Account Info
Log Out
English
Back
Log in to access Online Inquiry
Back to the Top
Chip stocks: A season of twists and turns
Views 34K Contents 211

NVIDIA's new synthetic data results: creating a "training data perpetual motion machine" for robots

AI's huge demand for data, data resources are nearly exhausted, so companies have begun to explore a "new way" to obtain data - their own "made" data. However, most of the previous synthetic data for AI model training, this time, NVIDIA for robot training to create a "data granary".

NVIDIA and the University of Texas at Austin, a new research paper, introduced a system called "MimicGen", only a small amount of human demonstration, can automatically generate large-scale robot training data sets. Jim Fan, a senior scientist at NVIDIA, said the company will open-source everything, including the generated data sets.

How large is the size of the generated data? With 10 human demonstrations, MimicGen can generate 1,000 synthetic examples; with 200 human demonstrations, MimicGen can generate 50,000 training data directly, covering 18 tasks and multiple simulated environments. $NVIDIA(NVDA.US)$ $Direxion Daily Semiconductor Bull 3x Shares ETF(SOXL.US)$ $PHLX Semiconductor Index(.SOX.US)$
Disclaimer: Community is offered by Moomoo Technologies Inc. and is for educational purposes only. Read more
5
+0
Translate
Report
65K Views
Comment
Sign in to post a comment
    990Followers
    62Following
    2715Visitors
    Follow