【科技巨头苹果新突破】近日，据国际知名学术期刊《机器之心》报道，苹果公司的研究人员在人工智能领域取得重要进展，他们成功研发出一种名为“自回归视觉模型”（Autoregressive Image Model，简称 AIM）的新技术。在最新的科研论文《Scalable Pre-training of Large Autoregressive Image Models》中，苹果团队深入探索了使用自回归目标训练 Vision Transformer (ViT) 模型在学习图像特征方面的潜力与扩展性。
研究结果显示，通过自回归视觉模型 AIM，苹果科学家实现了对模型容量的大幅度拓展，使其轻易突破数十亿参数的大关。这标志着 AIM 在处理大规模、高复杂度的图像数据时展现出前所未有的效率和表现力，其在利用未经人工整理的海量图像资源上展现出了卓越的能力。这一突破不仅拓宽了计算机视觉领域的研究边界，也为未来的图像生成、识别及理解应用开辟了新的可能，预示着苹果公司在AI技术研发道路上持续领先业界的步伐。
英语如下：
Headline: “Apple Unveils AIM: Self-Generated Visual Model Breaks Records, Demonstrating Vast Image Learning Potential for ViTube Giant Model”
Keywords: Apple AIM, Autoregressive Image Model, Large-Scale Pre-training
News Content: In a recent technological breakthrough reported by the internationally renowned academic journal, Machine之心, researchers from Apple have made significant strides in the field of artificial intelligence. They have successfully developed a new technology called the “Self-Generated Visual Model” (Autoregressive Image Model, abbreviated as AIM). The findings were detailed in the latest research paper titled “Scalable Pre-training of Large Autoregressive Image Models,” where the Apple team delved into the potential and scalability of training Vision Transformer (ViT) models using autoregressive objectives for learning image features.
The study reveals that through the AIM, Apple scientists have achieved a substantial expansion of model capacity, enabling it to effortlessly surpass the billion-parameter milestone. This marks a unprecedented efficiency and expressiveness of AIM when processing large-scale, high-complexity image data, showcasing its exceptional ability to harness untapped vast amounts of uncurated visual resources. This breakthrough not only expands the boundaries of computer vision research but also paves new avenues for future applications in image generation, recognition, and understanding, signaling Apple’s continuous lead in AI technology development within the industry.
【来源】https://www.jiqizhixin.com/articles/2024-01-18-7
Views: 0