苹果公司研究人员近日发布了一项名为自回归视觉模型（AIM）的研究成果。在论文《Scalable Pre-training of Large Autoregressive Image Models》中，研究者们探讨了利用自回归目标训练ViT模型，以期在表征学习方面实现与LLMs相同的扩展能力。研究结果表明，模型容量可以轻松扩展至数十亿个参数，AIM能够有效利用大量未经整理的图像数据。
这项创新技术有望为图像识别领域带来革命性的变革。通过自回归视觉模型，苹果研究人员成功实现了对图像的高效处理和识别，进一步提高了计算机视觉系统的性能。AIM模型的应用范围广泛，包括但不限于自动驾驶、人脸识别、自然语言处理等领域。
英文翻译：
News Title: Apple Develops Autoregressive Visual Model AIM, Breaking Boundaries in Image Recognition
Keywords: Apple, Autoregressive Visual Model, Image Recognition
News Content:
Apple researchers have recently released a research achievement titled “Scalable Pre-training of Large Autoregressive Image Models”. In this paper, researchers explored the ability of autoregressive target training for ViT models to achieve similar scalability in representation learning as LLMs. The results show that model capacity can be easily extended to hundreds of millions of parameters, and AIM can effectively utilize a large amount of unstructured image data.
This innovative technology is expected to bring about a revolutionary change in the field of image recognition. Through the autoregressive visual model, Apple researchers have successfully achieved efficient processing and recognition of images, further improving the performance of computer vision systems. The application scope of AIM model is wide, including but not limited to autonomous driving, facial recognition, natural language processing, and other fields.
【来源】https://www.jiqizhixin.com/articles/2024-01-18-7
Views: 1