A5008181744- Reinforcement Learning in Robotics
- Machine Learning and Algorithms
- Advanced Bandit Algorithms Research
- Advanced Vision and Imaging
- Evolutionary Algorithms and Applications
- Topic Modeling
- Advanced Image and Video Retrieval Techniques
- Energy Efficient Wireless Sensor Networks
- Adversarial Robustness in Machine Learning
- Interconnection Networks and Systems
- Complex Network Analysis Techniques
- Natural Language Processing Techniques
- Indoor and Outdoor Localization Technologies
- Mobile Crowdsensing and Crowdsourcing
- Video Surveillance and Tracking Methods
- Neural Networks and Applications
- Privacy-Preserving Technologies in Data
- Human Pose and Action Recognition
- Advanced Graph Neural Networks
- Recommender Systems and Techniques
- IoT and Edge/Fog Computing
- Opinion Dynamics and Social Influence
- Advanced Neural Network Applications
- Software-Defined Networks and 5G
- Formal Methods in Verification
East China Jiaotong University
2016-2025
University of Toronto
2025
Tsinghua University
2019-2025
Nanchang University
2025
Purdue University West Lafayette
2022-2024
Jilin Province Science and Technology Department
2024
Jilin University
2024
Shandong Institute of Business and Technology
2024
Shanghai Jiao Tong University
2013-2024
Southeast University
2024
Network-on-Chips (NoCs) are becoming integral parts of modern microprocessors as the number cores and modules integrated on a single chip continues to increase. Research development future NoC technology relies accurate modeling simulations evaluate performance impact analyze cost novel architectures. In this work, we present BookSim, cycle-accurate simulator for NoCs. The is designed simulation flexibility network components. It features modular design offers large set configurable...
We study the problem of off-policy value evaluation in reinforcement learning (RL), where one aims to estimate a new policy based on data collected by different policy. This is often critical step when applying RL real-world problems. Despite its importance, existing general methods either have uncontrolled bias or suffer high variance. In this work, we extend doubly robust estimator for bandits sequential decision-making problems, which gets best both worlds: it guaranteed be unbiased and...
We introduce a new photographing guidance (PhotoHelper) for amateur photographers to enhance their portrait photo quality using deep feature retrieval and fusion. In our model, we comprehensively integrate empirical aesthetic rules, traditional machine learning algorithms neural networks extract different kinds of features in both color space aspects. With these features, build modified random forest with structured photograph collection identify types photos. also define the composition...
Metallic elements having negative enthalpies of mixing tend to form characteristic local atomic clusters. In this review, we use the structural information in first nearest neighbour shell level, or first-shell cluster, derive composition rules two types complex alloy phases, quasicrystals and bulk metallic glasses, both being composed with mixing. We show phenomena quasicrystal-forming systems, where major such as cluster line, electron concentration size criteria are derived. Then analyse...
This paper studies systematic exploration for reinforcement learning with rich observations and function approximation. We introduce a new model called contextual decision processes, that unifies generalizes most prior settings. Our first contribution is complexity measure, the Bellman rank, we show enables tractable of near-optimal behavior in these processes naturally small many well-studied second algorithm engages to learn low rank. provably learns number samples polynomial all relevant...
With the rise of Internet Things (IoT) and fifth-generation (5G) networks, which have led to a surge in data processing increased transfer time, traditional cloud computing could no longer meet needs workers, so edge has emerged. Edge demand for low time consumption by at network then transmitting it third-party platform. However, since credibility platform is unknown can easily leak privacy workers. For transparent mechanism blockchain, two-stage protection based on blockchain proposed...
Using inertial measurement units mounted on foot is a feasible approach to improve the positioning accuracy for human motion capture system. This paper presents lightweight and low cost wireless system simultaneous reconstruction of body attitude displacement. First all, device based sensor networks distributes 15 nodes key limbs. Then, after an initial alignment with reduced error, zero-speed update algorithm used calculate In addition, constantly posture information, kind method gradient...
Custom accelerators improve the energy efficiency, area and performance of deep neural network (DNN) inference. This article presents a scalable DNN accelerator consisting 36 chips connected in mesh on multi-chip-module (MCM) using ground-referenced signaling (GRS). While previous fabricated single monolithic chip are optimal for specific sizes, proposed architecture enables flexible scaling efficient inference wide range DNNs, from mobile to data center domains. Communication is minimized...
Social trust assessment that characterizes a pairwise trustworthiness relationship can spur diversified applications. Extensive efforts have been put in exploration, but mainly focusing on applying graph convolutional network to establish social evaluation model, overlooking user feature factors related context-aware information prediction. In this article, we aim design new framework GATrust which integrates multi-aspect properties of users, including context-specific information,...
ABSTRACT Aims This study investigated the current status of nutrition literacy and related influencing factors in stroke patients, with a view to providing reference for development targeted interventions. Design Cross‐sectional study. Methods A convenience sampling method was used select 342 patients from June November 2024 as population, cross‐sectional survey conducted using General Information Questionnaire, Nutrition Literacy Scale, Herth Hope Chronic Disease Self‐Efficacy Scale Social...
Recently proposed high-radix interconnection networks [10] require global adaptive routing to achieve optimum performance. Existing direct methods are slow sense congestion remote from the source router and hence misroute many packets before such is detected. This paper introduces indirect (IAR) in which decision uses information that not directly available at router. We describe four IAR methods: credit round trip (CRT) [10], progressive (PAR), piggyback (PB), reservation (RES). evaluate...
For Markov decision processes with long horizons (i.e., discount factors close to one), it is common in practice use reduced during planning speed computation. However, perhaps surprisingly, when the model available agent estimated from data, as will be case most real-world problems, policy found using a shorter horizon can actually better than learned true horizon. In this paper we provide precise explanation for phenomenon based on principles of learning theory. We show formally that...
Matching the visual appearances of target over consecutive image frames is most critical issue in video-based object tracking. Choosing an appropriate distance metric for matching determines its accuracy and robustness, thus significantly influences tracking performance. Most existing methods employ fixed pre-specified metrics. However, this simple treatment problematic limited practice, because a does not likely to guarantee closest match be true interest. This paper presents new approach...
We offer an experimental benchmark and empirical study for off-policy policy evaluation (OPE) in reinforcement learning, which is a key problem many safety critical applications. Given the increasing interest deploying learning-based methods, there has been flurry of recent proposals OPE method, leading to need standardized analyses. Our work takes strong focus on diversity design enable stress testing methods. provide comprehensive benchmarking suite interplay different attributes method...
Text sentiment classification is of critical importance to improve the autonomous decision making and communication ability among object peers in Social Internet Things (SIoT). To classify polarity on a fine-grained level, aspect-level has become promising direction recent years. However, existing solutions typically ignore mutual information between sentences their respective aspect terms while generally performing by using simple attention mechanism. Thus, relevant results seem be...
In the participatory sensing framework, privacy protection of Internet Things (IoT) is very important. this article, cryptography-based methods are utilized to protect participants' information in unsecured network channels for dynamic and real-time tasks. The edge computing paradigm introduced traditional framework reduce latency. Then, Rivest Cipher 4 stream cipher logistic mapping combined deal with problems limited resources untruthful third-party platforms. Finally, product algebra...