Five FPT AI Research Papers Accepted by Leading Global Scientific Conferences

Five AI research papers involving FPT researchers and students have successfully passed international peer review and been accepted for publication at three prestigious scientific conferences. The achievement further affirms FPT’s growing research capabilities in artificial intelligence and demonstrates that Vietnamese researchers can contribute to solving globally significant scientific challenges.

The three conferences are ACL, PAKDD, and ICML 2026. This year, ACL received 12,148 submissions, with only 19% accepted into the main conference and 18% published in the Findings track. PAKDD accepted 184 papers out of 728 submissions, equivalent to an acceptance rate of approximately 25%. ICML recorded more than 23,900 submissions after the preliminary screening stage, with an acceptance rate of about 26.6%.

The conferences will take place at major global science and technology hubs: PAKDD in Hong Kong (June 9–12), ACL in San Diego, United States (July 2–7), and ICML in Seoul, South Korea (July 6–11).

Expanding AI’s Creative Capabilities

At PAKDD 2026, Associate Professor Dr. Phan Duy Hung, Deputy Director of the Quantum AI & Cyber Security Institute (QACI), FPT Corporation, together with Nguyen Dinh Hieu and Tran Minh Khuong, students of FPT University, developed the research paper “EquiFashion: Hybrid GAN–Diffusion Balancing Diversity–Fidelity for Fashion Design Generation.”

The EquiFashion project addresses a key challenge in AI image generation: producing a wide variety of designs while maintaining realism and adherence to user requirements. Today, fashion brands, e-commerce platforms, and advertising agencies increasingly use AI-generated product images to reduce production time and costs. However, existing tools still face limitations. When users describe clothing with specific colors, collars, or materials, AI models may misinterpret the request and generate unrealistic designs. As a result, users often need multiple rounds of revisions, consuming additional time and computational resources.

The EquiFashion research team at PAKDD 2026

To address this issue, the team proposed a two-stage approach. The first stage enables AI to generate a diverse set of design options, while the second stage refines the images to make garments appear more realistic, accurately capture details, and better align with the original description.

The project also introduces a dataset of 350,000 fashion image samples, covering more than 40 apparel categories, accompanied by text descriptions, sketches, model poses, and fabric information. As a result, the generated images achieve significantly higher realism and diversity compared to existing approaches.

Making Generative AI More Efficient

At ACL 2026, FPT had three research papers accepted: FastDiSS, CodeWiki, and SpecMind.

Among them, FastDiSS was developed by Dr. Hoang Thanh Tung, Director of the Multimodal AI Lab (QACI), FPT Corporation, together with researchers from FPT’s AI Residency Program.

The research focuses on diffusion language models, an emerging approach for AI text generation, with the goal of improving both efficiency and processing speed. AI systems must not only produce high-quality responses but also operate quickly and cost-effectively to serve large numbers of users simultaneously. Slow response times or excessive computational costs make large-scale deployment difficult.

The team proposed a new training methodology that enables AI models to perform fewer inference steps while maintaining output quality. On benchmark datasets, the method achieved up to 400 times faster text generation than conventional approaches without sacrificing performance.

This advancement is particularly relevant for digital assistants, customer-service chatbots, online education platforms, and e-commerce applications. By improving speed and reducing costs, AI becomes more practical for deployment in real-world products, ultimately providing users with smarter, faster, and more reliable services.

Dr. Hoang Thanh Tung, Director of the Multimodal AI Lab (QACI), FPT Corporation

In the CodeWiki project, researchers from the FPT AI Residency Program highlighted a common challenge in modern software systems: complex codebases are often supported by incomplete, outdated, or fragmented technical documentation.

When documentation is missing or poorly maintained, engineers spend substantial time understanding how systems operate, making debugging, upgrades, and project handovers more difficult.

CodeWiki aims to enable AI to automatically generate technical documentation for large software systems such as banking applications, shopping platforms, ticket-booking systems, and enterprise management software. The system can produce comprehensive documentation for an entire software project, including visual diagrams, across seven programming languages.

Researchers from the FPT AI Residency Program

In the SpecMind project, the research team focused on helping AI verify whether software behaves as expected.

During software development, engineers must define conditions for validating program correctness, such as checking whether outputs follow the correct format, calculations produce accurate results, or error-handling mechanisms function properly. This process is time-consuming, often manual, and prone to oversight.

SpecMind introduces an approach that allows AI to support this process more effectively. Instead of producing a single response, the AI can reason through multiple iterations, refine its analysis based on feedback, and progressively improve verification results. This method has the potential to reduce manual review efforts and improve productivity throughout the software development lifecycle.

Helping AI Make Better Decisions in Complex Environments

At ICML 2026, Dr. Tran The Hung, Director of the Decision Intelligence & Optimization Lab (QACI), FPT Corporation, together with colleagues and students, gained recognition with the paper “Variance Driven Exploration: A Provable and Efficient Methodology for Pure Exploration in Highly Stochastic Environments.”

Dr. Tran The Hung, Director of the Decision Intelligence & Optimization Lab (QACI), FPT Corporation

In reality, AI systems often operate in uncertain environments rather than idealized conditions. Applications such as autonomous robots, logistics systems, intelligent transportation networks, and process optimization platforms must constantly adapt to highly stochastic and unpredictable situations.

The research focuses on improving AI decision-making in such environments. The team proposed a novel methodology that reduces uncertainty during the learning process, thereby enhancing the effectiveness and reliability of AI systems when deployed in real-world scenarios.

The acceptance of multiple FPT research papers at ACL 2026, PAKDD 2026, and ICML 2026 highlights the research capabilities of FPT’s faculty members, researchers, and students, as well as their ability to contribute to solving global technological challenges.

Through research, education, and international collaboration, FPT continues to strengthen its foundational research capabilities and advance toward developing core technologies created and mastered by Vietnamese talent.