Search for: All records

The noise transition matrix plays a central role in the problem of learning with noisy labels. Among many other reasons, a large number of existing solutions rely on access to it. Identifying and estimating the transition matrix without ground truth labels is a critical and challenging task. When label noise transition depends on each instance, the problem of identifying the instance-dependent noise transition matrix becomes substantially more challenging. Despite recent works proposing solutions for learning from instance-dependent noisy labels, the field lacks a unified understanding of when such a problem remains identifiable. The goal of this paper is to characterize the identifiability of the label noise transition matrix. Building on Kruskal's identifiability results, we are able to show the necessity of multiple noisy labels in identifying the noise transition matrix for the generic case at the instance level. We further instantiate the results to explain the successes of the state-of-the-art solutions and how additional assumptions alleviated the requirement of multiple noisy labels. Our result also reveals that disentangled features are helpful in the above identification task and we provide empirical evidence. 

Designing robust loss functions is popular in learning with noisy labels while existing designs did not explicitly consider the overfitting property of deep neural networks (DNNs). As a result, applying these losses may still suffer from overfitting/memorizing noisy labels as training proceeds. In this paper, we first theoretically analyze the memorization effect and show that a lower-capacity model may perform better on noisy datasets. However, it is non-trivial to design a neural network with the best capacity given an arbitrary task. To circumvent this dilemma, instead of changing the model architecture, we decouple DNNs into an encoder followed by a linear classifier and propose to restrict the function space of a DNN by a representation regularizer. Particularly, we require the distance between two self-supervised features to be positively related to the distance between the corresponding two supervised model outputs. Our proposed framework is easily extendable and can incorporate many other robust loss functions to further improve performance. Extensive experiments and theoretical analyses support our claims. Code is available at https://github.com/UCSC-REAL/SelfSup_NoisyLabel. 

The generalization ability of machine learning models degrades significantly when the test distribution shifts away from the training distribution. We investigate the problem of training models that are robust to shifts caused by changes in the distribution of class-priors or group-priors. The presence of skewed training priors can often lead to the models overfitting to spurious features. Unlike existing methods, which optimize for either the worst or the average performance over classes or groups, our work is motivated by the need for finer control over the robustness properties of the model. We present an extremely lightweight post-hoc approach that performs scaling adjustments to predictions from a pre-trained model, with the goal of minimizing a distributionally robust loss around a chosen target distribution. These adjustments are computed by solving a constrained optimization problem on a validation set and applied to the model during test time. Our constrained optimization objective is inspired from a natural notion of robustness to controlled distribution shifts. Our method comes with provable guarantees and empirically makes a strong case for distributional robust post-hoc classifiers. An empirical implementation is available at https://github.com/weijiaheng/Drops. 

Strictly proper scoring rules (SPSR) are incentive compatible for eliciting information about random variables from strategic agents when the principal can reward agents after the realization of the random variables. They also quantify the quality of elicited information, with more accurate predictions receiving higher scores in expectation. In this paper, we extend such scoring rules to settings where a principal elicits private probabilistic beliefs but only has access to agents’ reports. We name our solution Surrogate Scoring Rules (SSR). SSR is built on a bias correction step and an error rate estimation procedure for a reference answer defined using agents’ reports. We show that, with a little information about the prior distribution of the random variables, SSR in a multi-task setting recover SPSR in expectation, as if having access to the ground truth. Therefore, a salient feature of SSR is that they quantify the quality of information despite the lack of ground truth, just as SPSR do for the setting with ground truth. As a by-product, SSR induce dominant uniform strategy truthfulness in reporting. Our method is verified both theoretically and empirically using data collected from real human forecasters. 

Many publications on COVID-19 were released on preprint servers such as medRxiv and bioRxiv. It is unknown how reliable these preprints are, and which ones will eventually be published in scientific journals. In this study, we use crowdsourced human forecasts to predict publication outcomes and future citation counts for a sample of 400 preprints with high Altmetric score. Most of these preprints were published within 1 year of upload on a preprint server (70%), with a considerable fraction (45%) appearing in a high-impact journal with a journal impact factor of at least 10. On average, the preprints received 162 citations within the first year. We found that forecasters can predict if preprints will be published after 1 year and if the publishing journal has high impact. Forecasts are also informative with respect to Google Scholar citations within 1 year of upload on a preprint server. For both types of assessment, we found statistically significant positive correlations between forecasts and observed outcomes. While the forecasts can help to provide a preliminary assessment of preprints at a faster pace than traditional peer-review, it remains to be investigated if such an assessment is suited to identify methodological problems in preprints. 

Wisdom of the crowd (Surowiecki, 2005a) disclosed a striking fact that the majority voting answer from a crowd is usually more accurate than a few individual experts. The same story is observed in machine learning - ensemble methods (Dietterich, 2000) leverage this idea to exploit multiple machine learning algorithms in various settings e.g., supervised learning and semi-supervised learning to achieve better performance by aggregating the predictions of different algorithms than that obtained from any constituent algorithm alone. Nonetheless, the existing aggregating rule would fail when the majority answer of all the constituent algorithms is more likely to be wrong. In this paper, we extend the idea proposed in Bayesian Truth Serum (Prelec, 2004) that “a surprisingly more popular answer is more likely to be the true answer instead of the majority one” to supervised classification further improved by ensemble final predictions method and semi-supervised classification (e.g., MixMatch (Berthelot et al., 2019)) enhanced by ensemble data augmentations method. The challenge for us is to define or detect when an answer should be considered as being “surprising”. We present two machine learning aided methods which can reveal the truth when the minority instead of majority has the true answer on both settings of supervised and semi-supervised classification problems. We name our proposed method the Machine Truth Serum. Our experiments on a set of classification tasks (image, text, etc.) show that the classification performance can be further improved by applying Machine Truth Serum in the ensemble final predictions step (supervised) and in the ensemble data augmentations step (semi-supervised). 

Label noise in real-world datasets encodes wrong correlation patterns and impairs the generalization of deep neural networks (DNNs). It is critical to find efficient ways to detect corrupted patterns. Current methods primarily focus on designing robust training techniques to prevent DNNs from memorizing corrupted patterns. These approaches often require customized training processes and may overfit corrupted patterns, leading to a performance drop in detection. In this paper, from a more data-centric perspective, we propose a training-free solution to detect corrupted labels. Intuitively, ``closer'' instances are more likely to share the same clean label. Based on the neighborhood information, we propose two methods: the first one uses ``local voting" via checking the noisy label consensuses of nearby features. The second one is a ranking-based approach that scores each instance and filters out a guaranteed number of instances that are likely to be corrupted. We theoretically analyze how the quality of features affects the local voting and provide guidelines for tuning neighborhood size. We also prove the worst-case error bound for the ranking-based method. Experiments with both synthetic and real-world label noise demonstrate our training-free solutions consistently and significantly improve most of the training-based baselines. 

In label-noise learning, estimating the transition matrix is a hot topic as the matrix plays an important role in building statistically consistent classifiers. Traditionally, the transition from clean labels to noisy labels (i.e., clean-label transition matrix (CLTM)) has been widely exploited to learn a clean label classifier by employing the noisy data. Motivated by that classifiers mostly output Bayes optimal labels for prediction, in this paper, we study to directly model the transition from Bayes optimal labels to noisy labels (i.e., Bayes-label transition matrix (BLTM)) and learn a classifier to predict Bayes optimal labels. Note that given only noisy data, it is ill-posed to estimate either the CLTM or the BLTM. But favorably, Bayes optimal labels have less uncertainty compared with the clean labels, i.e., the class posteriors of Bayes optimal labels are one-hot vectors while those of clean labels are not. This enables two advantages to estimate the BLTM, i.e., (a) a set of examples with theoretically guaranteed Bayes optimal labels can be collected out of noisy data; (b) the feasible solution space is much smaller. By exploiting the advantages, we estimate the BLTM parametrically by employing a deep neural network, leading to better generalization and superior classification performance. 

In this paper, we introduce DuelGAN, a generative adversarial network (GAN) solution to improve the stability of the generated samples and to mitigate mode collapse. Built upon the Vanilla GAN’s two-player game between the discriminator D1 and the generator G, we introduce a peer discriminator D2 to the min-max game. Similar to previous work using two discriminators, the first role of both D1, D2 is to distinguish between generated samples and real ones, while the generator tries to generate high-quality samples which are able to fool both discriminators. Different from existing methods, we introduce a duel between D1 and D2 to discourage their agreement and therefore increase the level of diversity of the generated samples. This property alleviates the issue of early mode collapse by preventing D1 and D2 from converging too fast. We provide theoretical analysis for the equilibrium of the min-max game formed among G,D1,D2. We offer convergence behavior of DuelGAN as well as stability of the min-max game. It’s worth mentioning that DuelGAN operates in the unsupervised setting, and the duel between D1 and D2 does not need any label supervision. Experiments results on a synthetic dataset and on real-world image datasets (MNIST, Fashion MNIST, CIFAR-10, STL-10, CelebA, VGG) demonstrate that DuelGAN outperforms competitive baseline work in generating diverse and high-quality samples, while only introduces negligible computation cost. Our code is publicly available at https://github.com/UCSC-REAL/DuelGAN.