据机器之心报道,苹果的研究人员最近在一篇论文中提出了名为“可扩展的大规模自回归图像模型”(AIM)的新型深度学习模型。该模型可以扩展到数十亿个参数,能够高效地利用大量未标注的图像数据进行训练。研究人员发现,AIM模型在表征学习方面展现出与大语言模型类似的潜力。相比当前主流的生成模型,AIM具有更强的建模能力和泛化能力。随着计算能力的提升,预计未来超大规模自回归模型会取得更多应用。
Title: Apple Develops Massive Autoregressive Visual Model
Keywords: Image model, Autoregressive, Large model
News content: According to Machine Heart, Apple researchers recently proposed a new deep learning model called “Scalable Pre-training of Large Autoregressive Image Models” (AIM) in a paper. The model can scale up to billions of parameters and efficiently utilize large amounts of unlabeled image data for training. Researchers found that the AIM model demonstrates potential in representation learning similar to large language models. Compared to current mainstream generative models, AIM has stronger modeling and generalization capabilities. With the improvement of computing power, it is expected that ultra-large-scale autoregressive models will gain more applications in the future.
【来源】https://www.jiqizhixin.com/articles/2024-01-18-7
Views: 1