Ensembling can improve the performance of Neural Networks, but existing approaches struggle when the architecture likelihood surface has dispersed, narrow peaks. Furthermore, existing methods construct equally weighted ensembles, and this is likely to be vulnerable to the failure modes of the weaker architectures. By viewing ensembling as approximately marginalising over architectures we construct ensembles using the tools of Bayesian Quadrature -- tools which are well suited to the exploration of likelihood surfaces with dispersed, narrow peaks. Additionally, the resulting ensembles consist of architectures weighted commensurate with their performance. We show empirically -- in terms of test likelihood, accuracy, and expected calibration error -- that our method outperforms state-of-the-art baselines, and verify via ablation studies that its components do so independently.

-- We hypothesize that similar objects should have similar outlier scores. To our knowledge, all existing outlier detectors calculate the outlier score for each object independently regardless of the outlier scores of the other objects. Therefore, they do not guarantee that similar objects have similar outlier scores. To verify our proposed hypothesis, we propose an outlier score post-processing technique for outlier detectors, called neighborhood averaging (NA), which pays attention to objects and their neighbors and guarantees them to have more similar outlier scores than their original scores. Given an object and its outlier score from any outlier detector, NA modifies its outlier score by combining it with its k nearest neighbors' scores. We demonstrate the effectivity of NA by using the well-known k-nearest neighbors (k-NN). Experimental results show that NA improves all 10 tested baseline detectors by 13% (from 0.70 to 0.79 AUC) on average evaluated on nine real-world datasets. Moreover, even outlier detectors that are already based on k-NN are also improved. The experiments also show that in some applications, the choice of detector is no more significant when detectors are jointly used with NA, which may pose a challenge to the generally considered idea that the data model is the most important factor. Outliers are objects that significantly deviate from other objects. Outliers can indicate useful information, which can be applied in applications such as fraud detection [1, 2], abnormal time series [3, 4], and traffic patterns [5, 6]. Outliers can also be harmful because they are generally unwanted, can be considered errors, and may have biased statistical analysis for applications like clustering [7, 8]. Recently, outlier detection has also been applied to manufacturing data [9] and industrial applications [10]. For these reasons, outliers need to be detected. Most outlier detectors calculate the so-called outlier score for every object independently and then calculate the threshold scores that deviate significantly from the others and label them as outliers [11].

Subgraph-wise sampling -- a promising class of mini-batch training techniques for graph neural networks (GNNs -- is critical for real-world applications. During the message passing (MP) in GNNs, subgraph-wise sampling methods discard messages outside the mini-batches in backward passes to avoid the well-known neighbor explosion problem, i.e., the exponentially increasing dependencies of nodes with the number of MP iterations. However, discarding messages may sacrifice the gradient estimation accuracy, posing significant challenges to their convergence analysis and convergence speeds. To address this challenge, we propose a novel subgraph-wise sampling method with a convergence guarantee, namely Local Message Compensation (LMC). To the best of our knowledge, LMC is the first subgraph-wise sampling method with provable convergence. The key idea is to retrieve the discarded messages in backward passes based on a message passing formulation of backward passes. By efficient and effective compensations for the discarded messages in both forward and backward passes, LMC computes accurate mini-batch gradients and thus accelerates convergence. Moreover, LMC is applicable to various MP-based GNN architectures, including convolutional GNNs (finite message passing iterations with different layers) and recurrent GNNs (infinite message passing iterations with a shared layer). Experiments on large-scale benchmarks demonstrate that LMC is significantly faster than state-of-the-art subgraph-wise sampling methods.

However, it is often a laborious process that involves While much effort has been devoted to deriving and analyzing manually designing suitable convex objectives and associated effective convex formulations of signal processing problems, convex constraints. Perhaps more important than the the gradients of convex functions also have critical applications objective function itself is the gradient of the function, since ranging from gradient-based optimization to optimal most convex problems are solved using computationally frugal transport. Recent works have explored data-driven methods gradient-based methods. Monotone gradient maps of convex for learning convex objective functions, but learning their functions also have critical applications in domains including monotone gradients is seldom studied. In this work, we propose gradient-based optimization, generalized linear models, C-MGN and M-MGN, two monotone gradient neural network linear inverse problems, and optimal transport. Therefore, architectures for directly learning the gradients of convex in this work, we propose to learn the gradient of convex functions. We show that, compared to state of the art functions in a data-driven manner using deep learning. Our methods, our networks are easier to train, learn monotone approach is a fundamental step toward blending strengths of gradient fields more accurately, and use significantly fewer both deep learning and convex optimization and offers a wide parameters. We further demonstrate their ability to learn optimal array of applications in data science and signal processing.

Linear $L_1$-regularized models have remained one of the simplest and most effective tools in data analysis, especially in information retrieval problems where n-grams over text with TF-IDF or Okapi feature values are a strong and easy baseline. Over the past decade, screening rules have risen in popularity as a way to reduce the runtime for producing the sparse regression weights of $L_1$ models. However, despite the increasing need of privacy-preserving models in information retrieval, to the best of our knoweledge, no differentially private screening rule exists. In this paper, we develop the first differentially private screening rule for linear and logistic regression. In doing so, we discover difficulties in the task of making a useful private screening rule due to the amount of noise added to ensure privacy. We provide theoretical arguments and experimental evidence that this difficulty arises from the screening step itself and not the private optimizer. Based on our results, we highlight that developing an effective private $L_1$ screening method is an open problem in the differential privacy literature.

We present our work on Track 2 in the Dialog System Technology Challenges 11 (DSTC11). DSTC11-Track2 aims to provide a benchmark for zero-shot, cross-domain, intent-set induction. In the absence of in-domain training dataset, robust utterance representation that can be used across domains is necessary to induce users' intentions. To achieve this, we leveraged a multi-domain dialogue dataset to fine-tune the language model and proposed extracting Verb-Object pairs to remove the artifacts of unnecessary information. Furthermore, we devised the method that generates each cluster's name for the explainability of clustered results. Our approach achieved 3rd place in the precision score and showed superior accuracy and normalized mutual information (NMI) score than the baseline model on various domain datasets.

In the academic world, the number of scientists grows every year and so does the number of authors sharing the same names. Consequently, it challenging to assign newly published papers to their respective authors. Therefore, Author Name Ambiguity (ANA) is considered a critical open problem in digital libraries. This paper proposes an Author Name Disambiguation (AND) approach that links author names to their real-world entities by leveraging their co-authors and domain of research. To this end, we use data collected from the DBLP repository that contains more than 5 million bibliographic records authored by around 2.6 million co-authors. Our approach first groups authors who share the same last names and same first name initials. The author within each group is identified by capturing the relation with his/her co-authors and area of research, represented by the titles of the validated publications of the corresponding author. To this end, we train a neural network model that learns from the representations of the co-authors and titles. We validated the effectiveness of our approach by conducting extensive experiments on a large dataset.

Billions of distributed, heterogeneous and resource constrained IoT devices deploy on-device machine learning (ML) for private, fast and offline inference on personal data. On-device ML is highly context dependent, and sensitive to user, usage, hardware and environment attributes. This sensitivity and the propensity towards bias in ML makes it important to study bias in on-device settings. Our study is one of the first investigations of bias in this emerging domain, and lays important foundations for building fairer on-device ML. We apply a software engineering lens, investigating the propagation of bias through design choices in on-device ML workflows. We first identify reliability bias as a source of unfairness and propose a measure to quantify it. We then conduct empirical experiments for a keyword spotting task to show how complex and interacting technical design choices amplify and propagate reliability bias. Our results validate that design choices made during model training, like the sample rate and input feature type, and choices made to optimize models, like light-weight architectures, the pruning learning rate and pruning sparsity, can result in disparate predictive performance across male and female groups. Based on our findings we suggest low effort strategies for engineers to mitigate bias in on-device ML.

Molecular fingerprints are widely used for predicting chemical properties, and selecting appropriate fingerprints is important. We generate new fingerprints based on the assumption that a performance of prediction using a more effective fingerprint is better. We generate effective interaction fingerprints that are the product of multiple base fingerprints. It is difficult to evaluate all combinations of interaction fingerprints because of computational limitations. Against this problem, we transform a problem of searching more effective interaction fingerprints into a quadratic unconstrained binary optimization problem. In this study, we found effective interaction fingerprints using QM9 dataset.

Tensor completion is an important problem in modern data analysis. In this work, we investigate a specific sampling strategy, referred to as tubal sampling. We propose two novel non-convex tensor completion frameworks that are easy to implement, named tensor $L_1$-$L_2$ (TL12) and tensor completion via CUR (TCCUR). We test the efficiency of both methods on synthetic data and a color image inpainting problem. Empirical results reveal a trade-off between the accuracy and time efficiency of these two methods in a low sampling ratio. Each of them outperforms some classical completion methods in at least one aspect.