Research Topics

Language AI

Language AI is a critical technology for Samsung to understand natural language voice or text input and generate intelligent and interesting responses. To this end, we have been working on advancing the state of the art in transformer-based natural language understanding & generation, common sense and knowledge augmentation, and leveraging large language models (LLMs) for Samsung’s unique language AI needs.

Multimodal AI

Real-world communication involves not only voice but also shared context to convey information. As AI technology advances, users anticipate AI agents to communicate in a similar multimodal way using camera or screen content. To enable responses to such user queries, we focus on advanced multimodal AI topics such as visual language grounding, emotion-aware AI, and generative AI that operates across text, audio and visual modalities.

On-device AI

Although many AI services currently rely on cloud computing, there is a growing need for on-device AI models that can provide real-time, always-available, and privacy-preserving services. To meet this need, we are working on developing lightweight transformer models or compressing large transformer models that can operate with small disk size, runtime memory, and computational resources. We are also pushing the boundary to make deep learning models tiny enough to run on chip with ultra-low power, memory and latency.