This paper proposes a new subspace learning method, named Quantized Fisher Discriminant Analysis (QFDA), which makes use of both machine learning and information theory. There is a lack of literature for combination of machine learning and information theory and this paper tries to tackle this gap. QFDA finds a subspace which discriminates the uniformly quantized images in the Discrete Cosine Transform (DCT) domain at least as well as discrimination of non-quantized images by Fisher Discriminant Analysis (FDA) while the images have been compressed. This helps the user to throw away the original images and keep the compressed images instead without noticeable loss of classification accuracy. We propose a cost function whose minimization can be interpreted as rate-distortion optimization in information theory. We also propose quantized Fisherfaces for facial analysis in QFDA.

Fairness is becoming a rising concern w.r.t. machine learning model performance. Especially for sensitive fields such as criminal justice and loan decision, eliminating the prediction discrimination towards a certain group of population (characterized by sensitive features like race and gender) is important for enhancing the trustworthiness of model. In this paper, we present a new general framework to improve machine learning fairness. The goal of our model is to minimize the influence of sensitive feature from the perspectives of both the data input and the predictive model. In order to achieve this goal, we reformulate the data input by removing the sensitive information and strengthen model fairness by minimizing the marginal contribution of the sensitive feature. We propose to learn the non-sensitive input via sampling among features and design an adversarial network to minimize the dependence between the reformulated input and the sensitive information. Extensive experiments on three benchmark datasets suggest that our model achieve better results than related state-of-the-art methods with respect to both fairness metrics and prediction performance.

Neural models for NLP typically use large numbers of parameters to reach state-of-the-art performance, which can lead to excessive memory usage and increased runtime. We present a structure learning method for learning sparse, parameter-efficient NLP models. Our method applies group lasso to rational RNNs (Peng et al., 2018), a family of models that is closely connected to weighted finite-state automata (WFSAs). We take advantage of rational RNNs' natural grouping of the weights, so the group lasso penalty directly removes WFSA states, substantially reducing the number of parameters in the model. Our experiments on a number of sentiment analysis datasets, using both GloVe and BERT embeddings, show that our approach learns neural structures which have fewer parameters without sacrificing performance relative to parameter-rich baselines. Our method also highlights the interpretable properties of rational RNNs. We show that sparsifying such models makes them easier to visualize, and we present models that rely exclusively on as few as three WFSAs after pruning more than 90% of the weights. We publicly release our code.

Ranking problems define a widely spread class of statistical learning problems with many applications, including fraud detection, document ranking or medicine. In this article, we systematically describe different types of ranking problems and investigate existing empirical risk minimization techniques to solve such ranking problems. Furthermore, we discuss whether a Boosting-type algorithm for continuous ranking problems is achievable by using surrogate loss functions.

Chi Thang Duong 1 Hongzhi Yin 2 Thanh Dat Hoang 3 Truong Giang Le Ba 3 Matthias Weidlich 4 Quoc Viet Hung Nguyen 5 Karl Aberer 1 1 EPFL 2 The University of Queensland 3 HUST 5 Griffith University 4 Humboldt-Universit at zu Berlin Abstract Graph embedding aims at learning a vector-based representation of vertices that incorporates the structure of the graph. This representation then enables inference of graph properties. Existing graph embedding techniques, however, do not scale well to large graphs. We therefore propose a framework for parallel computation of a graph embedding using a cluster of compute nodes with resource constraints. We show how to distribute any existing embedding technique by first splitting a graph for any given set of constrained compute nodes and then reconciling the embedding spaces derived for these sub-graphs. We also propose a new way to evaluate the quality of graph embeddings that is independent of a specific inference task. Based thereon, we give a formal bound on the difference between the embeddings derived by centralised and parallel computation. Experimental results illustrate that our approach for parallel computation scales well, while largely maintaining the embedding quality. 1 Introduction Graphs are a natural representation of relations between entities in complex systems, such as social networks or information networks. To enable inference on graphs, a graph embedding may be learned.

Cheolhyeong Kim 1, Haeseong Moon 2, and Hyung Ju Hwang † 1 1 Department of Mathematics, POSTECH 2 Department of Mathematics, University of California, San Diego Abstract Learning graph-structured data with graph neural networks (GNNs) has been recently emerging as an important field because of its wide applicability in bioinformatics, chemoinformatics, social network analysis and data mining. Recent GNN algorithms are based on neural message passing, which enables GNNs to integrate local structures and node features recursively. However, past GNN algorithms based on 1-hop neighborhood neural message passing are exposed to a risk of loss of information on local structures and relationships. In this paper, we propose Neighborhood Edge AggregatoR (NEAR), a novel framework that aggregates relations between the nodes in the neighborhood via edges. NEAR, which can be orthogonally combined with previous GNN algorithms, gives integrated information that describes which nodes in the neighborhood are connected. Therefore, GNNs combined with NEAR reflect each node's local structure beyond the nodes themselves. Experimental results on multiple graph classification tasks show that our algorithm achieves state-of-the-art results. 1 Introduction Interest in learning graph structured data has risen rapidly in recent years because of its wide applicability in bioinformatics, chemoinformatics, social network analysis and data mining. For learning graph-structured data, we need an algorithm that can effectively represent the graph structure and relations between the graph nodes.

Fine-tuning a deep network trained with the standard cross-entropy loss is a strong baseline for few-shot learning. When fine-tuned transductively, this outperforms the current state-of-the-art on standard datasets such as Mini-Imagenet, Tiered-Imagenet, CIFAR-FS and FC-100 with the same hyper-parameters. The simplicity of this approach enables us to demonstrate the first few-shot learning results on the Imagenet-21k dataset. We find that using a large number of meta-training classes results in high few-shot accuracies even for a large number of test classes. We do not advocate our approach as the solution for few-shot learning, but simply use the results to highlight limitations of current benchmarks and few-shot protocols. We perform extensive studies on benchmark datasets to propose a metric that quantifies the "hardness" of a test episode. This metric can be used to report the performance of few-shot algorithms in a more systematic way.

Recent works have shown superiorities of decentralized SGD to centralized counterparts in large-scale machine learning, but their theoretical gap is still not fully understood. In this paper, we propose a decentralized stochastic gradient tracking (DSGT) algorithm over peer-to-peer networks for empirical risk minimization problems, and explicitly evaluate its convergence rate in terms of key parameters of the problem, e.g., algebraic connectivity of the communication network, mini-batch size, and gradient variance. Importantly, it is the first theoretical result that can \emph{exactly} recover the rate of the centralized SGD, and has optimal dependence on the algebraic connectivity of the networks when using stochastic gradients. Moreover, we explicitly quantify how the network affects speedup and the rate improvement over existing works. Interestingly, we also point out for the first time that both linear and sublinear speedup can be possible. We empirically validate DSGT on neural networks and logistic regression problems, and show its advantage over the state-of-the-art algorithms.

The framework of information theoretic learning (ITL) has been verified as a powerful approach for robust machine learning, which improves robustness significantly in regression, feature extraction, dimensionality reduction and so on. Nevertheless, few studies utilize ITL for robust classification. In this study, we attempt to improve the robustness of the logistic regression, a fundamental method in classification, through analyzing the characteristic when the model is affected by outliers. We propose an ITL-based variant that learns by the error distribution, the performance of which is experimentally evaluated on two toy examples and several public datasets, compared with two traditional methods and two states of the art. The results demonstrate that the novel method can outperform the states of the art evidently in some cases, and behaves with desirable potential to achieve better robustness in complex situations than existing methods.

In this paper, we solve the arms exponential exploding issue in multivariate Multi-Armed Bandit (Multivariate-MAB) problem when the arm dimension hierarchy is considered. We propose a framework called path planning (TS-PP) which utilizes decision graph/trees to model arm reward success rate with m-way dimension interaction, and adopts Thompson sampling (TS) for heuristic search of arm selection. Naturally, it is quite straightforward to combat the curse of dimensionality using a serial processes that operates sequentially by focusing on one dimension per each process. For our best acknowledge, we are the first to solve Multivariate-MAB problem using graph path planning strategy and deploying alike Monte-Carlo tree search ideas. Our proposed method utilizing tree models has advantages comparing with traditional models such as general linear regression. Simulation studies validate our claim by achieving faster convergence speed, better efficient optimal arm allocation and lower cumulative regret.