Mathieu
Andrew Ng has established himself as an authority in the field of artificial intelligence through his work at Stanford University, Google Brain, and Baidu. He recently spoke to IEEE Spectrum about his current focus on Landing AI and the data-centric AI movement. Ng believes that large models trained on large data sets still have potential to grow, especially in the case of building “foundation models” for computer vision; however, he acknowledges that there is a need for small data solutions to address issues such as efficiency, accuracy, and bias. In order to build a foundation model for computer vision, Ng explains that there is a scalability problem due to the significant compute power required to process large volumes of video data. He also notes the limitations of the “big data” approach in industries where such data sets do not exist, and advocates for a shift towards “good data.” Landing AI uses pre-trained models for visual inspection, but Ng emphasizes the importance of providing tools for manufacturers to fine-tune the model with a consistent set of labeled images.
