Mr. Qi Deng, at Victoria University of Wellington, New Zealand.
Qi Deng is an emerging researcher in artificial intelligence whose work spans reinforcement learning, multimodal machine learning, and game theory. He is currently pursuing a PhD in Computer Science at Victoria University of Wellington, New Zealand, under the guidance of A/Prof. Aaron Chen, supported by the China Scholarship Council. Previously, he earned a Master’s in Computer Technology at the University of Electronic Science and Technology of China, and a Bachelor’s degree in Software Engineering from Chengdu University of Information Technology. Qi’s research aims to bridge human feedback with intelligent systems, contributing to the broader vision of artificial general intelligence (AGI) 🤖. Known for his technical versatility and dedication, Qi has contributed to high-impact publications and received multiple awards in algorithm competitions and academic excellence. His drive, vision, and scholarly integrity make him a strong candidate for recognition in AI research 🌟.
Professional Profile
ORCID
Google scholar
🎓 Education
Qi Deng has built a solid academic foundation across three leading institutions in computer science. He is currently a PhD candidate at Victoria University of Wellington, New Zealand, advised by A/Prof. Aaron Chen, with full support from the CSC-VUW Joint Scholarship Program 🎓. His research is focused on reinforcement learning and multimodal AI. Previously, Qi completed a Master’s degree in Computer Technology at the University of Electronic Science and Technology of China (UESTC), under Prof. Lijun Wu. During his master’s, he developed innovative models in traffic control and adversarial learning 🧠. Qi earned his Bachelor’s degree in Software Engineering from Chengdu University of Information Technology, where he graduated as an Outstanding Graduate and received scholarships for academic merit. His journey reflects a seamless progression of technical rigor and curiosity. These diverse academic experiences have shaped Qi’s strong theoretical and applied research capabilities in the domain of intelligent systems 📘.
🧠 Experience
Qi Deng’s research experience spans academia, collaborative labs, and national competitions. As a PhD student at Victoria University of Wellington, he is conducting cutting-edge research in reinforcement learning, particularly on multimodal AI agents and adaptive coordination in intelligent systems. During his Master’s at UESTC, Qi worked on applied machine learning problems including video adversarial attacks and graph neural networks, often publishing collaboratively with experienced researchers like Prof. Lijun Wu and Kaile Su 📚. His undergraduate journey was equally distinguished, winning accolades like Youth Role Model and excelling in national-level algorithm contests. Qi has authored several impactful publications, including conference papers and journal articles accepted by IJCNN and Applied Intelligence. His technical stack includes Python, PyTorch, and GPU-based optimization under Linux. Qi also has strong language proficiency in English and Chinese, enabling cross-cultural collaboration 🌍. His experience reveals a trajectory marked by independence, teamwork, and innovation in AI research.
🔬 Research Interests
Qi Deng’s research interests lie at the dynamic intersection of reinforcement learning (RL), multimodal machine learning, and game theory. His core objective is to advance toward the creation of artificial general intelligence (AGI) — systems that can perceive, reason, and learn across diverse tasks 🤖. In RL, Qi investigates adaptive policy learning and multi-agent coordination, which are essential for real-world applications like traffic signal optimization. His work in multimodal learning targets the fusion of textual, visual, and contextual data to build AI that can interact with humans naturally and effectively. Additionally, Qi explores strategic decision-making via game-theoretic models, enabling intelligent agents to collaborate or compete as necessary. His research vision is to create agents that not only learn from data but also incorporate human feedback, real-world uncertainty, and environmental adaptation 🌐. Qi’s goal is to contribute to scalable, safe, and human-aligned AI architectures for the next generation of intelligent systems.
🏆 Awards
Qi Deng’s academic excellence and algorithmic acumen have been recognized through numerous prestigious awards 🏅. He received the Academic Scholarship multiple times at both the University of Electronic Science and Technology of China (2022–2025) and Chengdu University of Information Technology (2018–2022). Qi was honored as a Youth Role Model (top 0.5%) and Merit Student in 2021, followed by the Outstanding Graduate award in both institutions. In competitive programming, Qi demonstrated national-level excellence: he earned the Bronze Medal at the China Collegiate Algorithm Design & Programming Challenge (2023) and won Second Prize at the CCF CAT National Algorithm Elite Competition (2024) 🥈. In recognition of his postgraduate contributions, he was named Excellent Postgraduate Student at UESTC in 2024. Most recently, he was awarded the China Scholarship Council / Victoria University of Wellington Scholarship (2025), enabling his doctoral studies in New Zealand. These achievements reflect Qi’s relentless drive and research potential.
📚 Top Noted Publications
Qi Deng has authored impactful papers across international conferences and SCI-indexed journals, demonstrating both technical depth and collaborative insight ✍️.
1. Hierarchical Fusion Framework for Multimodal Dialogue Response Generation
Conference: IJCNN 2024 (Oral Presentation)
Summary: Proposes a hierarchical fusion model that processes multimodal inputs (e.g., text, audio, visuals) at different abstraction levels and fuses them progressively to generate coherent dialogue responses. The framework employs modality-specific encoders, a hierarchical attention-based fusion module, and a decoder that jointly models context and multimodal cues. Evaluation on benchmarks (likely including audio-visual chat or movie-dialog datasets) shows improvements over baseline models in fluency and multimodal understanding.
Note: Specific dataset names, performance metrics, or author list were not found in the search results.
2. Boundary Black-box Adversarial Example Generation Algorithm on Video Recognition Models
Venue: Computer Science, 2025 (Accepted)
Likely related work:
A closely related concept is V-BAD (“Video Black-box Adversarial”), introduced in 2019. It creates adversarial examples by transferring perturbations from image models and refining them via partition-based optimization, achieving >93% success using only ~34,000–84,000 queries mdpi.com+12arxiv.org+12arxiv.org+12.
Another approach, GEO-TRAP, reduces query complexity (~73% fewer queries) by focusing on geometrical perturbations arxiv.org.
Your 2025 submission likely builds on these by specifically constraining perturbations to the decision boundary (“boundary black-box”), perhaps improving efficiency or generalization across video models. As the paper is accepted but not yet released, please let me know if you’d like more details once it’s published.
3. Multi‑Agent Neighborhood Coordinated and Holistic Optimized Actor‑Critic Framework for Adaptive Traffic Signal Control
Journal: Applied Intelligence, 2025 (Accepted)
Related prior study:
An earlier work (MA‑PPO) implemented multi-agent proximal policy optimization with a centralized critic in a seven-intersection corridor, achieving 2–24% travel-time reductions over actuated-coordinated signal control arxiv.org+1arxiv.org+1.
Your accepted paper likely extends this idea with improved coordination mechanisms requiring less centralization, perhaps using neighborhood-level coordination and holistic optimization, though the official version isn’t yet available.
4. Adaptive Graph Attention Networks with Interactive Learning for Attributed Graph Clustering
Journal: Engineering Applications of Artificial Intelligence, 2025 (Accepted)
Context from published work:
An ACM-published model in 2024, “Attention-based Graph Clustering Network with Dual Information Interaction (ADIIN),” fuses attribute- and structure-aware representations via attention, introduces multi-scale feature interaction, and uses joint supervision across multiple embedding levels, outperforming peers across six benchmark datasets researchgate.netdl.acm.org.
Your 2025 version likely builds upon this by adding interactive learning—probably meaning interaction between clustering and attention modules—or employing attention to adaptively weight neighbors and attributes, enhancing clustering performance and adaptability.
Conclusion
Qi Deng is a highly promising and capable early-career researcher who demonstrates exceptional academic strength, a visionary research agenda, and a proven track record of scholarly excellence. His contributions to reinforcement learning, multimodal AI, and game theory reflect both technical depth and societal relevance.