Log in
Research report
计算机行业AI前沿系列（一）：混合专家模型技术（MoE）-20230818-中信建投-24页
核心观点：混合专家模型技术(MoE)是由专家模型和门控模型组成稀疏门控制的深度学习技术。MoE通过门控网络实现任务/训练数据在不同专家模型间的分配，让每个模型专注处理其最擅长的任务，进而实现模型的稀疏性。随着大模型参数量的持续提升，MoE在提升计算效率方面展现出强大的能力。同时，近年来MoE逐渐与不同领域的模型相结合，包括自然语言、图像、推荐系统、多模态等，并展现出优秀的性能。后续MoE技术将在提
Read full report →Related industries
No related institutions
Similar reports
- Environmental stewardship for leaner AI agent systems: The case for architectural restraintArchitectural restraint can help AI agents reduce unnecessary computation, use computing resources more efficiently, and strengthen governance.Article•9-min read
- What Does Venture Capital Actually Finance? Scale and Structure around the World
- You are not a model. Don’t price per token.
- 2025 State of AI Report: The Builder’s Playbook
- AI Adoption Index
- Cursor + SpaceXAI: the fastest iterating team wins