A5059407219- EEG and Brain-Computer Interfaces
- Topic Modeling
- Domain Adaptation and Few-Shot Learning
- Multimodal Machine Learning Applications
- Natural Language Processing Techniques
- Human Pose and Action Recognition
- Text and Document Classification Technologies
- Neural Networks and Applications
- Advanced Memory and Neural Computing
- Autonomous Vehicle Technology and Safety
- Advanced Neural Network Applications
- Robotic Locomotion and Control
- IoT and Edge/Fog Computing
- Reinforcement Learning in Robotics
- Machine Learning and Data Classification
- Functional Brain Connectivity Studies
- Sleep and Work-Related Fatigue
- Advanced Image and Video Retrieval Techniques
- Robotic Path Planning Algorithms
- Speech and dialogue systems
- Neurobiology of Language and Bilingualism
- Neural dynamics and brain function
- Advanced Wireless Communication Technologies
- Data Stream Mining Techniques
- Cognitive Computing and Networks
University of Technology Sydney
2021-2024
University of British Columbia
2019-2021
Ping An (China)
2021
The University of Sydney
2021
Okanagan University College
2019
Continual learning methods aim at training a neural network from sequential data with streaming labels, relieving catastrophic forgetting. However, existing are based on and designed for convolutional networks (CNNs), which have not utilized the full potential of newly emerged powerful vision transformers. In this paper, we propose novel attention-based framework Lifelong Vision Transformer (LVT), to achieve better stability-plasticity trade-off continual learning. Specifically, an...
Brain-computer interfaces (BCIs) present a promising avenue by translating neural activity directly into text, eliminating the need for physical actions. However, existing non-invasive BCI systems have not successfully covered entire alphabet, limiting their practicality. In this paper, we propose novel EEG-based system with Curriculum-based Neural Spelling Framework, which recognizes all 26 alphabet letters decoding signals associated handwriting first, and then apply Generative AI (GenAI)...
Continual learning is an intellectual ability of artificial agents to learn new streaming labels from sequential data. The main impediment continual catastrophic forgetting, a severe performance degradation on previously learned tasks. Although simply replaying all previous data or continuously adding the model parameters could alleviate issue, it impractical in real-world applications due limited available resources. Inspired by mechanism human brain deepen its past impression, we propose...
Identifying meaningful brain activities is critical in brain-computer interface (BCI) applications. Recently, an increasing number of neural network approaches have been proposed to recognize EEG signals. However, these depend heavily on using complex structures improve the performance recognition and suffer from deficit training data. Inspired by waveform characteristics processing methods shared between speech signals, we propose Speech2EEG, a novel method that leverages pretrained...
The translation of brain dynamics into natural language is pivotal for brain-computer interfaces (BCIs). With the swift advancement large models, such as ChatGPT, need to bridge gap between and languages becomes increasingly pressing. Current methods, however, require eye-tracking fixations or event markers segment word-level features, which can restrict practical application these systems. To tackle issues, we introduce a novel framework, DeWave, that integrates discrete encoding sequences...
Decoding language from brain dynamics is an important open direction in the realm of brain-computer interface (BCI), especially considering rapid growth large models. Compared to invasive-based signals which require electrode implantation surgery, non-invasive neural (e.g. EEG, MEG) have attracted increasing attention their safety and generality. However, exploration not adequate three aspects: 1) previous methods mainly focus on EEG but none works address this problem MEG with better signal...
Decoding natural language from noninvasive brain signals has been an exciting topic with the potential to expand applications of brain-computer interface (BCI) systems. However, current methods face limitations in decoding sentences electroencephalography (EEG) signals. Improving performance requires development a more effective encoder for EEG modality. Nonetheless, learning generalizable representations remains challenge due relatively small scale existing datasets. In this paper, we...
Distributed artificial intelligence (AI) is becoming an efficient approach to fulfill the high and diverse requirements for future vehicular networks. However, distributed tasks generated by vehicles often require resources. A customized resource provision scheme required improve utilization of multi-dimensional In this work, a slice selection-based online offloading (SSOO) algorithm proposed in First, response time energy consumption are reduced processing locally on vehicles. Then,...
Streaming label learning aims to model newly emerged labels for multilabel classification systems, which requires plenty of new data training. However, in changing environments, only a small amount can practically be collected. In this work, we formulate and study few-shot streaming (FSLL), models emerging with few annotated examples by utilizing the knowledge learned from past labels. We propose meta-learning framework, semantic inference network (SIN), learn infer correlation between adapt...
It is desirable to include more controllable attributes enhance the diversity of generated responses in open-domain dialogue systems. However, existing methods can generate with only one attribute or lack a flexible way them multiple attributes. In this paper, we propose Progressively trained Hierarchical Encoder-Decoder (PHED) tackle task. More specifically, PHED deploys Conditional Variational AutoEncoder (CVAE) on Transformer aspect at stage. A vital characteristic CVAE separate latent...
The differences in brain dynamics across human subjects, commonly referred to as artifacts, have long been a challenge the field, severely limiting generalizability of recognition models. Traditional methods for artifact removal typically employ spectrum filtering or blind source separation, based on simple prior distribution assumptions, which ultimately constrain capacity model each subject's domain variance. In this paper, we propose novel approach generative denoising process, capable...
Recently, humanoid robots have made significant advances in their ability to perform challenging tasks due the deployment of Reinforcement Learning (RL), however, inherent complexity robots, including difficulty designing complicated reward functions and training entire sophisticated systems, still poses a notable challenge. To conquer these challenges, after many iterations in-depth investigations, we meticulously developed full-size robot, "Adam", whose innovative structural design greatly...
The utilization of Large Language Models (LLMs) within the realm reinforcement learning, particularly as planners, has garnered a significant degree attention in recent scholarly literature. However, substantial proportion existing research predominantly focuses on planning models for robotics that transmute outputs derived from perception into linguistic forms, thus adopting `pure-language' strategy. In this research, we propose hybrid End-to-End learning framework autonomous driving by...
Recent advancements in humanoid robotics, including the integration of hierarchical reinforcement learning-based control and utilization LLM planning, have significantly enhanced ability robots to perform complex tasks. In contrast highly developed robots, human factors involved remain relatively unexplored. Directly controlling with brain has already appeared many science fiction novels, such as Pacific Rim Gundam. this work, we present E2H (EEG-to-Humanoid), an innovative framework that...
Few-shot learning can adapt the classification model to new labels with only a few labeled examples. Previous studies mainly focus on scenario of single category label per example but have not solved more challenging multi-label exponential-sized output space and low-data effectively. In this paper, we propose semantic-aware meta-learning for few-shot learning. Our approach learn infer semantic correlation between unseen historical quickly tasks from Specifically, features be mapped into...