Artificial Intelligence
We conduct fundamental research on various machine learning algorithms as well as computer vision and language understanding. Recently SAIT strengthens its fundamentals in AI research and increases competitiveness with an opportunity to work closely with AI communities.
AI Algorithms
Understanding users and environments with human-level intelligence
MACHINE LEARNING Recent technological breakthroughs in the field of artificial intelligence have been enabled by machine learning. There are still many challenges toward human level AI
including understanding the physical world, learning the common sense, utilizing the human knowledge, and figuring out causality. Our research efforts aim to movemachines and systems further along from perception to cognition. We conduct fundamental research on machine learning,
in partucular, deep learning theories and pursue innovations in core machine algorithms.
1. AI BASED MATERIAL DESIGN Developing new functional materials can take several years using conventional methods. However, by utilizing AI technology, the process can be significantly accelerated.
We are using cutting-edge algorithms such as graph neural networks for designing molecules and predicting their properties. Additionally, we are exploring using machine learning.
to generate synthesis plans that can optimize target properties and their yields. We are also working on combining machine learning algorithms with human expert knowledge to discover new materials.
2. AI FOR SEMICONDUCTORS Our AI technology deploys deep learning on semiconductor development and manufacturing to maximize R&D efficiency and achieve ultimate manufacturing performance
for automated chip design & validation, detection of potential faults, process optimization and yield prediction & analysis. We also pursue solving industrial problems by developing computer vision technology to automate the manufacturing systems.
Our research topics include depth estimation, image generation, representation learning, anomaly detection, domain adaptation, continual learning and physics based AI simulation.
COMPUTER VISION Computer Vision enables high-level understanding from visual data. It is a challenging problem for computers to understand the environments and reasoem about context from images.
While deep learning based computer vision techniques have successfully advanced applications such as biometerics , augmented reality, and driver assistive technology,
it is still behind the human-level understanding of environments in real world. We seek to invent innovative algorithms for machines to see and interpret the world,
as well as develop systems to enhance the visual systems of machines and sensors
1.CAMERA SYSTEM & IMAGE PROCESSING AI image signal processing has become a key technology for innovating image quality of mobile camera.
We are trying to transform the existing signal processing-based mobile camera ISP pipeline to AI ISP for maximizing perceptual image quality.
Major research topics are not only limited to deep learning-based SW algorithms for image restoration/enhancement.
We are intensively investigating tunable end-to-end AI ISP architecture by co-optimizing HW computing units and SW algorithms. ￦
AI camera design methodology such as deep optics will be further exploited in terms of total camera system, which covers from algorithm, computing, sensor/optics, and material.
2. AUTONOMOUS DRIVING We are creating an autonomous EV where we can carry our loved ones. The warm hearted researchers in Samsung are focusing on developing core technologies that facilitate L4 autonomous driving.
Also, Autonomous EV platform is thoroughly explored to prepare upcoming EV world. With the developed technology, we are aiming L4 Autonomous driving Sensor, SoC solution.
This solution will be scalable and fully customizable SoC solution. We also develop server-side AIs which are to enable breakthrough autonomous driving by large-scale AIs.
Our research scope spans over 2D/3D visual perception, generative AI, and multi-task/multi-modal learning. Key applications of research include: (1) an auto-label AI and (2) a synthetic data generation for AI model training.
We will seek answers to realize self-evolving AI that surpasses human visual intelligence and pursue development of safer and more reliable ADAS/AD technologies.
