Recently, at EPFL's machine learning summer camp, Turing Award winner Yann LeCun's latest report "From Machine Learning to Autonomous Intelligence" elaborated on the recent specific concepts of autonomous intelligence, which is very worthy of attention!


With the continuous development of machine learning, researchers in the field began to think about a question: how far are we from artificial general intelligence (AGI)?

To achieve AGI, the most critical point is to let the machine understand how the world works and have a wide range of real-world knowledge.

This is also a question recently explored by Turing Award winner LeCun. He once said: It has been his lifelong pursuit to make machines act like humans or animals.

LeCun believes that the functioning of the animal brain can be seen as a simulation of the real world, which he calls a world model. Babies learn the basics by observing the world in the first few months of life, LeCun said. Watching a small ball drop hundreds of times, the average baby will have a basic understanding of the existence and operation of gravity even if they don't understand physics.

Not long ago, LeCun said he had built an early version of the world model for basic object recognition, and he is now working on training it to make predictions. In a paper published yesterday, LeCun detailed this vision.

picture



Expertise

, like 17

picture

AI/ML/DL Applications Today

    • Relying heavily on supervised deep learning. There are some on Deep RL.

    • Increasing reliance on pre-training self-supervised learning

ML/DL sucks now compared to humans and animals

    • Humans and animals learn models of the world

Self-Supervised Learning

    • Main problems: representing uncertainty, learning abstractions.

energy-based model

    • Contrastive Learning Methods Samples

    • Non-contrastive learning methods

picture

Deep SSL is a enabler for the next AI revolution

I will try to convince you:

    • Ditching Supervision and Reinforcement Learning

        Well, not exactly, but as much as possible.

    • Abandonment probability model

        Use an energy-based framework instead

     Abandon generative models

    Use a federated embedding architecture instead

    Using a Hierarchical Latent Variable-Based Energy Model

    Enable machines to reason and plan.

    See manuscript: "The Path to Autonomous Machine Intelligence"

    https://openreview.net/forum?id=BZ5a1r-kVsf


picture

picture

picture

picture

picture

picture


Expert knowledge and convenient viewing

Easy to download , please pay attention to the official account of Zhizhi (click the blue Zhizhi above to follow)

  • Reply " A192 " in the background to get the download link of " From Machine Learning to Autonomous Intelligence", the latest report of Turing Award winner Yann LeCun, with 192 pages of ppt and video .

picture
                       

Expertise, professional and credible AI knowledge distribution , making cognitive collaboration faster and better! Welcome to register and log in to www.zhuanzhi.ai to get 100,000 + AI (AI and military, medicine, public security, etc.) topic dry goods knowledge materials!
picture
Welcome to WeChat scan to join the AI ​​knowledge planet group , get the latest AI professional dry goods knowledge tutorial materials and communicate with experts !
picture
Click " Read the original text " to learn about the use of know- how , and view the information on obtaining 100,000 + AI topic knowledge