在人工智能领域，情感识别一直是研究的热点之一。近日，一场旨在提升情感AI能力的高规格赛事圆满落幕，这是第二届多模态情感识别挑战赛（MER24）。该赛事由清华大学陶建华教授、中国科学院自动化研究所连政、帝国理工学院Björn W.Schuller、奥卢大学赵国英以及南洋理工大学Erik Cambra联合在AI顶会IJCAI2024上发起举办。
MER24挑战赛设置了三个赛道，包括半监督学习赛道（Semi）、噪声鲁棒性赛道（Noise）和开放式词汇情感识别赛道（Ov）。其中，Semi赛道最受关注，因为它要求参赛队伍利用少量有标签和大量无标签数据视频数据来训练模型，并对模型在无标签数据集上的表现和泛化能力进行评估。
在激烈的竞争中，社交平台Soul App的语音技术团队凭借创新的技术方案赢得了Semi赛道的第一名。他们的成功关键在于改进半监督学习技术，提高了模型情感识别的性能。自5月大赛启动以来，来自全球近百支参赛队伍展开了角逐，包括知名高校和创新企业。
此次挑战赛的成功举办，不仅推动了多模态情感识别技术的发展，也加速了相关技术在真实人机交互场景中的应用。随着AI技术的不断进步，未来AI将更加理解人类的情感，提供更加人性化的服务。
英语如下：
News Title: “Domestic AI Emotion Recognition Technology Makes New Breakthrough: Soul App Wins Multimodal Challenge”
Keywords: AI Emotion, Multimodal Recognition, Technology Competition
News Content: In the field of artificial intelligence, emotion recognition has been a hot topic of research. Recently, a high-profile competition aimed at enhancing the capabilities of AI emotion recognition came to a successful conclusion. This was the second Multimodal Emotion Recognition Challenge (MER24). The competition was jointly initiated and hosted by Professor Tao Jianhua from Tsinghua University, Lian Zheng from the Institute of Automation, Chinese Academy of Sciences, Björn W. Schuller from Imperial College London, Zhao Guoying from the University of Oulu, and Erik Cambra from Nanyang Technological University at the AI top conference IJCAI2024.
The MER24 challenge set up three tracks, including the Semi-supervised Learning track (Semi), the Noise Robustness track (Noise), and the Open-vocabulary Emotion Recognition track (Ov). Among these, the Semi-supervised Learning track received the most attention, as it required competing teams to train models using only a small amount of labeled and a large amount of unlabeled video data, and to evaluate the model’s performance and generalization ability on an unlabeled dataset.
In the fierce competition, the voice technology team of the social platform Soul App won first place in the Semi-supervised Learning track with an innovative solution. Their key to success lay in improving semi-supervised learning technology, which enhanced the model’s performance in emotion recognition. Since the competition was launched in May, over 100 teams from around the globe, including prestigious universities and innovative enterprises, have engaged in the competition.
The successful hosting of this challenge not only promoted the development of multimodal emotion recognition technology but also accelerated the application of related technologies in real human-computer interaction scenarios. As AI technology continues to advance, AI will become even more capable of understanding human emotions and providing more humanized services.
【来源】https://www.jiqizhixin.com/articles/2024-07-31-9
Views: 2