Presentation Lineup

NAVER has grown from Korea’s leading IT company into a global platform company spanning search, commerce, fintech, content, and cloud. This keynote traces NAVER’s journey toward AI-native products, including its Korean-centric foundation model HyperCLOVA X, and AI-native services, and shows how each step builds on the last. Together, these efforts mark a shift toward proactive agents that infer user intent, plan, and complete real-world tasks across NAVER’s services. This talk will close a look at with where this is heading: a flywheel that compounds data, services, and research together.

Since LLMs’ late emergence, “Agentic AI” has rapidly reshaped software and high-income professions while converging with robotics to advance “Physical AI” for industrial and everyday use. Following LG AI Research’s establishment in December 2020, EXAONE development began in 2021 under a government-designated AI program, evolving into the global frontier model K-EXAONE. This progress enabled LG AI Research to secure global recognition in the Stanford HAI AI Index Report. Today, EXAONE delivers tangible results across manufacturing, R&D, healthcare, finance, and public services through collaborations with LG affiliates, domestic institutions, and global partners. As industries worldwide accelerate AI transformation (AX), most remain in early implementation stages, addressing data integration and structural shifts while acknowledging AI’s pivotal role in national competitiveness—projected to fundamentally reshape research and industrial ecosystems—making academia-industry partnerships essential for sustainable positioning. Continued AI technological advancement is expected to significantly transform the shape of our future.

The rise of LLM-based AI services is making memory a central bottleneck in computing infrastructure. Transformer-based models require massive memory capacity and bandwidth for model parameters, long-context processing, and inference-time data movement. As a result, memory cost and supply constraints are becoming critical challenges for scalable AI services. To overcome this, research is advancing in two complementary directions: reducing memory demand through efficient algorithms, and using memory more effectively to reduce computation and improve system-level efficiency. Both ultimately aim to lower the cost of AI services and enable broader market growth. This challenge will become even more important in the era of agentic AI, where autonomous agents may operate continuously and dramatically increase infrastructure demand. In this talk, we discuss why memory is becoming the defining factor of AI efficiency and how SK Hynix aims to enable scalable AI infrastructure through next-generation AI memory solutions.

This talk introduces the strategic direction of manufacturing AI and presents the transformation journey of HD Hyundai in adopting AI-driven innovation. As the manufacturing industry faces increasing complexity and demand for efficiency, AI has become a critical enabler for operational excellence and sustainable growth. In this presentation, we discuss the key challenges in implementing AI within large-scale industrial environments, including data integration, legacy system constraints, and organizational transformation. HD Hyundai’s approach focuses on building a unified AI framework, leveraging advanced analytics, and integrating AI into core production processes. We share real-world use cases demonstrating how AI applications have improved productivity, optimized operations, and enhanced decision-making capabilities across multiple business units. Furthermore, we highlight lessons learned from the company’s AI transformation journey, including governance models and cross-functional collaboration. Finally, the talk outlines future directions for manufacturing AI and emphasizes the importance of continuous innovation in shaping the next generation of smart industries.

Games are among the most demanding testbeds for applied AI. They combine real-time interaction, massive multimodal data, adversarial behavior, and millions of concurrent users whose experience hinges on every millisecond. This keynote presents how KRAFTON, the game company behind the PUBG franchise, has been embedding AI as part of both gameplay and live operations at global scale.
We begin with a brief overview of KRAFTON’s product portfolio and global footprint, including the recognition of PUBG as an official Asian Games esports title.
The core of the talk focuses on three concrete AI initiatives that translate frontier research into production. First, Ally, an on-device AI companion designed to talk and play with the user as a duo partner, powered by a small language model tuned for character behavior and dialogue. Second, anti-cheat pipelines that leverage both gameplay logs and vision-language models to protect fair play through agentic detection workflows. Third, AI for eSports analytics, where models for win probability estimation, engagement prediction, and trajectory forecasting bring new depth to PUBG broadcasts. We close with KRAFTON’s vision for AI-native game development and discuss what running these systems at global player scale has taught us, and the research community, about deploying AI in interactive environments.

Samsung DS’s journey from deploying AI in production to building AI‑ready data and knowledge foundations is presented. Over the past years, we have integrated 400+ systems, 50,000+ tables, and 5M+ documents, enabling 1,000+ of AI agents and LLM chatbots. More recently, our focus is on making enterprise knowledge accessible; secure RAG pipelines, document access controls, and self‑service AI platforms that empower domain experts to innovate. The next decade will shift from data governance to knowledge governance, establishing trusted enterprise knowledge as the foundation for AI agents.

Clinical dermatology and the beauty industry serve different audiences—patients and consumers—yet share a common goal: promoting healthy skin. Clinical practice provides diagnostic rigor and medical expertise, whereas beauty platforms offer accessibility, scalability, and continuity of care. Recognizing these complementary strengths, we present a dermatologist-in-the-loop AI skin assessment framework developed through a collaboration between Incheon St. Mary’s Hospital and AMOREPACIFIC.
While AI-based skin analysis can support everyday skincare, accurate evaluation of sensitive skin and skin lesions requires clinically informed interpretation. To address this challenge, dermatologist expertise was integrated throughout the development process, including data acquisition, annotation, model design, and validation. As an initial step, dedicated AI models for skin-lesion analysis were developed and evaluated against dermatologist reference standards.
Building on this foundation, we propose a future roadmap incorporating multimodal skin foundation models, explainable AI, dynamic skin-state prediction, and privacy-preserving federated learning. Beyond research, these technologies are being translated into consumer-facing digital and retail platforms to expand access to evidence-based skincare guidance.