arXiv subjects

Payman Arabshahi

Publications and source records attributed to Payman Arabshahi.

10 recordsLinked to original sources

arXiv2024

Removing Spurious Correlation from Neural Network Interpretations

The existing algorithms for identification of neurons responsible for undesired and harmful behaviors do not consider the effects of confounders such as topic of the conversation. In this work, we show that confounders can create spurious correlations and propose a new causal mediation approach that controls the impact of the topic. In experiments with two large language models, we study the localization hypothesis and show that adjusting for the effect of conversation topic, toxicity becomes less localized.

cs.CL↗

arXiv2024

Fast Training Dataset Attribution via In-Context Learning

We investigate the use of in-context learning and prompt engineering to estimate the contributions of training data in the outputs of instruction-tuned large language models (LLMs). We propose two novel approaches: (1) a similarity-based approach that measures the difference between LLM outputs with and without provided context, and (2) a mixture distribution model approach that frames the problem of identifying contribution scores as a matrix factorization task. Our empirical comparison demonstrates that the mixture model approach is more robust to retrieval noise in in-context learning, providing a more reliable estimation of data contributions.

cs.CL↗

arXiv2024

Formulating the Restoration of Distribution Networks as a Multiple Traveling Salesman Problem

Severe weather events can cause extensive damage to electrical distribution networks, requiring a multi-day restoration effort. Optimizing the dispatch of repair crews minimizes the severe socio-economic consequences of such events. Considering both repair times and travel times, we use graphical manipulations to transform this multiple crew scheduling problem into a type of traveling salesman problem(TSP). Specifically, we demonstrate that the restoration problem bears major resemblance to an instance of a cost constrained reward maximizing mTSP (multiple TSP) on node and edge weighted (doubly weighted) graphs (a variant we dub the CCRM-mTSP-DW), where the objective is to maximize the aggregate reward earned during the upcoming restoration window, provided no crew violates its time budget and electrical continuity constraints are met. Despite the rich history of research on the TSP and its variants, this CCRM-mTSP-DW variant has not been studied before, although its closest cousin happens to be the "Selective TSP" (S-TSP). This reinterpretation of the restoration problem not only opens up the possibility of drawing on existing solution methods developed for the TSP and its variants, it also adds a new chapter in the annals of research on "TSP-like'' problems. In this paper, we propose a "TSP-like'' mixed integer linear programming (MILP) model for solving the restoration problem and validate it on the IEEE PES 123-node test feeder network.

eess.SY↗

arXiv2020

Stabilizing Transient Disturbances With Utility-Scale Inverter-Based Resources

This paper presents a trajectory tracking control strategy that modulates the active power injected by geographically distributed inverter-based resources to support transient stability. Each resource is independently controlled, and its response drives the local bus voltage angle toward a trajectory that tracks the angle of the center of inertia. The center-of-inertia angle is estimated in real time from wide-area measurements. The main objectives are to stabilize transient disturbances and increase the amount of power that can be safely transferred over key transmission paths without loss of synchronism. Here we envision the actuators as utility-scale energy storage systems; however, equivalent examples could be developed for partially-curtailed photovoltaic generation and/or Type 4 wind turbine generators. The strategy stems from a time-varying linearization of the equations of motion for a synchronous machine. The control action produces synchronizing torque in a special reference frame that accounts for the motion of the center of inertia. This drives the system states toward the desired trajectory and promotes rotor angle stability. For testing we employ a reduced-order dynamic model of the North American Western Interconnection. The results show that this approach improves system reliability and can increase capacity utilization on stability-limited transmission corridors.

eess.SY↗

arXiv2020

CoronaSurveys: Using Surveys with Indirect Reporting to Estimate the Incidence and Evolution of Epidemics

The world is suffering from a pandemic called COVID-19, caused by the SARS-CoV-2 virus. National governments have problems evaluating the reach of the epidemic, due to having limited resources and tests at their disposal. This problem is especially acute in low and middle-income countries (LMICs). Hence, any simple, cheap and flexible means of evaluating the incidence and evolution of the epidemic in a given country with a reasonable level of accuracy is useful. In this paper, we propose a technique based on (anonymous) surveys in which participants report on the health status of their contacts. This indirect reporting technique, known in the literature as network scale-up method, preserves the privacy of the participants and their contacts, and collects information from a larger fraction of the population (as compared to individual surveys). This technique has been deployed in the CoronaSurveys project, which has been collecting reports for the COVID-19 pandemic for more than two months. Results obtained by CoronaSurveys show the power and flexibility of the approach, suggesting that it could be an inexpensive and powerful tool for LMICs.

cs.DC↗

arXiv2019

A Generalized PSS Architecture for Balancing Transient and Small-Signal Response

For decades, power system stabilizers paired with high initial response automatic voltage regulators have served as an effective means of meeting sometimes conflicting system stability requirements. Driven primarily by increases in power electronically-coupled generation and load, the dynamics of large-scale power systems are rapidly changing. Electric grids are losing inertia and traditional sources of voltage support and oscillation damping. The system load is becoming stiffer with respect to changes in voltage. In parallel, advancements in wide-area measurement technology have made it possible to implement control strategies that act on information transmitted over long distances in nearly real time. In this paper, we present a power system stabilizer architecture that can be viewed as a generalization of the standard $Δω$-type stabilizer. The control strategy utilizes a real-time estimate of the center-of-inertia speed derived from wide-area measurements. This approach creates a flexible set of trade-offs between transient and small-signal response, making synchronous generators better able to adapt to changes in system dynamics. The phenomena of interest are examined using a two-area test case and a reduced-order model of the North American Western Interconnection. To validate the key findings under realistic conditions, we employ a state-of-the-art co-simulation platform called HELICS to combine high-fidelity power system and communication network models. The benefits of the proposed control strategy are retained even under pessimistic assumptions of communication network performance.

eess.SY↗

arXiv2019

Post-disaster Repair Scheduling in Partially Automated Electricity Distribution Networks

Natural disasters require repairing all damaged components in electricity distribution networks. Optimal scheduling repair crews to minimize the aggregate duration of interruptions reduces the harm. We consider the fact that the number of switches is much smaller than the number of edges. The problem is modeled by a parallel identical machine scheduling with group soft precedence constraints to minimize the total weighted energization time. We propose an LP-based list scheduling algorithm and a conversion algorithm and analyze their theoretical performances.

math.OC↗

arXiv2018

Scheduling Post-Disaster Repairs in Electricity Distribution Networks

Natural disasters, such as hurricanes, earthquakes and large wind or ice storms, typically require the repair of a large number of components in electricity distribution networks. Since power cannot be restored before these repairs have been completed, optimally scheduling the available crews to minimize the cumulative duration of the customer interruptions reduces the harm done to the affected community. Considering the radial network structure of the distribution system, this repair and restoration process can be modeled as a scheduling problem with soft precedence constraints. As a benchmark, we first formulate this problem as a time-indexed ILP with valid inequalities. Three practical methods are then proposed to solve the problem: (i) an LP-based list scheduling algorithm, (ii) a single to multi-crew repair schedule conversion algorithm, and (iii) a dispatch rule based on $ρ$-factors which can be interpreted as Component Importance Measures. We show that the first two algorithms are $2$ and $\left(2 - \frac{1}{m}\right)$ approximations respectively. We also prove that the latter two algorithms are equivalent. Numerical results validate the effectiveness of the proposed methods.

math.OC↗

arXiv2017

Distribution Systems Hardening against Natural Disasters

Distribution systems are often crippled by catastrophic damage caused by a natural disaster. Well-designed hardening can significantly improve the performance of post-disaster restoration operations. Such performance is quantified by a resilience measure associated with the operability trajectory. The distribution system hardening problem can be formulated as a two-stage stochastic problem, where the inner operational problem addresses the proper sequencing of post-disaster repairs and the outer problem the judicious selection of components to harden. We propose a deterministic robust reformulation with two solution methods, an MILP formulation and a heuristic approach. We provide computational evidence on various IEEE test feeders which illustrates that the heuristic approach provides near-optimal hardening solutions efficiently.

math.OC↗

arXiv2016

A Modified CSMA/CA Protocol for OFDM Underwater Networks: Cross Layer Design

The underwater acoustic channel continues to present significant challenges to efficient throughput performance of underwater acoustic sensor networks (UASNs) in varying scenarios. As a result, cross-layer approaches that explore joint PHY/MAC strategies are worthy of further exploration. We consider a recent high-speed OFDM modem and propose a new cross-layer solution based on modified CSMA/CA, for a canonical star network topology with few nodes (the most common scenario in UASNs). Some innovations to an adaptive OFDM PHY link are developed to jointly select the modulation, convolutional coding and frequency diversity order (different transmission modes) for matching varying channel conditions. Additionally, receiver logic that disambiguates the cause of packet loss between a) that caused by channel vs. b) that due to collisions is used to modify the ARQ/backoff logic for retransmissions with CSMA/CA random access. Simulation results reveal that the cross-layer design can effectively increase network throughput.

cs.IT↗