Boyu Li

ORCID: 0000-0003-2425-936X
Publications
Citations
Views
---
Saved
---
About
Contact & Profiles
Research Areas
  • Speech Recognition and Synthesis
  • Privacy-Preserving Technologies in Data
  • Data Quality and Management
  • Neural Networks and Reservoir Computing
  • Advanced Graph Neural Networks
  • Adversarial Robustness in Machine Learning
  • Natural Language Processing Techniques
  • Cryptography and Data Security
  • Semantic Web and Ontologies
  • Domain Adaptation and Few-Shot Learning
  • Service-Oriented Architecture and Web Services
  • Anomaly Detection Techniques and Applications
  • Text and Document Classification Technologies
  • Explainable Artificial Intelligence (XAI)
  • Machine Learning and Data Classification
  • Internet Traffic Analysis and Secure E-voting
  • Reinforcement Learning in Robotics
  • Face and Expression Recognition
  • Complex Network Analysis Techniques
  • Neural Networks and Applications
  • Speech and dialogue systems
  • AI in cancer detection
  • Imbalanced Data Classification Techniques
  • Access Control and Trust
  • Human Mobility and Location-Based Analysis

Institute of Automation
2024-2025

Chinese Academy of Sciences
2021-2025

Hainan University
2025

University of Technology Sydney
2024

Zhongnan Hospital of Wuhan University
2023-2024

Wuhan University
2023-2024

Beijing Academy of Artificial Intelligence
2024

University of Chinese Academy of Sciences
2021-2024

Nanjing University
2024

Henan University
2023

When classes are nonseparable or overlapping, training samples in a local neighborhood may come from different classes. In this situation, the with class labels be comparable of query. As consequence, conventional nearest neighbor classifier, such as kappa-nearest scheme, produce wrong prediction. To address issue, paper, we propose new classification method, which performs task based on probabilistic centers each class. This method works by reducing number negative contributing points,...

10.1109/tsmcb.2007.908363 article EN IEEE Transactions on Systems Man and Cybernetics Part B (Cybernetics) 2008-01-16

10.1109/tsmc.2025.3541926 article EN IEEE Transactions on Systems Man and Cybernetics Systems 2025-01-01

Discrimination can occur when the underlying unbiased labels are overwritten by an agent with potential bias, resulting in biased datasets that unfairly harm specific groups and cause classifiers to inherit these biases. In this paper, we demonstrate despite only having access labels, it is possible eliminate bias filtering fairest instances within framework of confident learning. context learning, low self-confidence usually indicates label errors; however, not always case. Instances,...

10.1609/aaai.v38i15.29634 article EN Proceedings of the AAAI Conference on Artificial Intelligence 2024-03-24

A chip-scale chaotic laser system with optoelectronic delayed feedback is proposed and analyzed by numerical simulation. This chip eliminates the need for bulky delay components such as long optical fibers, free propagation external cavities, relying solely on internal devices waveguides to achieve delay. approach simplifies integration, maintaining a compact size. According results, exhibits rich dynamics, including periodicity, quasi-periodicity, states. Chaos resembling Gaussian white...

10.1364/oe.515058 article EN cc-by Optics Express 2024-04-02

Anonymization technique has been extensively studied and widely applied for privacy-preserving data publishing. In most previous approaches, a microdata table consists of three categories attribute: explicit-identifier, quasi-identifier (QI), sensitive attribute. Actually, different individuals may have view on the sensitivity attributes. Therefore, there is another type attribute that contains both QI values values, namely, semi-sensitive Based such observation, we propose new anonymization...

10.1016/j.jksuci.2022.12.008 article EN cc-by-nc-nd Journal of King Saud University - Computer and Information Sciences 2022-12-28

A number of processes and pathways have been reported in the development Group I pulmonary hypertension (Group PAH); however, novel biomarkers need to be identified for a better diagnosis management. We employed robust rank aggregation (RRA) algorithm shortlist key differentially expressed genes (DEGs) between PAH patients controls. An optimal diagnostic model was obtained by comparing seven machine learning algorithms verified an independent dataset. The functional roles DEGs were analyzed...

10.3390/ijms24098050 article EN International Journal of Molecular Sciences 2023-04-28

Training latency is critical for the success of numerous intrigued applications ignited by federated learning (FL) over heterogeneous mobile devices. By revolutionarily overlapping local gradient transmission with continuous computing, FL can remarkably reduce its training homogeneous clients, yet encounter severe model staleness, drifts, memory cost and straggler issues in environments. To unleash full potential overlapping, we propose, FedEx, a novel \underline{fed}erated approach to...

10.48550/arxiv.2407.00943 preprint EN arXiv (Cornell University) 2024-06-30

This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version no longer accessible. Unsupervised cross-domain Reinforcement Learning (RL) pre-training shows great potential challenging continuous visual control but poses a big challenge. In paper, we propose \textbf{C}ross-domain \textbf{R}andom \textbf{P}re-\textbf{T}raining with \textbf{pro}totypes (CRPTpro), novel, efficient, and effective self-supervised RL...

10.48550/arxiv.2302.05614 preprint EN other-oa arXiv (Cornell University) 2023-01-01

Robots are playing a more and important role in people's production life, recently. However, robot control dynamic environment is still difficulty. With the great breakthrough of deep reinforcement learning field video games, this method also extended to robots. Due gap between simulation real environment, algorithm trained difficult be applied environment. Aiming at gimbal with two degrees freedom (DOF), pipline combining system identification proposed. On one hand, shooting accuracy moving...

10.1109/iccss53909.2021.9722012 article EN 2020 7th International Conference on Information, Cybernetics, and Computational Social Systems (ICCSS) 2021-12-10

The traditional Infrastructure as a Service (IaaS) cloud platform tends to realize high data availability by introducing dedicated storage devices. However, this heterogeneous architecture has maintenance cost and might reduce the performance of virtual machines. In homogeneous IaaS platform, servers in would uniformly provide computing resources resources, which effectively solve above problems, although corresponding mechanisms need be introduced improve availability. Efficient resource...

10.1155/2022/7832117 article EN Journal of Sensors 2022-10-04

BACKGROUND Nuclear prelamin A recognition factor-like (NARFL) plays a crucial role in cytosolic iron-sulfur protein assembly (CIA) and protects cells against oxidative stress. In our previous study, we identified novel homozygous mutation NARFL that led to decreased expression consanguineous family with diffuse pulmonary arteriovenous malformations (DPAVMs) secondary hypertension. Additionally, observed narfl deletion zebrafish resulted larvae lethality, subintestinal vessel malformation,...

10.1101/2024.02.06.24302421 preprint EN cc-by-nc-nd medRxiv (Cold Spring Harbor Laboratory) 2024-02-08

Recent studies have demonstrated the success of foundation agents in specific tasks or scenarios. However, existing cannot generalize across different scenarios, mainly due to their diverse observation and action spaces semantic gaps, reliance on task-specific resources. In this work, we propose General Computer Control (GCC) setting: building that can master any computer task by taking only screen images (and possibly audio) as input, producing keyboard mouse operations output, similar...

10.48550/arxiv.2403.03186 preprint EN arXiv (Cornell University) 2024-03-05

Agent-agnostic reinforcement learning aims to learn a universal control policy that can simultaneously set of robots with different morphologies. Recent studies have suggested using the transformer model address variations in state and action spaces caused by morphologies, morphology information is necessary improve performance. However, existing methods limitations exploiting morphological information, where rationality observation integration cannot be guaranteed. We propose Morphological...

10.1109/tcds.2024.3383158 article EN IEEE Transactions on Cognitive and Developmental Systems 2024-04-01

Reservoir computation models form a subclass of recurrent neural networks with fixed non-trainable input and dynamic coupling weights. Only the static readout from state space (reservoir) is trainable, thus avoiding known problems propagation gradient information backwards through time. have been successfully applied in variety tasks were shown to be universal approximators time-invariant fading memory filters under various settings. Simple cycle reservoirs (SCR) suggested as severely...

10.48550/arxiv.2308.10793 preprint EN other-oa arXiv (Cornell University) 2023-01-01

Deep biasing for the Transducer can improve recognition performance of rare words or contextual entities, which is essential in practical applications, especially streaming Automatic Speech Recognition (ASR). However, deep with large-scale remains challenging, as drops significantly when more distractors exist and there are similar grapheme sequences bias list. In this paper, we combine phoneme textual information Transducers to distinguish pronunciation spelling. Moreover, introduction...

10.1109/asru57964.2023.10389716 article EN 2021 IEEE Automatic Speech Recognition and Understanding Workshop (ASRU) 2023-12-16

Natural language communication between humans and machines is a popular direction of artificial intelligence. As technology for realizing direct dialogue machines, speech recognition can convert human signals into information, providing technical support human-machine communication. In this paper, we conduct study. First, collect more than 1,000 pieces data, preprocess the extract three different characteristics such as LPC, LPCC, MFCC, divide data training set test set. Two naive Bayesian...

10.1109/iciscae51034.2020.9236812 article EN 2020-09-27

Fast gradient sign attack series are popular methods that used to generate adversarial examples. However, most of the approaches based on fast cannot balance indistinguishability and transferability due limitations basic structure. To address this problem, we propose a method, called Adam Iterative Gradient Tanh Method (AI-FGTM), indistinguishable examples with high transferability. Besides, smaller kernels dynamic step size also applied for further increasing success rates. Extensive...

10.48550/arxiv.2007.03838 preprint EN other-oa arXiv (Cornell University) 2020-01-01

Domain adaptation using text-only corpus is challenging in end-to-end(E2E) speech recognition. Adaptation by synthesizing audio from text through TTS resource-consuming. We present a method to learn Unified Speech-Text Representation Conformer Transducer(USTR-CT) enable fast domain the corpus. Different previous textogram method, an extra encoder introduced our work representation and removed during inference, so there no modification for online deployment. To improve efficiency of...

10.48550/arxiv.2306.04076 preprint EN cc-by arXiv (Cornell University) 2023-01-01

This paper presents three models for predicting, classifying, and evaluating game data, which is collected from Wordle's Twitter account. Model I a predictive model based on an improved LSTM algorithm that uses Sliding window to improve accuracy. By using short-term predictions, the training set updated, reducing long-term uncertainty. II distribution PLS MLP algorithms, used predict linear part, while compensates deficiency in nonlinear part. III K-means classification elbow rule determine...

10.1109/cvidl58838.2023.10166102 article EN 2023-05-12

Deep biasing for the Transducer can improve recognition performance of rare words or contextual entities, which is essential in practical applications, especially streaming Automatic Speech Recognition (ASR). However, deep with large-scale remains challenging, as drops significantly when more distractors exist and there are similar grapheme sequences bias list. In this paper, we combine phoneme textual information Transducers to distinguish pronunciation spelling. Moreover, introduction...

10.48550/arxiv.2311.08966 preprint EN cc-by arXiv (Cornell University) 2023-01-01

Discrimination can occur when the underlying unbiased labels are overwritten by an agent with potential bias, resulting in biased datasets that unfairly harm specific groups and cause classifiers to inherit these biases. In this paper, we demonstrate despite only having access labels, it is possible eliminate bias filtering fairest instances within framework of confident learning. context learning, low self-confidence usually indicates label errors; however, not always case. Instances,...

10.48550/arxiv.2312.08749 preprint EN other-oa arXiv (Cornell University) 2023-01-01

Ontology is being widely used for data integration and knowledge discovery in the field of engineering. ranking one important functions ontology search engine, which ranks searched ontologies a reasonable order. A good method can help user acquiring satisfied efficiently from considerable amount results. However, existing methods literatures are unable to rank well meet user's demand due their inherent shortcomings. In this paper, novel OntoDUIA proposed presented. It evaluates user-query...

10.1504/ijhpsa.2017.10013582 article EN International Journal of High Performance Systems Architecture 2017-01-01
Coming Soon ...