Multimodal Large Language Models (MLLM) are advanced artificial intelligence models designed to understand and process multiple modalities, including text, images, audio, etc. By applying deep learning techniques and large sets of data for training, these models are able to figure out patterns and relationships between models.

Application Area 1: Natural Language Processing
- NLP model structure innovation
- Learning Feedback Enhancement
- Distributed Training
- Model Optimization
- Other related technologies.

Application Area 2: Computer Vision and Pattern Recognition
- Image Segmentation and Classification
- Image Semantic Understanding
- Semi-supervised Learning
- Weakly Supervised Learning, etc.

Application Area 3: Multimodal and Multi-perspective Learning
Explore how to combine multiple sources of information and learn information in datasets from multiple perspectives. These applications include graphic retrieval, 3D object detection, machine translation, etc.

Application Area 4: Medical Image Analysis
The purpose of Medical Image Analysis is to detect the interest areas and targets in medical images qualitatively or quantitatively, and to explore the connotation of images, so as to provide information reference for clinicians or researchers. The main research fields include image classification and segmentation, disease diagnosis, image registration, medical report generation, etc. 

AIGC aims at using artificial intelligence algorithms to generate content with certain creativity and quality. It applies model training and data learning to generate content based on input criteria and guidance. For instance, if we input keywords, descriptions or samples, AIGC will generate articles, images, audio, etc. accordingly.

Embodied intelligence is a subfield of AI research. It focuses on enabling intelligent systems to perceive, understand, and engage with their environment through actual interaction. Embodied intelligence aims to integrate sensors, action mechanisms, and decision-making capabilities to enable intelligent systems to act and interact with the physical world as humans do.

By blending perception, cognition, and behavior, embodied intelligence models are able to learn the structure and rules of the environment and take appropriate actions. This comprehensive approach allows embodied intelligence to be applied to robotics, autonomous vehicles, augmented reality and more, facilitating closer interaction and cooperation between intelligent systems and the human environment.

AI for Science means pre-training models with strong generalization capabilities, thus enabling them to tackle multiple downstream tasks related to major bottlenecks in science, which include controllable targeted peptide design, virus mutation prediction, antibody generation and key property prediction, protein structure prediction, RNA structure and function prediction, etc.