Recently, Professor Wu Enda, an authoritative scholar in the field of AI and machine learning around the world and former chief scientist of Baidu, said of his next development trend of AI, "Don't believe the hype (AI advantages based on massive data). ”
We all know that the development of AI is inseparable from a large amount of data. Big data and big models constitute the two legs of AI, and one is indispensable. When it comes to AI, it will be said how big the amount of data for training the model is, so far the data parameter scale of the top AI training model has reached the trillion level, which is not small, it can be said that it is really big data.
Then AI is not necessarily the more data, the larger the model is the best, in many scenarios, especially in the traditional industry, large models pre-trained based on open data seem to be ineffective. Many times, not so much data is needed, and a large amount of data will envelop a lot of unused data, affecting the efficiency of computing and the final output.
With the implementation of data and privacy security regulations in various countries, the amount of public data that can be obtained or the amount of data that can be used legally can no longer be the same as in previous years, and the big data dividend period of AI has passed.
According to Gartner's report, by 2025, 70 percent of organizations will be forced to shift their focus from big data to small data and pan-data, and reduce the need for data from AI. But that doesn't mean the development of big data and AI will stall.
In the past, big data blindly pursued the large scale of data parameters, and the control of data quality was relatively weak, and it was not common enough in data application. On the contrary, small data and pan-data are more in line with the reality of domestic enterprises, and specific data models can be built according to specific scenarios to generate business insights and realize automated decision-making.
At present, the processing schemes of small data are also diverse, and the common ones are small sample learning, knowledge graph, transfer learning, self-supervised learning, synthetic data, etc.
Big data is good, but small data is the norm, especially after countries have introduced policies and regulations on personal data protection, and large-scale collection of personal data collection has become difficult. But even so, through scientific data collection and data governance, small data can also play a big role.
The GI big data middle office system independently developed by LightPoint Technology is to help customers establish data standards for massively dispersed and heterogeneous data, collect and store them after unifying the caliber, and then refine the value through big data, machine learning and other methods to form enterprise data assets, provide support for business insights and automated decision-making of government and enterprise units, and empower front-end business.
Data has undoubtedly become the most important factor of production in this era, as an important strategic resource of government and enterprise units, those high-quality small data, even if the parameter magnitude is not large, the potential can not be underestimated, in the near future will play a greater value.