Network method of moments arXiv:1202.5101 is an important tool for nonparametric network inferences. However, there has been little investigation on accurate descriptions of the sampling distributions of network moment statistics. In this paper, we present the first higher-order accurate approximation to the sampling CDF of a studentized network moment by Edgeworth expansion. In sharp contrast to classical literature on noiseless U-statistics, we showed that the Edgeworth expansion of a network moment statistic as a noisy U-statistic can achieve higher-order accuracy without non-lattice or smoothness assumptions but just requiring weak regularity conditions. Behind this result is our surprising discovery that the two typically-hated factors in network analysis, namely, sparsity and edge-wise observational errors, jointly play a blessing role, contributing a crucial self-smoothing effect in the network moment statistic and making it analytically tractable. Our assumptions match the minimum requirements in related literature. For practitioners, our empirical Edgeworth expansion is highly accurate and computationally efficient. It is also easy to implement. These were demonstrated by comprehensive simulation studies. We showcase three applications of our results in network inference. We proved, to our knowledge, for the first time that some network bootstraps enjoy higher-order accuracy, and provided theoretical guidance for tuning network sub-sampling. We also derived a one-sample test and Cornish-Fisher confidence interval for any given moment, both with analytical formulation and explicit error rates.

Recently, NLP has seen a surge in the usage of large pre-trained models. Users download weights of models pre-trained on large datasets, then fine-tune the weights on a task of their choice. This raises the question of whether downloading untrusted pre-trained weights can pose a security threat. In this paper, we show that it is possible to construct ``weight poisoning'' attacks where pre-trained weights are injected with vulnerabilities that expose ``backdoors'' after fine-tuning, enabling the attacker to manipulate the model prediction simply by injecting an arbitrary keyword. We show that by applying a regularization method, which we call RIPPLe, and an initialization procedure, which we call Embedding Surgery, such attacks are possible even with limited knowledge of the dataset and fine-tuning procedure. Our experiments on sentiment classification, toxicity detection, and spam detection show that this attack is widely applicable and poses a serious threat. Finally, we outline practical defenses against such attacks. Code to reproduce our experiments is available at https://github.com/neulab/RIPPLe.

Equity research analysts at financial institutions play a pivotal role in capital markets; they provide an efficient conduit between investors and companies' management and facilitate the efficient flow of information from companies, promoting functional and liquid markets. However, previous research in the academic finance and behavioral economics communities has found that analysts' estimates of future company earnings and other financial quantities can be affected by a number of behavioral, incentive-based and discriminatory biases and systematic errors, which can detrimentally affect both investors and public companies. We propose a Bayesian latent variable model for analysts' systematic errors and biases which we use to generate a robust bias-adjusted consensus estimate of company earnings. Experiments using historical earnings estimates data show that our model is more accurate than the consensus average of estimates and other related approaches.

Computer vision algorithms, e.g. for face recognition, favour groups of individuals that are better represented in the training data. This happens because of the generalization that classifiers have to make. It is simpler to fit the majority groups as this fit is more important to overall error. We propose to create a balanced training dataset, consisting of the original dataset plus new data points in which the group memberships are intervened, minorities become majorities and vice versa. We show that current generative adversarial networks are a powerful tool for learning these data points, called contrastive examples. We experiment with the equalized odds bias measure on tabular data as well as image data (CelebA and Diversity in Faces datasets). Contrastive examples allow us to expose correlations between group membership and other seemingly neutral features. Whenever a causal graph is available, we can put those contrastive examples in the perspective of counterfactuals.

In the past twenty years there has been an explosion of the use of graphical models to represent the relationship between a vector of random variables and perform distributed inference which takes advantage of the underlying graphical representations. Bayesian networks (BNs) (Darwiche 2009; Fenton and Neil 2012) are nowadays the most used graphical models, with applications to a wide array of domains and implementation in various software: for instance, the R packages bnlearn by Scutari (2010) and gRain by Højsgaard (2012), among others. However, BNs can only represent symmetric conditional independences which in practical applications may not be fully justified. For this reason, a variety of models that can take into account the asymmetric nature of real-world data have been proposed; for example, context-specific BNs (Boutilier, Friedman, Goldszmidt, and Koller 1996), labeled directed acyclic graphs (Pensar, Nyman, Koski, and Corander 2015) and probabilistic decision graphs (Jaeger, Nielsen, and Silander 2006). Unlike most of its competitors, the chain event graph (CEG) (Collazo, Görgen, and Smith 2018; Smith and Anderson 2008; Riccomagno and Smith 2004, 2009) can capture all (context-specific) conditional independences in a unique graph, obtained by a coalescence over the vertices of an appropriately constructed probability tree, called staged tree.

This article reviews bias-correction models for measurement error of exposure variables in the field of nutritional epidemiology. Measurement error usually attenuates estimated slope towards zero. Due to the influence of measurement error, inference of parameter estimate is conservative and confidence interval of the slope parameter is too narrow. Bias-correction in estimators and confidence intervals are of primary interest. We review the following bias-correction models: regression calibration methods, likelihood based models, missing data models, simulation based methods, nonparametric models and sampling based procedures.

Many important real-world applications involve time-series data with skewed distribution. Compared to conventional imbalance learning problems, the classification of imbalanced time-series data is more challenging due to high dimensionality and high inter-variable correlation. This paper proposes a structure preserving Oversampling method to combat the High-dimensional Imbalanced Time-series classification (OHIT). OHIT first leverages a density-ratio based shared nearest neighbor clustering algorithm to capture the modes of minority class in high-dimensional space. It then for each mode applies the shrinkage technique of large-dimensional covariance matrix to obtain accurate and reliable covariance structure. Finally, OHIT generates the structure-preserving synthetic samples based on multivariate Gaussian distribution by using the estimated covariance matrices. Experimental results on several publicly available time-series datasets (including unimodal and multi-modal) demonstrate the superiority of OHIT against the state-of-the-art oversampling algorithms in terms of F-value, G-mean, and AUC.

We study the problem of learning from aggregate observations where supervision signals are given to sets of instances instead of individual instances, while the goal is still to predict labels of unseen individuals. A well-known example is multiple instance learning (MIL). In this paper, we extend MIL beyond binary classification to other problems such as multiclass classification and regression. We present a probabilistic framework that is applicable to a variety of aggregate observations, e.g., pairwise similarity for classification and mean/difference/rank observation for regression. We propose a simple yet effective method based on the maximum likelihood principle, which can be simply implemented for various differentiable models such as deep neural networks and gradient boosting machines. Experiments on three novel problem settings -- classification via triplet comparison and regression via mean/rank observation indicate the effectiveness of the proposed method.

Conventional machine learning applications in the mobile/IoT setting transmit data to a cloud-server for predictions. Due to cost considerations (power, latency, monetary), it is desirable to minimise device-to-server transmissions. The budget learning (BL) problem poses the learner's goal as minimising use of the cloud while suffering no discernible loss in accuracy, under the constraint that the methods employed be edge-implementable. We propose a new formulation for the BL problem via the concept of bracketings. Concretely, we propose to sandwich the cloud's prediction, $g,$ via functions $h^-, h^+$ from a `simple' class so that $h^- \le g \le h^+$ nearly always. On an instance $x$, if $h^+(x)=h^-(x)$, we leverage local processing, and bypass the cloud. We explore theoretical aspects of this formulation, providing PAC-style learnability definitions; associating the notion of budget learnability to approximability via brackets; and giving VC-theoretic analyses of their properties. We empirically validate our theory on real-world datasets, demonstrating improved performance over prior gating based methods.

We introduce a trainable coded modulation scheme that enables joint optimization of the bit-wise mutual information (BMI) through probabilistic shaping, geometric shaping, bit labeling, and demapping for a specific channel model and for a wide range of signal-to-noise ratios (SNRs). Compared to probabilistic amplitude shaping (PAS), the proposed approach is not restricted to symmetric probability distributions, can be optimized for any channel model, and works with any code rate $k/m$, $m$ being the number of bits per channel use and $k$ an integer within the range from $1$ to $m-1$. The proposed scheme enables learning of a continuum of constellation geometries and probability distributions determined by the SNR. Additionally, the PAS architecture with Maxwell-Boltzmann (MB) as shaping distribution was extended with a neural network (NN) that controls the MB shaping of a quadrature amplitude modulation (QAM) constellation according to the SNR, enabling learning of a continuum of MB distributions for QAM. Simulations were performed to benchmark the performance of the proposed joint probabilistic and geometric shaping scheme on additive white Gaussian noise (AWGN) and mismatched Rayleigh block fading (RBF) channels.