The most popular and comprehensive Open Source ECM platform
Let us know more
Artificial Intelligence: Quality of Training Data is Key to Success
Artificial Intelligence (AI) projects create a model and then feed the model data. A general rule of thumb is that the more data that is input into the model, the better the results.
But one thing noticed is that there are often diminishing returns as more and more data is added. Part of the reason for this is that the focus is often on the quantity rather than the quality of the data. Researchers, like Andrew Ng, now promote the use of smaller high-quality data sets — they call this approach Data-Centric AI, as opposed to Model-Centric AI where greater emphasis is placed on the AI design architecture.
Peter Gao, CEO at Aquarium Learning, wrote for Medium that “the moral of this story is: if you have petabytes of data, hundreds of GPUs to scale model training with, and millions of dollars to spend on research, you can push through diminishing returns with brute force by throwing more data and compute at the problem.”
Daniel de la Fuente, Vice President at IBM, said that “no matter how advanced your AI is, it will only be as good as the data that is fed into the system. Creating a strong data foundation is always a smart move to put AI to work.”
Businesses that use AI have gotten the message. 72 percent of companies now say that data management and data quality issues can jeopardize AI achievement, according to an MIT Technology Review survey.
Laurel Ruma, Global Director of Custom Content, MIT Technology Review Insights, said that “data issues are more likely than not to be the reason if companies fail to achieve their AI goals. Improving processing speeds, governance, and quality of data, as well as its sufficiency for models, are the main data imperatives to ensure AI can be scaled.”
