arXiv subjects

Yanru Zhang

Publications and source records attributed to Yanru Zhang.

30 records · Page 2Linked to original sources

arXiv2023

Deep Reinforcement Learning-Assisted Federated Learning for Robust Short-term Utility Demand Forecasting in Electricity Wholesale Markets

Short-term load forecasting (STLF) plays a significant role in the operation of electricity trading markets. Considering the growing concern of data privacy, federated learning (FL) is increasingly adopted to train STLF models for utility companies (UCs) in recent research. Inspiringly, in wholesale markets, as it is not realistic for power plants (PPs) to access UCs' data directly, FL is definitely a feasible solution of obtaining an accurate STLF model for PPs. However, due to FL's distributed nature and intense competition among UCs, defects increasingly occur and lead to poor performance of the STLF model, indicating that simply adopting FL is not enough. In this paper, we propose a DRL-assisted FL approach, DEfect-AwaRe federated soft actor-critic (DearFSAC), to robustly train an accurate STLF model for PPs to forecast precise short-term utility electricity demand. Firstly. we design a STLF model based on long short-term memory (LSTM) using just historical load data and time data. Furthermore, considering the uncertainty of defects occurrence, a deep reinforcement learning (DRL) algorithm is adopted to assist FL by alleviating model degradation caused by defects. In addition, for faster convergence of FL training, an auto-encoder is designed for both dimension reduction and quality evaluation of uploaded models. In the simulations, we validate our approach on real data of Helsinki's UCs in 2019. The results show that DearFSAC outperforms all the other approaches no matter if defects occur or not.

cs.DC↗

arXiv2022

FedTADBench: Federated Time-Series Anomaly Detection Benchmark

Time series anomaly detection strives to uncover potential abnormal behaviors and patterns from temporal data, and has fundamental significance in diverse application scenarios. Constructing an effective detection model usually requires adequate training data stored in a centralized manner, however, this requirement sometimes could not be satisfied in realistic scenarios. As a prevailing approach to address the above problem, federated learning has demonstrated its power to cooperate with the distributed data available while protecting the privacy of data providers. However, it is still unclear that how existing time series anomaly detection algorithms perform with decentralized data storage and privacy protection through federated learning. To study this, we conduct a federated time series anomaly detection benchmark, named FedTADBench, which involves five representative time series anomaly detection algorithms and four popular federated learning methods. We would like to answer the following questions: (1)How is the performance of time series anomaly detection algorithms when meeting federated learning? (2) Which federated learning method is the most appropriate one for time series anomaly detection? (3) How do federated time series anomaly detection approaches perform on different partitions of data in clients? Numbers of results as well as corresponding analysis are provided from extensive experiments with various settings. The source code of our benchmark is publicly available at https://github.com/fanxingliu2020/FedTADBench.

cs.LG↗

arXiv2022

Deep Reinforcement Learning for Optimal Power Flow with Renewables Using Graph Information

Renewable energy resources (RERs) have been increasingly integrated into large-scale distributed power systems. Considering uncertainties and voltage fluctuation issues introduced by RERs, in this paper, we propose a deep reinforcement learning (DRL)-based strategy leveraging spatial-temporal (ST) graphical information of power systems, to dynamically search for the optimal operation, i.e., optimal power flow (OPF), of power systems with a high uptake of RERs. Specifically, we formulate the OPF problem as a multi-objective optimization problem considering generation cost, voltage fluctuation, and transmission loss, and employ deep deterministic policy gradient (DDPG) to learn an optimal allocation strategy for OPF. Moreover, given that the nodes in power systems are self-correlated and interrelated in temporal and spatial views, we develop a multi-grained attention-based spatial-temporal graph convolution network (MG-ASTGCN) for extracting ST graphical correlations and features, aiming to provide prior knowledge of power systems for its sequential DDPG algorithm to more effectively solve OPF. We validate our algorithm on modified IEEE 33, 69, and 118-bus radial distribution systems and demonstrate that our algorithm outperforms other benchmark algorithms. Our experimental results also reveal that our MG-ASTGCN can significantly accelerate DDPG's training process and performance in solving OPF.

cs.LG↗

arXiv2022

DearFSAC: An Approach to Optimizing Unreliable Federated Learning via Deep Reinforcement Learning

In federated learning (FL), model aggregation has been widely adopted for data privacy. In recent years, assigning different weights to local models has been used to alleviate the FL performance degradation caused by differences between local datasets. However, when various defects make the FL process unreliable, most existing FL approaches expose weak robustness. In this paper, we propose the DEfect-AwaRe federated soft actor-critic (DearFSAC) to dynamically assign weights to local models to improve the robustness of FL. The deep reinforcement learning algorithm soft actor-critic is adopted for near-optimal performance and stable convergence. Besides, an auto-encoder is trained to output low-dimensional embedding vectors that are further utilized to evaluate model quality. In the experiments, DearFSAC outperforms three existing approaches on four datasets for both independent and identically distributed (IID) and non-IID settings under defective scenarios.

cs.LG↗

arXiv2022

Protum: A New Method For Prompt Tuning Based on "[MASK]"

Recently, prompt tuning \cite{lester2021power} has gradually become a new paradigm for NLP, which only depends on the representation of the words by freezing the parameters of pre-trained language models (PLMs) to obtain remarkable performance on downstream tasks. It maintains the consistency of Masked Language Model (MLM) \cite{devlin2018bert} task in the process of pre-training, and avoids some issues that may happened during fine-tuning. Naturally, we consider that the "[MASK]" tokens carry more useful information than other tokens because the model combines with context to predict the masked tokens. Among the current prompt tuning methods, there will be a serious problem of random composition of the answer tokens in prediction when they predict multiple words so that they have to map tokens to labels with the help verbalizer. In response to the above issue, we propose a new \textbf{Pro}mpt \textbf{Tu}ning based on "[\textbf{M}ASK]" (\textbf{Protum}) method in this paper, which constructs a classification task through the information carried by the hidden layer of "[MASK]" tokens and then predicts the labels directly rather than the answer tokens. At the same time, we explore how different hidden layers under "[MASK]" impact on our classification model on many different data sets. Finally, we find that our \textbf{Protum} can achieve much better performance than fine-tuning after continuous pre-training with less time consumption. Our model facilitates the practical application of large models in NLP.

cs.CL↗

arXiv2021

Feature Encoding with AutoEncoders for Weakly-supervised Anomaly Detection

Weakly-supervised anomaly detection aims at learning an anomaly detector from a limited amount of labeled data and abundant unlabeled data. Recent works build deep neural networks for anomaly detection by discriminatively mapping the normal samples and abnormal samples to different regions in the feature space or fitting different distributions. However, due to the limited number of annotated anomaly samples, directly training networks with the discriminative loss may not be sufficient. To overcome this issue, this paper proposes a novel strategy to transform the input data into a more meaningful representation that could be used for anomaly detection. Specifically, we leverage an autoencoder to encode the input data and utilize three factors, hidden representation, reconstruction residual vector, and reconstruction error, as the new representation for the input data. This representation amounts to encode a test sample with its projection on the training data manifold, its direction to its projection and its distance to its projection. In addition to this encoding, we also propose a novel network architecture to seamlessly incorporate those three factors. From our extensive experiments, the benefits of the proposed strategy are clearly demonstrated by its superior performance over the competitive methods.

cs.LG↗

arXiv2019

Deep Anomaly Detection in Packet Payload

With the widespread adoption of cloud services, especially the extensive deployment of plenty of Web applications, it is important and challenging to detect anomalies from the packet payload. For example, the anomalies in the packet payload can be expressed as a number of specific strings which may cause attacks. Although some approaches have achieved remarkable progress, they are with limited applications since they are dependent on in-depth expert knowledge, e.g., signatures describing anomalies or communication protocol at the application level. Moreover, they might fail to detect the payload anomalies that have long-term dependency relationships. To overcome these limitations and adaptively detect anomalies from the packet payload, we propose a deep learning based framework which consists of two steps. First, a novel feature engineering method is proposed to obtain the block-based features via block sequence extraction and block embedding. The block-based features could encapsulate both the high-dimension information and the underlying sequential information which facilitate the anomaly detection. Second, a neural network is designed to learn the representation of packet payload based on Long Short-Term Memory (LSTM) and Convolutional Neural Networks (CNN). Furthermore, we cast the anomaly detection as a classification problem and stack a Multi-Layer Perception (MLP) on the above representation learning network to detect anomalies. Extensive experimental results on three public datasets indicate that our model could achieve a higher detection rate, while keeping a lower false positive rate compared with five state-of-the-art methods.

eess.SP↗

arXiv2017

Non-Cash Auction for Spectrum Trading in Cognitive Radio Networks: A Contract Theoretical Model with Joint Adverse Selection and Moral Hazard

In cognitive radio networks (CRNs), spectrum trading is an efficient way for secondary users (SUs) to achieve dynamic spectrum access and to bring economic benefits for the primary users (PUs). Existing methods requires full payment from SU, which blocked many potential "buyers", and thus limited the PU's expected income. To better improve PUs' revenue from spectrum trading in a CRN, we introduce a financing contract, which is similar to a sealed non-cash auction that allows SU to do a financing. Unlike previous mechanism designs in CRN, the financing contract allows the SU to only pay part of the total amount when the contract is signed, known as the down payment. Then, after the spectrum is released and utilized, the SU pays the rest of payment, known as the installment payment, from the revenue generated by utilizing the spectrum. The way the financing contract carries out and the sealed non-cash auction works similarly. Thus, contract theory is employed here as the mathematical framework to solve the non-cash auction problem and form mutually beneficial relationships between PUs and SUs. As the PU may not have the full acknowledgement of the SU's financial status, nor the SU's capability in making revenue, the problems of adverse selection and moral hazard arise in the two scenarios, respectively. Therefore, a joint adverse selection and moral hazard model is considered here. In particular, we present three situations when either or both adverse selection and moral hazard are present during the trading. Furthermore, both discrete and continuous models are provided in this paper. Through extensive simulations, we show that the adverse selection and moral hazard cases serve as the upper and lower bounds of the general case where both problems are present.

cs.NI↗

arXiv2017

Multi-Dimensional Payment Plan in Fog Computing with Moral Hazard

Recently, the concept of fog computing which aims at providing time-sensitive data services has become popular. In this model, computation is performed at the edge of the network instead of sending vast amounts of data to the cloud. Thus, fog computing provides low latency, location awareness to end users, and improves quality-of-services (QoS). One key feature in this model is the designing of payment plan from network operator (NO) to fog nodes (FN) for the rental of their computing resources, such as computation capacity, spectrum, and transmission power. In this paper, we investigate the problem of how to design the efficient payment plan to maximize the NO's revenue while maintaining FN's incentive to cooperate through the moral hazard model in contract theory. We propose a multi-dimensional contract which considers the FNs' characteristics such as location, computation capacity, storage, transmission bandwidth, and etc. First, a contract which pays the FNs by evaluating the resources they have provided from multiple aspects is proposed. Then, the utility maximization problem of the NO is formulated. Furthermore, we use the numerical results to analyze the optimal payment plan, and compare the NO's utility under different payment plans.

cs.NI↗

arXiv2015

Offloading in Software Defined Network at Edge with Information Asymmetry: A Contract Theoretical Approach

The proliferation of highly capable mobile devices such as smartphones and tablets has significantly increased the demand for wireless access. Software defined network (SDN) at edge is viewed as one promising technology to simplify the traffic offloading process for current wireless networks. In this paper, we investigate the incentive problem in SDN-at-edge of how to motivate a third party access points (APs) such as WiFi and smallcells to offload traffic for the central base stations (BSs). The APs will only admit the traffic from the BS under the precondition that their own traffic demand is satisfied. Under the information asymmetry that the APs know more about own traffic demands, the BS needs to distribute the payment in accordance with the APs' idle capacity to maintain a compatible incentive. First, we apply a contract-theoretic approach to model and analyze the service trading between the BS and APs. Furthermore, other two incentive mechanisms: optimal discrimination contract and linear pricing contract are introduced to serve as the comparisons of the anti adverse selection contract. Finally, the simulation results show that the contract can effectively incentivize APs' participation and offload the cellular network traffic. Furthermore, the anti adverse selection contract achieves the optimal outcome under the information asymmetry scenario.

cs.NI↗

arXiv2015

Exploring Social Ties for Enhanced Device-to-Device Communications in Wireless Networks

Device-to-device (D2D) communications is seen as a major technology to overcome the imminent wireless capacity crunch and to enable novel application services. In this paper, we propose a novel, social-aware approach for optimizing D2D communications by exploiting two network layers: the social network and the physical, wireless network. First we formulate the physical layer D2D network according to users' encounter histories. Subsequently, we propose a novel approach, based on the so-called Indian Buffet Process, so as to model the distribution of contents in users' online social networks. Given the online and offline social relations collected by the Evolved Node B, we jointly optimize the traffic offload process in D2D communication. Simulation results show that the proposed approach offload the traffic of Evolved Node B successfully.

cs.SI↗

arXiv2015

Social Network Enhanced Device-to-Device Communication Underlaying Cellular Networks

Device-to-device (D2D) communication has seen as a major technology to overcome the imminent wireless capacity crunch and to enable new application services. In this paper, we propose a social-aware approach for optimizing D2D communication by exploiting two layers: the social network and the physical wireless layers. First we formulate the physical layer D2D network according to users' encounter histories. Subsequently, we propose an approach, based on the so-called Indian Buffet Process, so as to model the distribution of contents in users' online social networks. Given the social relations collected by the Evolved Node B (eNB), we jointly optimize the traffic offloading process in D2D communication. In addition, we give the Chernoff bound and approximated cumulative distribution function (CDF) of the offloaded traffic. In the simulation, we proved the effectiveness of the bound and CDF. The numerical results based on real traces show that the proposed approach offload the traffic of eNB's successfully.

cs.SI↗