‘Small data’ is necessary for intelligent machine learning
Titled ‘Small Data’s Big AI Potential’ the report underlines how, whilst ‘big data’ is commonly associated with AI, small data is also crucial in machine learning.
The report points to ‘Transfer learning’ is a key example of how small data makes a difference. Also known as ‘fine-tuning’, transfer learning is used when a developer has little data on a task of interest but a lot of data on a similar problem. By training a model using a big data set, it is possible to retrain by using a smaller, related data set.
“For example, by starting with an ImageNet classifier, researchers in Bangalore, India, used transfer learning to train a model to locate kidneys in ultrasound images using only 45 training examples,” the report explains.
Research into transfer learning is rapidly growing, especially in the US and China. CSET states that for this reason, the transfer learning approach is likely to work better than other approaches, such as reinforcement learning, and become more widely used in the future.
By allowing the use of AI with less data, organisations can strengthen progress in areas where there is not as much data, such as forecasting rare natural hazards or predicting the risk of disease for a population that does not have digital health records.
The report states that some analysts believe that AI has been applied more successfully to issues where data was most available. So, approaches such as transfer learning will be increasingly important as more organisations seek to diversify AI application areas and venture into previously underexplored domains.
Writers of the report, Husanjot Chahal and Helen Toner stated that “the existence of techniques such as transfer learning does not seem to have reached the awareness of the broader space of policy makers and business leaders in positions of making important decisions about AI funding and adoption,” despite the surge in transfer learning research.
They added, “by acknowledging the success of small data techniques like transfer learning – and allocating resources to support their widespread use – we can help overcome some of the pervasive misconceptions regarding the role of data in AI and foster innovation in new directions.”
Subscribe to our Editor's weekly newsletter