Zijian (Longino) ZHAO 赵子健
The Hong Kong University of Science and Technology
Biography
I’m a Ph.D. student in Department of Civil and Environmental Engineering and Center for Scientific Computation at The Hong Kong University of Science and Technology, under the supervision of Prof. Sen Li. My current research primarily focuses on Deep Reinforcement Learning, Intelligent Transportation, Mobile Computing, and AI Ethics.
I received my B.Eng. degree from School of Computer Science and Engineering at Sun Yat-sen University in 2024, mentored by Prof. Kai Huang (Robot RL), Prof. Ning Liu and Prof. Chengying Gao (Music AI). I was also a visiting student in Shenzhen Research Institute of Big Data at The Chinese University of Hong Kong (Shenzhen) from 2023 to 2024, under the guidance of Prof. Guangxu Zhu (Deep Wireless Sensing).
Aside from academics, I’m also the vocalist and guitarist for several metal and rock bands, including Tokamak Disruption, NEWS (personal band) and Rights of Lethe (dissolved). If you are interested in collaborating with me (whether in academia or music), please feel free to reach out!
- Deep Learning & Machine Learning & Reinforcement Learning & Multi-Agent
- Interdisciplinary AI & AI Ethics & AI Arts
- Smart City & Intelligent Transportation & Mobile Computing
- Integrated Sensing, AI, and Communication & Network and Edge Intelligence
- Music Information Retrieval & Natural Language Processing & Multi-Modal & Multimedia
Ph.D. in Civil Engineering (Scientific Computation), 2024~
The Hong Kong University of Science and Technology (Clearwater Bay Campus, Hong Kong)
B.Eng. in Computer Science and Technology (National Basic Subject Talent Training Plan), 2020~2024
Sun Yat-sen University (Guangzhou Campus)
Latest News
2025.10: Our paper “Enhancing Reasoning Reliability through Formal Logic Verification in Reinforcement Learning” has been submitted for possible publication.
2025.09: Our paper “The Impacts of Data Privacy Regulations on Food-Delivery Platforms” has been accepted by Transportation Research Part C.
2025.09: Our papers “Ride-Hailing Vehicle Dispatch with a Mixture of On-Demand and Pre-Booked Requests: A Multi-Agent Reinforcement Learning Approach” and “Zero-Effort Image-to-Music Generation: An Interpretable RAG-Based VLM Approach” have been submitted for possible publication.
2025.09: Our proposal “Towards Fairness in Transportation Gig Markets: Identifying, Imitating, and Mitigating Algorithm Discrimination via Deep Reinforcement Learning” has been submitted for possible publication.
2025.09: Our papers “LoFi: Vision-Aided Label Generator for Wi-Fi Localization and Tracking” and “Modelling the 5G Energy Consumption using Real-world Data: Energy Fingerprint is All You Need” have been accepted by IEEE Globecom GenAI NGN Workshop 2025.
2025.09: Our paper “IfWorld: A Multi-Agent Framework for Cross-Disciplinary Counterfactual Scenario Reasoning” has been submitted for possible publication.
2025.08: We have uploaded the datasets WiGesture, WiFall, WiCount, LoFi, and RPMC-L2 in Hugging Face.
2025.08: Our paper “Towards Advanced Mathematical Reasoning for LLMs via First-Order Logic Theorem Proving” has been accepted by EMNLP 2025 (main conference).
2025.08: We have uploaded the code and paramters of CSI-BERT, CSI-BERT2, Adversarial-MidiBERT, PianoBART, and Skip-BART in Hugging Face.
2025.08: Our paper “CSI-BERT2: A BERT-inspired Framework for Efficient CSI Prediction and Classification in Wireless Communication and Sensing” has been submitted for possible publication.
2025.07: Our papers “One Step is Enough: Multi-Agent Reinforcement Learning based on One-Step Policy Optimization for Order Dispatch on Ride-Sharing Platforms” and “Each to Their Own: Exploring the Optimal Embedding in RAG” have been submitted for possible publication.
2025.06: Our paper “Does MMD Really Align? A Cross Domain Wireless Sensing Method via Local Distribution” has been accepted by IEEE/CIC ICCC 2025.
2025.06: Our paper “An Overview of Domain-Specific Foundation Model: Key Technologies, Applications and Challenges” has been accepted by SCIS.
2025.05: Our papers “Triple-BERT: Do We Really Need MARL for Ride-Sharing Order Dispatch?”, “Automatic Stage Lighting Control: Is it a Rule-Driven Process or Generative Task?”, and “Towards Advanced Mathematical Reasoning for LLMs via First-Order Logic Theorem Proving” have been submitted for possible publication.
2025.05: Our technical report “A Short Overview of Multi-Modal Wi-Fi Sensing” has been available in ArXiv.
2025.04: Our paper “Let Network Decide What to Learn: Symbolic Music Understanding Model Based on Large-scale Adversarial Pre-training” has been accepted by ACM ICMR 2025.
2025.04: Our paper “Multi-Agent Reinforcement Learning for Order Assignment and Payment Setting on Food-Delivery Platforms: The Implicit Algorithmic Biases” has been accepted by ISTDM 2025.
2025.04: Our paper “The Impacts of Data Privacy Regulations on Food-Delivery Platforms” has been submitted for possible publication.
2025.03: Our technical report “Label Unbalance in High-frequency Trading” has been available in ArXiv and has been reported in QuantML.
2025.02: We have uploaded the WiGesture dataset in Sensing Dataset Platform (SDP), which can be used for Wi-Fi gesture recognition, people identification, and cross-domain tasks.
2025.02: Our paper “CrossFi: A Cross Domain Wi-Fi Sensing Framework Based on Siamese Network” has been accepted by IEEE IOT-J.
2025.02: We have released the first stage lighting dataset Rock, Punk, Metal, Core - Livehouse Lighting (RPMC-L2). The corresponding paper will be avilable soon.
2025.02: I am awarded the Elite Scholarship from Likelihood Lab.
2025.01: Our papers “KNN-MMD: Cross Domain Wireless Sensing via Local Distribution Alignment” and a conference version “Does MMD Really Align? A Cross Domain Wireless Sensing Method via Local Distribution” and “LoFi: Vision-Aided Label Generator for Wi-Fi Localization and Tracking” have been submitted for possible publication.
2025.01: Our paper “Discriminatory Order Assignment and Payment-Setting of On-Demand Food-Delivery Platforms: A Multi-Action and Multi-Agent Reinforcement Learning Framework” and a conference version “Multi-Agent Reinforcement Learning for Order Assignment and Payment Setting on Food-Delivery Platforms: The Implicit Algorithmic Biases” have been submitted for possible publication.
Activity and Service
Code: C/C++ (CCF-CSP:320, Top 0.8%), Python, Matlab Music: Guitar, Bass, Keyboard
Interview: HKUST JUPAS 2025
Society Membership: IEEE Student Member, ACM Student Member, CCF Student MemberTPC Membership: IEEE WCNC Workshop 2024-2025, IEEE PIMRC Workshop 2024-2025, IEEE Globecom Workshop 2025, IEEE/CIC ICCC Workshop 2025Technical Reviewer: IEEE ICASSP 2024-2026, AAAI 2026, IEEE ICME 2024-2025, IEEE WCNC 2024-2025, IEEE PIMRC 2024-2025, HKSTS 2024-2025, ICLR 2025, ACL ARR 2025, IEEE Globecom 2025, IEEE IJCNN 2025, IEEE/CIC ICCC 2025, IEEE AVSS 2025, IEEE MLSP 2025, IET IRC 2025, BTR 2025, IEEE SMC 2023, IEEE TMC, IEEE WCL, MTAP, Bentham Science Book
Experience
Accomplishments
Featured Publications
This paper examines the impacts of data privacy regulations (such as the General Data Protection Regulation), on the on-demand food-delivery market. Specifically, we consider a food-delivery platform that determines order bundling, order assignment, and courier payments at each step, alongside a group of couriers who make short-term decisions on whether to accept assigned orders and long-term decisions on whether to allow the platform to use their historical behavioral data (e.g., order acceptance and rejection history) for operational decision-making. We formulate a Markov Decision Process (MDP) to simulate the platform’s operational strategies and a Multi-Agent Contextual Multi-Armed Bandit (MA-CMAB) framework to simulate the couriers’ decisions under the data privacy regulation. The platform’s MDP involves mixed-integer decisions, and a novel hybrid multi-agent reinforcement learning framework is proposed to combine a Double Deep Q-Network (DDQN) for discrete order assignment strategies and Proximal Policy Optimization with KL and CLIP (PPO-KL-CLIP) for continuous payment decisions. For the couriers, we develop a Maximum Likelihood Estimation (MLE)-based Thompson Sampling method to derive their optimal strategies. To address the interaction between the platform and the couriers, we employ a two-stage training framework that the first stage trains a general policy for the platform that adapts to any courier strategy, while the second stage derives the optimal strategies for the couriers. The proposed model and algorithm are validated using real-world food-delivery data from Hong Kong, comparing scenarios under data privacy regulations with a benchmark case without such regulations. Interestingly, the findings reveal that, contrary to initial expectations, data privacy regulations not only protect couriers but may also result in higher platform profits and improved customer experiences. By giving couriers the flexibility to decide whether to share their work-related data for order assignments and payment decisions, the regulations attract more active couriers, thereby improving overall system performance. Specifically, the number of active couriers increases under regulation, particularly during peak hours, leading to more food-delivery orders served and higher platform profits. Customers also benefit from shorter delivery times and lower overtime rates. These findings highlight the potential of data privacy regulations to reshape labor dynamics in the gig economy, creating a win-win scenario for platforms, couriers, and customers. The code of this paper is publicly available at https://github.com/RS2002/GDPR-Food-Delivery .
In recent years, Wi-Fi sensing has garnered significant attention due to its numerous benefits, such as privacy protection, low cost, and penetration ability. Extensive research has been conducted in this field, focusing on areas such as gesture recognition, people identification, and fall detection. However, many data-driven methods encounter challenges related to domain shift, where the model fails to perform well in environments different from the training data. One major factor contributing to this issue is the limited availability of Wi-Fi sensing datasets, which makes models learn excessive irrelevant information and over-fit to the training set. Unfortunately, collecting large-scale Wi-Fi sensing datasets across diverse scenarios is a challenging task. To address this problem, we propose CrossFi, a siamese network-based approach that excels in both in-domain scenario and cross-domain scenario, including few-shot, zero-shot scenarios, and even works in few-shot new-class scenario where testing set contains new categories. The core component of CrossFi is a sample-similarity calculation network called CSi-Net, which improves the structure of the siamese network by using an attention mechanism to capture similarity information, instead of simply calculating the distance or cosine similarity. Based on it, we develop an extra Weight-Net that can generate a template for each class, so that our CrossFi can work in different scenarios. Experimental results demonstrate that our CrossFi achieves state-of-the-art performance across various scenarios. In gesture recognition task, our CrossFi achieves an accuracy of 98.17% in in-domain scenario, 91.72% in one-shot cross-domain scenario, 64.81% in zero-shot cross-domain scenario, and 84.75% in one-shot new-class scenario. The code for our model is publicly available at https://github.com/RS2002/CrossFi.
Despite the development of various deep learning methods for Wi-Fi sensing, package loss often results in noncontinuous estimation of the Channel State Information (CSI), which negatively impacts the performance of the learning models. To overcome this challenge, we propose a deep learning model based on Bidirectional Encoder Representations from Transformers (BERT) for CSI recovery, named CSI-BERT. CSI-BERT can be trained in an self-supervised manner on the target dataset without the need for additional data. Furthermore, unlike traditional interpolation methods that focus on one subcarrier at a time, CSI-BERT captures the sequential relationships across different subcarriers. Experimental results demonstrate that CSIBERT achieves lower error rates and faster speed compared to traditional interpolation methods, even when facing with high loss rates. Moreover, by harnessing the recovered CSI obtained from CSI-BERT, other deep learning models like Residual Network and Recurrent Neural Network can achieve an average increase in accuracy of approximately 15% in Wi-Fi sensing tasks. The collected dataset WiGesture and code for our model are publicly available at https://github.com/RS2002/CSI-BERT.
Recent Publications
Patents
Talks
