OEM CNC machining metal parts plastic pulley
Recently, the second meeting of the multi-modal artificial intelligence industry consortium with the theme of “creating new heights of industrial intelligence” was held in Wuhan. “Multimodal Basic Large Model Technical White Paper” led by the company. OEM CNC machining metal parts plastic pulley
“Currently, artificial intelligence is moving from single-modal, supervised learning to the era of multi-modal, self-supervised learning.” said Wang Jinqiao, researcher at the Institute of Automation, Chinese Academy of Sciences and president of the Wuhan Institute of Artificial Intelligence. The future of artificial intelligence is bound to be bright.”
Dependence on data labeling is reduced by more than an order of magnitude
To understand multimodal artificial intelligence, you first need to understand what a modality is. “Generally speaking, each source or form of information can be called a modality.” Wang Jinqiao said. OEM CNC machining metal parts plastic pulley
Humans use multimodal input and output methods in information acquisition, environmental perception, knowledge learning and expression. For example, if a person wants to find a blooming flower on a lawn, he can either look for it through the modality of vision, that is, look directly with his eyes; or he can search through the modality of smell, that is, with his nose Smell; it can also be explored through the modality of touch, that is, touch. Faced with the problem of finding flowers, generally speaking, people will adopt multimodal methods such as vision and smell. “To a certain extent, multimodal input and output methods are one of the important manifestations of human intelligence,” said Wang Jinqiao.
The current network data includes different modalities such as images, videos, texts, and audios. For artificial intelligence, in order to better grasp, analyze, and utilize data on the Internet, it is necessary to be able to systematically coordinate and analyze these multimodal data. OEM CNC machining metal parts plastic pulley
Wang Jinqiao said: “Technological innovation is one of the important driving forces to promote the development of the multimodal artificial intelligence industry. Since the start of multimodal learning in the 1970s, with the recent generative pre-training, Transformer-based bidirectional encoder expression With the rapid emergence of large-scale pre-training models, the field of artificial intelligence is undergoing a large-scale pre-training paradigm shift from supervised learning to unsupervised learning under the condition of ‘big data + big models’, and the development of multimodal artificial intelligence is ushering in a new peak.”
In recent years, multimodal artificial intelligence has made great technological breakthroughs in scene generalization and dependence on data. OEM CNC machining metal parts plastic pulley
“Multimodal artificial intelligence learns massive unlabeled data through self-supervised learning, and at the same time conducts labeled learning and fine-tuning of small data for specific task scenarios. Compared with single-modal artificial intelligence, its dependence on data labeling is reduced. more than an order of magnitude.” Wang Jinqiao said.
The “multi-modal large model + small model” model may become mainstream
In 2017, the State Council issued the “New Generation Artificial Intelligence Development Plan”, and the industrial transformation triggered by artificial intelligence technology is accelerating its evolution. At present, new industrial applications, new business forms, and new models created by artificial intelligence technology in all walks of life are constantly emerging. my country’s artificial intelligence industrialization is gaining momentum, and the multimodal artificial intelligence industry has also made many breakthroughs. OEM CNC machining metal parts plastic pulley
Today, the multimodal artificial intelligence industry is making great strides towards scenario-based and practical use. “At present, model parameters and data scale are no longer the focus of competition among R&D institutions, and the multi-modal artificial intelligence industry is moving towards a new stage of scene application.” Wang Jinqiao said.
“For example, the ‘Omnimedia Multimodal Model’ is based on the technical accumulation of the ‘Wenhai’ multimodal media big data and the ‘Zidong Taichu’ trimodal large model of the Institute of Automation, Chinese Academy of Sciences, combined with the Xinhua News Agency’s full media The massive data accumulation and media integration business scenarios are established. The ‘all-media multi-modal model’ will accelerate the application of artificial intelligence in multi-media business scenarios such as video dubbing, voice broadcasting, title summaries, and poster creation.” Wang Jinqiao said .
In addition, multi-modal artificial intelligence also has broad application scenarios in many fields such as smart cities, financial technology, and people’s livelihood services.
Wang Jinqiao said that in the future, the model of “multi-modal large model + small model” may become the mainstream of multi-modal artificial intelligence, which can effectively solve the problems of fragmentation and diversification of demand. OEM CNC machining metal parts plastic pulley
Wang Jinqiao said that my country should build a multi-modal large model ecology of the industrialization paradigm, continue to improve the domestic basic software and hardware support system, increase the integration of artificial intelligence with scientific research fields such as medicine, materials, and climate, and improve the artificial intelligence industry application policy. standards, and vigorously cultivate artificial intelligence compound talents. OEM CNC machining metal parts plastic pulley