More Like this

1. Network Modification using a Novel Gramian-based Edge Centrality

   https://doi.org/10.1109/CDC40024.2019.9028860  

   Chanekar, Prasad Vilas; Nozari, Erfan; Cortes, Jorge (December 2019, IEEE Conference on Decision and Control)

   null (Ed.)

   Modifying the structure of man-made and natural networked systems has become increasingly feasible due to recent technological advances. This flexibility offers great opportunities to save resources and improve controllability and energy efficiency. In contrast (and dual) to the well-studied optimal actuator placement problem, this work focuses on improving network controllability by adding and/or re-weighting network edges while keeping the actuation structure fixed. First a novel energy-based edge centrality measure is proposed and then its relationship with the gradient (with respect to edge weights) of the trace of the controllability Gramian is rigorously characterized. Finally, a network modification algorithm based on the proposed measure is proposed and its efficacy in terms of computational complexity and controllability enhancement is numerically demonstrated. 

   more » « less
2. Weighting Methods for Rare Event Identification From Imbalanced Datasets

   https://doi.org/10.3389/fdata.2021.715320  

   He, Jia; Cheng, Maggie X. (December 2021, Frontiers in Big Data)

   In machine learning, we often face the situation where the event we are interested in has very few data points buried in a massive amount of data. This is typical in network monitoring, where data are streamed from sensing or measuring units continuously but most data are not for events. With imbalanced datasets, the classifiers tend to be biased in favor of the main class. Rare event detection has received much attention in machine learning, and yet it is still a challenging problem. In this paper, we propose a remedy for the standing problem. Weighting and sampling are two fundamental approaches to address the problem. We focus on the weighting method in this paper. We first propose a boosting-style algorithm to compute class weights, which is proved to have excellent theoretical property. Then we propose an adaptive algorithm, which is suitable for real-time applications. The adaptive nature of the two algorithms allows a controlled tradeoff between true positive rate and false positive rate and avoids excessive weight on the rare class, which leads to poor performance on the main class. Experiments on power grid data and some public datasets show that the proposed algorithms outperform the existing weighting and boosting methods, and that their superiority is more noticeable with noisy data. 

   more » « less
3. A Model Predictive Approach for Voltage Support in Microgrids using Energy Storage Systems

   https://doi.org/10.1109/PESGM46819.2021.9638037  

   Bhujel, Niranjan; Rai, Astha; Hansen, Timothy M.; Tonkoski, Reinaldo; Tamrakar, Ujjwol (January 2021, 2021 IEEE Power & Energy Society General Meeting (PESGM))

   Low voltage microgrid systems are characterized by high sensitivity to both active and reactive power for voltage support. Also, the operational conditions of microgrids connected to active distribution systems are time-varying. Thus, the ideal controller to provide voltage support must be flexible enough to handle technical and operational constraints. This paper proposes a model predictive control (MPC) approach to provide dynamic voltage support using energy storage systems. This approach uses a simplified predictive model of the system along with operational constraints to solve an online finite-horizon optimization problem. Control signals are then computed such that the defined cost function is minimized. By proper selection of MPC weighting parameters, the quality of service provided can be adjusted to achieve the desired performance. A simulation study in Matlab/Simulink validates the proposed approach for a simplified version of a 100 kVA, 208 V microgrid using typical parameters. Results show that performance of the voltage support can be adjusted depending on the choice of weight and constraints of the controller. 

   more » « less
4. Recovering from Biased Data: Can Fairness Constraints Improve Accuracy?

   Blum, Avrim; Stangl, Kevin (June 2020, Symposium on Foundations of Responsible Computing (FORC))

   Multiple fairness constraints have been proposed in the literature, motivated by a range of concerns about how demographic groups might be treated unfairly by machine learning classifiers. In this work we consider a different motivation; learning from biased training data. We posit several ways in which training data may be biased, including having a more noisy or negatively biased labeling process on members of a disadvantaged group, or a decreased prevalence of positive or negative examples from the disadvantaged group, or both. Given such biased training data, Empirical Risk Minimization (ERM) may produce a classifier that not only is biased but also has suboptimal accuracy on the true data distribution. We examine the ability of fairness-constrained ERM to correct this problem. In particular, we find that the Equal Opportunity fairness constraint [Hardt et al., 2016] combined with ERM will provably recover the Bayes optimal classifier under a range of bias models. We also consider other recovery methods including re-weighting the training data, Equalized Odds, and Demographic Parity, and Calibration. These theoretical results provide additional motivation for considering fairness interventions even if an actor cares primarily about accuracy. 

   more » « less
5. Revealing the Structure of Deep Neural Networks via Convex Duality

   Ergen, T.; Pilanci, M. (January 2021, International Conference on Machine Learning (ICML) 2021)

   We study regularized deep neural networks (DNNs) and introduce a convex analytic framework to characterize the structure of the hidden layers. We show that a set of optimal hidden layer weights for a norm regularized DNN training problem can be explicitly found as the extreme points of a convex set. For the special case of deep linear networks, we prove that each optimal weight matrix aligns with the previous layers via duality. More importantly, we apply the same characterization to deep ReLU networks with whitened data and prove the same weight alignment holds. As a corollary, we also prove that norm regularized deep ReLU networks yield spline interpolation for one-dimensional datasets which was previously known only for two-layer networks. Furthermore, we provide closed-form solutions for the optimal layer weights when data is rank-one or whitened. The same analysis also applies to architectures with batch normalization even for arbitrary data. Therefore, we obtain a complete explanation for a recent empirical observation termed Neural Collapse where class means collapse to the vertices of a simplex equiangular tight frame. 

   more » « less