARD-Deck based on their spectral-similarity with test samples. By leveraging complementary frequency biases of different compressed models, the proposed approach builds a winning hand of CARDs that establishes a new state-of-the-art on CIFAR-10-C accuracies (i.e., 96.8% clean and 92.75% robust) with dramatically better memory usage than their non-compressed counterparts. We also present some theoretical evidences supporting our empirical findings.
suggested questions
Log in to be able to interact and post comments
comments
Fetching comments
Sign in to be able to follow your search criteria
