[ad_1]
\nThis is an excerpt of the article that was recollected from flipboard, as to generate a clip of interesting news from the topics that i think are important and I want to save for further references and share with my readers.<\/p>\n
Reducing the data-dependency of deep learning is currently among the top priorities of AI researchers.<\/p>\n
In his <\/span>keynote speech at the AAAI conference<\/a>, computer scientist Yann LeCun discussed the limits of current deep learning techniques and presented the blueprint for \u201cself-supervised learning,\u201d his roadmap to solve deep learning\u2019s data problem. LeCun is one of the <\/span>godfathers of deep learning<\/a> <\/span>and the inventor of <\/span>convolutional neural networks (CNN)<\/a>, one of the key elements that have spurred a revolution in artificial intelligence in the past decade.<\/p>\n Self-supervised learning is one of several plans to create data-efficient artificial intelligence systems. At this point, it\u2019s really hard to predict which technique will succeed in creating the next AI revolution (or if we\u2019ll end up adopting a totally different strategy). But here\u2019s what we know about LeCun\u2019s masterplan.<\/p>\n First, LeCun clarified that what is often referred to as the limitations of deep learning is, in fact, a limit of <\/span>supervised learning<\/a>. Supervised learning is the category of machine learning algorithms that require annotated training data. For instance, if you want to create an image classification model, you must train it on a vast number of images that have been labeled with their proper class.<\/p>\n \u201c[Deep learning] is not supervised learning. It\u2019s not just <\/span>neural networks<\/a>…<\/div>\nA clarification on the limits of deep learning<\/h2>\n