AI products
First, the data becomes more generalized and easy to obtain. Machine learning needs vast amounts of data, past data has not been so abundant, and even now there are many free data supply machine learning. Various disciplines in the various data sets has become the fuel of machine learning system, for example, Open Images Dataset is a data set consists of more than 9 million images covering thousands of types of composition.
Second, computing power has made rapid progress. Development of artificial intelligence need to be very abundant force calculation, calculation of force development is currently showing exponential growth. For example, Tensor Processing Units (TPUs), rather than within a few hours can, in just a few minutes to complete training machine learning models. TPUs set of computing power is 10,000 times 1990 computer.
Third, faster to build more complex models and techniques. For example, in the field of natural language understanding, researchers are making incredible progress
AI
data path convergence road daily, store, process (ETL process).
Currently summarize some common scenarios
delivered to the RFM model, AI product manager confirmed the significance of business data, say Joe Smith is a high-value customers, John Doe is a high-potential customers, then those data on behalf of Joe Smith is a high-value customers, RFM model recently purchased time R, purchase frequency F, the amount of consumption M, exactly how to verify that Joe Smith is a high-value customers, this time AI product manager will be on the business itself naturally present delivery model to help optimize training model.
In the process of doing, verifies find RFM model flawed, is not well on the Air Division customers intelligent identification, then the RFM model is gradually iteration LRFM model, specific system module again and then deployed for various business scenarios project.
After the bifurcation is another route for data analysis, such as a takeaway platform hopes to user history and time series data to improve consumer LineLian the target of consumer goods LineLian, then the need to consume inventory data LineLian analysis, needs analysis here LineLian past consumption data, for example LineLian past consumption are dinner (IT male overtime), POI point of consumption dinner, dinner consumption category, and so on, and then design algorithms analyze LineLian hundreds of thousands of users, from which can often set dinner to lunch also set goals for lunch.