Despite the rapid progress in multihop question-answering (QA), models still have trouble explaining why an answer is correct, with limited explanation training data available to learn from. To address this, we introduce three explanation datasets in which explanations formed from corpus facts are annotated. Our first dataset, eQASC, contains over 98K explanation annotations for the multihop question answering dataset QASC, and is the first that annotates multiple candidate explanations for each answer. The second dataset eQASC-perturbed is constructed by crowd-sourcing perturbations (while preserving their validity) of a subset of explanations in QASC, to test consistency and generalization of explanation prediction models. The third dataset eOBQA is constructed by adding explanation annotations to the OBQA dataset to test generalization of models trained on eQASC. We show that this data can be used to significantly improve explanation quality (+14% absolute F1 over a strong retrieval baseline) using a BERT-based classifier, but still behind the upper bound, offering a new challenge for future research. We also explore a delexicalized chain representation in which repeated noun phrases are replaced by variables, thus turning them into generalized reasoning chains (for example: "X is a Y" AND "Y has Z" IMPLIES "X has Z"). We find that generalized chains maintain performance while also being more robust to certain perturbations.

When we were invited to write a retrospective article about our AAAI-99 paper on mutual bootstrapping (Riloff and Jones 1999), our first reaction was hesitation because, well, that algorithm seems old and clunky now. But upon reflection, it shaped a great deal of subsequent work on bootstrapped learning for natural language processing, both by ourselves and others. So our second reaction was enthusiasm, for the opportunity to think about the path from 1999 to 2017 and to share the lessons that we learned about bootstrapped learning along the way. This article begins with a brief history of related research that preceded and inspired the mutual bootstrapping work, to position it with respect to that period of time. We then describe the general ideas and approach behind the mutual bootstrapping algorithm.

In a previous article we discussed why it's a good idea to prefer probability models to "hard" classification models, and why you should delay setting "hard" classification rules as long as possible. But decisions have to be made, and eventually you will have to set that threshold. A good threshold balances classifier precision/recall or sensitivity/specificity in a way that best meets the project or business needs. One way to quantify and think about this balance is the notion of model utility, which maps the performance of a model to some notion of the value achieved by that performance. In this article, we demonstrate the use of sigr::model_utility() to estimate model utility and pick model thresholds for classification problems.

Pruning and quantization are widely applied techniques for compressing deep neural networks, often driven by the resource constraints of deploying models to mobile phones or embedded devices (Esteva et al., 2017; Lane & Warden, 2018). To-date, discussion around the relative merits of different compression methods has centered on the tradeoff between level of compression and top-line metrics such as top-1 and top-5 accuracy (Blalock et al., 2020). Along this dimension, compression techniques are remarkably successful. It is possible to prune the majority of weights (Gale et al., 2019; Evci et al., 2019) or heavily quantize the bit representation (Jacob et al., 2017) with negligible decreases to test-set accuracy. However, recent work by Hooker et al. (2019a) has found that the minimal changes to top-line metrics obscure critical differences in generalization between pruned and non-pruned networks. The authors establish that pruning disproportionately impacts predictive performance on a small subset of the dataset. We build upon this work and focus on the implications of these findings for a dataset with sensitive protected attributes such as gender and age. Our work addresses the question: Does compression amplify existing algorithmic bias?

In this paper, we provide a precise characterize of generalization properties of high dimensional kernel ridge regression across the under- and over-parameterized regimes, depending on whether the number of training data $n$ exceeds the feature dimension $d$. By establishing a novel bias-variance decomposition of the expected excess risk, we show that, while the bias is independent of $d$ and monotonically decreases with $n$, the variance depends on $n,d$ and can be unimodal or monotonically decreasing under different regularization schemes. Our refined analysis goes beyond the double descent theory by showing that, depending on the data eigen-profile and the level of regularization, the kernel regression risk curve can be a double-descent-like, bell-shaped, or monotonic function of $n$. Experiments on synthetic and real data are conducted to support our theoretical findings.

With the abundance of industrial datasets, imbalanced classification has become a common problem in several application domains. Oversampling is an effective method to solve imbalanced classification. One of the main challenges of the existing oversampling methods is to accurately label the new synthetic samples. Inaccurate labels of the synthetic samples would distort the distribution of the dataset and possibly worsen the classification performance. This paper introduces the idea of weakly supervised learning to handle the inaccurate labeling of synthetic samples caused by traditional oversampling methods. Graph semi-supervised SMOTE is developed to improve the credibility of the synthetic samples' labels. In addition, we propose cost-sensitive neighborhood components analysis for high dimensional datasets and bootstrap based ensemble framework for highly imbalanced datasets. The proposed method has achieved good classification performance on 8 synthetic datasets and 3 real-world datasets, especially for high imbalance and high dimensionality problems. The average performances and robustness are better than the benchmark methods.

Discriminative deep neural networks (DNNs) do well at classifying input associated with the classes they have been trained on. However, out-of-distribution (OOD) input poses a great challenge to such models and consequently represents a major risk when these models are used in safety-critical systems. In the last two years, extensive research has been performed in the domain of OOD detection. This research has relied mainly on training the model with OOD data or requiring additional computation for OOD detection. Such methods may not be applicable in many real world use cases. In this paper, we propose GLOD -- Gaussian likelihood out of distribution detector -- an extended DNN classifier capable of efficiently detecting OOD samples with no additional runtime overhead and without auxiliary training data. GLOD uses a layer that models the Gaussian density function of the trained classes. The layer outputs are used to estimate a Log-Likelihood Ratio which is employed to detect OOD samples. We evaluate GLOD's detection performance on SVHN, CIFAR-10 and CIFAR-100.

Machine learning techniques have been widely used in natural language processing (NLP). However, as revealed by many recent studies, machine learning models often inherit and amplify the societal biases in data. Various metrics have been proposed to quantify biases in model predictions. In particular, several of them evaluate disparity in model performance between protected groups and advantaged groups in the test corpus. However, we argue that evaluating bias at the corpus level is not enough for understanding how biases are embedded in a model. In fact, a model with similar aggregated performance between different groups on the entire data may behave differently on instances in a local region. To analyze and detect such local bias, we propose LOGAN, a new bias detection technique based on clustering. Experiments on toxicity classification and object classification tasks show that LOGAN identifies bias in a local region and allows us to better analyze the biases in model predictions.

We propose test suite accuracy to approximate semantic accuracy for Text-to-SQL models. Our method distills a small test suite of databases that achieves high code coverage for the gold query from a large number of randomly generated databases. At evaluation time, it computes the denotation accuracy of the predicted queries on the distilled test suite, hence calculating a tight upper-bound for semantic accuracy efficiently. We use our proposed method to evaluate 21 models submitted to the Spider leader board and manually verify that our method is always correct on 100 examples. In contrast, the current Spider metric leads to a 2.5% false negative rate on average and 8.1% in the worst case, indicating that test suite accuracy is needed. Our implementation, along with distilled test suites for eleven Text-to-SQL datasets, is publicly available.

A University of Central Florida researcher is part of a new study showing that artificial intelligence can be nearly as accurate as a physician in diagnosing COVID-19 in the lungs. The study, recently published in Nature Communications, shows the new technique can also overcome some of the challenges of current testing. Researchers demonstrated that an AI algorithm could be trained to classify COVID-19 pneumonia in computed tomography (CT) scans with up to 90 percent accuracy, as well as correctly identify positive cases 84 percent of the time and negative cases 93 percent of the time. CT scans offer a deeper insight into COVID-19 diagnosis and progression as compared to the often-used reverse transcription-polymerase chain reaction, or RT-PCR, tests. These tests have high false negative rates, delays in processing and other challenges.