We take an information-theoretic approach to identify nonlinear feature redundancies in unsupervised learning. We define a subset of features as sufficiently-informative when the joint entropy of all the input features equals to that of the chosen subset. We argue that the rest of the features are redundant as all the accessible information about the data can be captured from sufficiently-informative features. Next, instead of directly estimating the entropy, we propose a Fourier-based characterization. For that, we develop a novel Fourier expansion on the Boolean cube incorporating correlated random variables. This generalization of the standard Fourier analysis is beyond product probability spaces. Based on our Fourier framework, we propose a measure of redundancy for features in the unsupervised settings. We then, consider a variant of this measure with a search algorithm to reduce its computational complexity as low as with being the number of samples and the number of features. Besides the theoretical justifications, we test our method on various real-world and synthetic datasets. Our numerical results demonstrate that the proposed method outperforms state-of-the-art feature selection techniques.
Resource Category: AI/ML
A fundamental obstacle in learning information from data is the presence of nonlinear redundancies and dependencies in it. To address this, we propose a Fourier-based approach to extract relevant information in the supervised setting. We first develop a novel Fourier expansion for functions of correlated binary random variables. This is a generalization of the standard Fourier expansion on the Boolean cube beyond product probability spaces. We further extend our Fourier analysis to stochastic mappings. As an important application of this analysis, we investigate learning with feature subset selection. We reformulate this problem in the Fourier domain, and introduce a computationally efficient measure for selecting features. Bridging the Bayesian error rate with the Fourier coefficients, we demonstrate that the Fourier expansion provides a powerful tool to characterize nonlinear dependencies in the features-label relation. Via theoretical analysis, we show that our proposed measure finds provably asymptotically optimal feature subsets. Lastly, we present an algorithm based on our measure and verify our findings via numerical experiments on various datasets.
In temporal ordered clustering , given a single snapshot of a dynamic network in which nodes arrive at distinct time instants, we aim at partitioning its nodes into K ordered clusters C_1≺⋯≺C_K such that for i<j , nodes in cluster C_i arrived before nodes in cluster C_j , with K being a data-driven parameter and not known upfront. Such a problem is of considerable significance in many applications ranging from tracking the expansion of fake news to mapping the spread of information. We first formulate our problem for a general dynamic graph, and propose an integer programming framework that finds the optimal clustering, represented as a strict partial order set, achieving the best precision (i.e., fraction of successfully ordered node pairs) for a fixed density (i.e., fraction of comparable node pairs). We then develop a sequential importance procedure and design unsupervised and semi-supervised algorithms to find temporal ordered clusters that efficiently approximate the optimal solution. To illustrate the techniques, we apply our methods to the vertex copying (duplication-divergence) model which exhibits some edge-case challenges in inferring the clusters as compared to other network models. Finally, we validate the performance of the proposed algorithms on synthetic and real-world networks.
Short-term probabilistic forecasts of the trajectory of the COVID-19 pandemic in the United States have served as a visible and important communication channel between the scientific modeling community and both the general public and decision-makers. Forecasting models provide specific, quantitative, and evaluable predictions that inform short-term decisions such as healthcare staffing needs, school closures, and allocation of medical supplies. In 2020, the COVID-19 Forecast Hub collected, disseminated, and synthesized hundreds of thousands of specific predictions from more than 50 different academic, industry, and independent research groups. This manuscript systematically evaluates 23 models that regularly submitted forecasts of reported weekly incident COVID-19 mortality counts in the US at the state and national level. One of these models was a multi-model ensemble that combined all available forecasts each week. The performance of individual models showed high variability across time, geospatial units, and forecast horizons. Half of the models evaluated showed better accuracy than a naïve baseline model. In combining the forecasts from all teams, the ensemble showed the best overall probabilistic accuracy of any model. Forecast accuracy degraded as models made predictions farther into the future, with probabilistic accuracy at a 20-week horizon more than 5 times worse than when predicting at a 1-week horizon. This project underscores the role that collaboration and active coordination between governmental public health agencies, academic modeling teams, and industry partners can play in developing modern modeling capabilities to support local, state, and federal response to outbreaks.
Controlling recurrent infectious diseases is a vital yet complicated problem in global health. During the long period of time from patients becoming infected to finally seeking treatment, their close contacts are exposed and vulnerable to the disease they carry. Active screening (or case finding) methods seek to actively discover undiagnosed cases by screening contacts of known infected people to reduce the spread of the disease. Existing practice of active screening methods often screen all contacts of an infected person, requiring a large budget. In cooperation with a research institute in India, we develop a model of the active screening problem and present a software agent, REMEDY. This agent assists maximizing effectiveness of active screening under real world budgetary constraints and limited contact information. Our contributions are: (1) A new approach to modeling multi-round network-based screening/contact tracing under uncertainty and proof of its NP-hardness; (2) Two novel algorithms, Full- and Fast-REMEDY. Full-REMEDY considers the effect of future actions and provides high solution quality, whereas Fast-REMEDY scales linearly in the size of the network; (3) Evaluation of Full- and Fast-REMEDY on several real-world datasets which emulate human contact to show that they control diseases better than the baselines. We also show that the software agent is robust to errors in estimates of disease parameters, and incomplete information of the contact network. Our software agent is currently under review before deployment as a means to improve the efficiency of district-wise active screening for tuberculosis in India.