High-dimensional problems have long been considered the Achilles' heel of Bayesian optimization. Spurred by the curse of dimensionality, a large collection of algorithms aim to make it more performant in this sett...
详细信息
High-dimensional problems have long been considered the Achilles' heel of Bayesian optimization. Spurred by the curse of dimensionality, a large collection of algorithms aim to make it more performant in this setting, commonly by imposing various simplifying assumptions on the objective. In this paper, we identify the degeneracies that make vanilla Bayesian optimization poorly suited to high-dimensional tasks, and further show how existing algorithms address these degeneracies through the lens of lowering the model complexity. Moreover, we propose an enhancement to the prior assumptions that are typical to vanilla Bayesian optimization, which reduces the complexity to manageable levels without imposing structural restrictions on the objective. Our modification - a simple scaling of the Gaussian process lengthscale prior with the dimensionality - reveals that standard Bayesian optimization works drastically better than previously thought in high dimensions, clearly outperforming existing state-of-the-art algorithms on multiple commonly considered real-world high-dimensional tasks. Copyright 2024 by the author(s)
This paper proposes an optimization algorithm for a sparse wideband planar antenna array based on the covariance matrix adaptive evolutionary strategy algorithm. By dividing the planar antenna array into multiple regi...
详细信息
Offline preference optimization allows fine-tuning large models directly from offline data, and has proved effective in recent alignment practices. We propose generalized preference optimization (GPO), a family of off...
详细信息
Offline preference optimization allows fine-tuning large models directly from offline data, and has proved effective in recent alignment practices. We propose generalized preference optimization (GPO), a family of offline losses parameterized by a general class of convex functions. GPO enables a unified view over preference optimization, encompassing existing algorithms such as DPO, IPO and SLiC as special cases, while naturally introducing new variants. The GPO framework also sheds light on how offline algorithms enforce regularization, through the design of the convex function that defines the loss. Our analysis and experiments reveal the connections and subtle differences between the offline regularization and the KL divergence regularization intended by the canonical RLHF formulation. In a controlled setting akin to Gao et al. (2023), we also show that different GPO variants achieve similar trade-offs between regularization and performance, though the optimal values of hyper-parameter might differ as predicted by theory. In all, our results present new algorithmic toolkits and empirical insights to alignment practitioners. Please see https://***/pdf/2402.05749 for the full version of the paper. Copyright 2024 by the author(s)
This paper explores the dynamic assortment optimization problem in the context of live-streaming sales. In response to the dynamic characteristics of live-streaming e-commerce, we propose a new choice model that exten...
详细信息
Mixed-variable optimization problems (MVOPs) involve both discrete and continuous decision variables which lead to a lot of difficulties solving these problems. In this paper, a hybrid differential evolution for MVOPs...
详细信息
The Backtracking Search Algorithm (BSA) stands out as a contemporary stochastic technique that has showcased its prowess in tackling intricate engineering challenges. Thus, a novel hybrid evolutionary algorithm design...
详细信息
Hierarchical decision making problems, such as bilevel programs and Stackelberg games, are attracting increasing interest in both the engineering and machine learning communities. Yet, existing solution methods lack e...
详细信息
Hierarchical decision making problems, such as bilevel programs and Stackelberg games, are attracting increasing interest in both the engineering and machine learning communities. Yet, existing solution methods lack either convergence guarantees or computational efficiency, due to the absence of smoothness and convexity. In this work, we bridge this gap by designing a first-order hypergradient-based algorithm for Stackelberg games and mathematically establishing its convergence using tools from nonsmooth analysis. To evaluate the hypergradient, namely, the gradient of the upper-level objectve, we develop an online scheme that simultaneously computes the lower level equilibrium and its Jacobian. Crucially, this scheme exploits and preserves the original hierarchical and distributed structure of the problem, which renders it scalable and privacy-preserving. We numerically verify the computational efficiency and scalability of our algorithm on a large-scale hierarchical demand-response model.
A stand-alone hybrid energy system is needed to mitigate the growing demand for future energy and to reduce emissions generated by conventional fuel sources. This work shows and proposes an optimum hybrid energy syste...
详细信息
A stand-alone hybrid energy system is needed to mitigate the growing demand for future energy and to reduce emissions generated by conventional fuel sources. This work shows and proposes an optimum hybrid energy system configuration with less environmental pollution for a coastal region of Bangladesh with a load demand of 292.20 kWh/d and a peak load of 41.14 kW. This research presents the cost of energy (COE) and net present cost (NPC) for a PV-wind hybrid system with five alternative fuel generator technologies. For each possible configuration, the effects of three alternative dispatch algorithms, load following (LF), cycle charging (CC), and combined dispatch (CD), are examined. All three algorithms were validated using a genetic algorithm (GA), Cuckoo search algorithm (CUSA), Constrained Particle Swarm optimization (CPSO), Harmony search algorithm (HSA), and Non-dominated Sorting Genetic Algorithm (NSGA-II). Furthermore, a sensitivity analysis is also conducted by utilizing the gasoline price, discount rate, battery cost, PV cost, and inflation rate. The results show that the PV-wind-natural gas-based system provides the minimum COE (0.196 USD/kWh) and NPC (270,483 USD) when the CD algorithm is followed. For the same optimal hybrid energy system, the COE is 0.201 USD/ kWh, 0.0998 USD/kWh, 0.101 USD/kWh, 0.101 USD/kWh, and 0.0987 USD/kWh respectively for GA, CUSA, CPSO, HSA, and NSGA-II. However, COE increases by 8 % (0.212 USD/kWh) and 15 % (0.225 USD/kWh) when the CC and LF algorithms are followed. In addition, a comparison of all configurations reveals that the PV-windbiomass configuration with the CD algorithm excludes biomass generators to lower the COE, making the system emission-free.
The COVID-19 outbreak has negatively impacted the income of many bank users. Many users without emergency funds had difficulty coping with this unexpected event and had to use credit or apply to the government for bai...
详细信息
The COVID-19 outbreak has negatively impacted the income of many bank users. Many users without emergency funds had difficulty coping with this unexpected event and had to use credit or apply to the government for bailout funds. Therefore, it is necessary to develop spending plans and deposit plans based on transaction data of users to assist them in saving sufficient emergency funds to cope with unexpected events. In this paper, an emergency fund model is proposed, and two optimization algorithms are applied to solve the optimal solution of the model. Secondly, an early warning mechanism is proposed, i.e. an unexpected prevention index and a consumption index are proposed to measure the ability of users to cope with unexpected events and the reasonableness of their expenditure respectively, which provides early warning to users. Finally, the model is experimented with real bank users and the performance of the model is analysed. The experiments show that compared to the no-planning scenario, the model helps users to save more emergency funds to cope with unexpected events, furthermore, the proposed model is real-time and sensitive.
Coal and gas outbursts are a major cause of fatalities in underground coal mines and pose a threat to coal power generation worldwide. Among the current mitigation efforts include monitoring methane gas levels using s...
详细信息
Coal and gas outbursts are a major cause of fatalities in underground coal mines and pose a threat to coal power generation worldwide. Among the current mitigation efforts include monitoring methane gas levels using sen-sors, employing geophysical surveys to identify geological structures and zones prone to outbursts, and using empirical modeling for outburst predictions. However, in the wake of industry 4.0 technologies, several studies have been conducted on applying artificial intelligence methods to predict outbursts. The proposed models and their results vary significantly in the literature. This study reviews the application of machine learning (ML) to predict coal and gas outbursts in underground mines using a mixed-method approach. Most of the available literature, with a focus on ML applications in coal and gas outburst prediction, was investigated in China. Findings indicate that researchers proposed diverse ML models mostly combined with different optimization algorithms, including particle swarm optimization (PSO), genetic algorithm (GA), rough set (RS), and fruit fly optimization algorithm (IFOA) for outburst prediction. The number and type of input parameters used for prediction differed significantly, with initial gas velocity being the most dominant parameter for gas outbursts, and coal seam depth as the dominant parameter for coal outbursts. The datasets for training and testing the proposed ML models in the literature varied significantly but were mostly insufficient - which questions the reliability of some of the applied ML models. Future research should investigate the effect of data size and input parameters on coal and gas outburst prediction.
暂无评论