arXiv subjects

Zhiyuan Yang

Publications and source records attributed to Zhiyuan Yang.

60 records · Page 4Linked to original sources

arXiv2022

Pareto Set Learning for Expensive Multi-Objective Optimization

Expensive multi-objective optimization problems can be found in many real-world applications, where their objective function evaluations involve expensive computations or physical experiments. It is desirable to obtain an approximate Pareto front with a limited evaluation budget. Multi-objective Bayesian optimization (MOBO) has been widely used for finding a finite set of Pareto optimal solutions. However, it is well-known that the whole Pareto set is on a continuous manifold and can contain infinite solutions. The structural properties of the Pareto set are not well exploited in existing MOBO methods, and the finite-set approximation may not contain the most preferred solution(s) for decision-makers. This paper develops a novel learning-based method to approximate the whole Pareto set for MOBO, which generalizes the decomposition-based multi-objective optimization algorithm (MOEA/D) from finite populations to models. We design a simple and powerful acquisition search method based on the learned Pareto set, which naturally supports batch evaluation. In addition, with our proposed model, decision-makers can readily explore any trade-off area in the approximate Pareto set for flexible decision-making. This work represents the first attempt to model the Pareto set for expensive multi-objective optimization. Experimental results on different synthetic and real-world problems demonstrate the effectiveness of our proposed method.

cs.NE↗

arXiv2022

Pareto Set Learning for Neural Multi-objective Combinatorial Optimization

Multiobjective combinatorial optimization (MOCO) problems can be found in many real-world applications. However, exactly solving these problems would be very challenging, particularly when they are NP-hard. Many handcrafted heuristic methods have been proposed to tackle different MOCO problems over the past decades. In this work, we generalize the idea of neural combinatorial optimization, and develop a learning-based approach to approximate the whole Pareto set for a given MOCO problem without further search procedure. We propose a single preference-conditioned model to directly generate approximate Pareto solutions for any trade-off preference, and design an efficient multiobjective reinforcement learning algorithm to train this model. Our proposed method can be treated as a learning-based extension for the widely-used decomposition-based multiobjective evolutionary algorithm (MOEA/D). It uses a single model to accommodate all the possible preferences, whereas other methods use a finite number of solutions to approximate the Pareto set. Experimental results show that our proposed method significantly outperforms some other methods on the multiobjective traveling salesman problem, multiobjective vehicle routing problem, and multiobjective knapsack problem in terms of solution quality, speed, and model efficiency.

cs.LG↗

arXiv2021

Template NeRF: Towards Modeling Dense Shape Correspondences from Category-Specific Object Images

We present neural radiance fields (NeRF) with templates, dubbed Template-NeRF, for modeling appearance and geometry and generating dense shape correspondences simultaneously among objects of the same category from only multi-view posed images, without the need of either 3D supervision or ground-truth correspondence knowledge. The learned dense correspondences can be readily used for various image-based tasks such as keypoint detection, part segmentation, and texture transfer that previously require specific model designs. Our method can also accommodate annotation transfer in a one or few-shot manner, given only one or a few instances of the category. Using periodic activation and feature-wise linear modulation (FiLM) conditioning, we introduce deep implicit templates on 3D data into the 3D-aware image synthesis pipeline NeRF. By representing object instances within the same category as shape and appearance variation of a shared NeRF template, our proposed method can achieve dense shape correspondences reasoning on images for a wide range of object classes. We demonstrate the results and applications on both synthetic and real-world data with competitive results compared with other methods based on 3D information.

cs.CV↗

arXiv2021

Controllable Pareto Multi-Task Learning

A multi-task learning (MTL) system aims at solving multiple related tasks at the same time. With a fixed model capacity, the tasks would be conflicted with each other, and the system usually has to make a trade-off among learning all of them together. For many real-world applications where the trade-off has to be made online, multiple models with different preferences over tasks have to be trained and stored. This work proposes a novel controllable Pareto multi-task learning framework, to enable the system to make real-time trade-off control among different tasks with a single model. To be specific, we formulate the MTL as a preference-conditioned multiobjective optimization problem, with a parametric mapping from preferences to the corresponding trade-off solutions. A single hypernetwork-based multi-task neural network is built to learn all tasks with different trade-off preferences among them, where the hypernetwork generates the model parameters conditioned on the preference. For inference, MTL practitioners can easily control the model performance based on different trade-off preferences in real-time. Experiments on different applications demonstrate that the proposed model is efficient for solving various MTL problems.

cs.LG↗

arXiv2018

Cyber-Induced Risk Modeling for Microprocessor-Based Relays in Substations

Once critical substations are compromised, attack agents can coordinate among their peers to plot for maximizing disruption using local control devices. For defenders, it is critical to enumerate and identify all digital relays to determine the systemic risks. Any combination of disruptive switching via the compromised relays can result in misoperation or immediate effect to the system. The resulting consequence of these attack's initial events would possibly incur cascading failure to a grid. This paper quantifies the criticality of substation protective relays with the combination of the outage level and its corresponding severity risk index. The proposed hypothesized outages are based on the type of protective relaying, bus configuration of a substation, and commonly implemented relaying schemes, such as bus differential, directional overcurrent, and distance relays, are studied. This preliminary work also provides three approaches of determination in probabilities for sensitivity analysis. The proposed risk indices are evaluated using IEEE test systems.

eess.SP↗

arXiv2018

Impact Assessment of Hypothesized Cyberattacks on Interconnected Bulk Power Systems

The first-ever Ukraine cyberattack on power grid has proven its devastation by hacking into their critical cyber assets. With administrative privileges accessing substation networks/local control centers, one intelligent way of coordinated cyberattacks is to execute a series of disruptive switching executions on multiple substations using compromised supervisory control and data acquisition (SCADA) systems. These actions can cause significant impacts to an interconnected power grid. Unlike the previous power blackouts, such high-impact initiating events can aggravate operating conditions, initiating instability that may lead to system-wide cascading failure. A systemic evaluation of "nightmare" scenarios is highly desirable for asset owners to manage and prioritize the maintenance and investment in protecting their cyberinfrastructure. This survey paper is a conceptual expansion of real-time monitoring, anomaly detection, impact analyses, and mitigation (RAIM) framework that emphasizes on the resulting impacts, both on steady-state and dynamic aspects of power system stability. Hypothetically, we associate the combinatorial analyses of steady state on substations/components outages and dynamics of the sequential switching orders as part of the permutation. The expanded framework includes (1) critical/noncritical combination verification, (2) cascade confirmation, and (3) combination re-evaluation. This paper ends with a discussion of the open issues for metrics and future design pertaining the impact quantification of cyber-related contingencies.

cs.CR↗