Apple's Latest AI Research Could Revolutionize iPhone Experience
-
Recently, California-based tech giant Apple has made another major breakthrough in the field of artificial intelligence, introducing new technologies about 3D avatars and efficient language model inference through two new research papers. These innovations promise to deliver more immersive visual experiences and enable complex AI systems to run on consumer devices like iPhones and iPads.
HUGS Technology: Generating Realistic 3D Avatars from Videos
In the first research paper, Apple scientists introduced HUGS (Human Gaussian Splats) technology, which generates dynamic 3D avatars from short monocular videos (i.e., videos captured by a single camera). Lead author Muhammed Kocabas stated: "Our method uses only a monocular video containing a small number of frames (50-100 frames) and automatically learns to separate static scenes from fully animatable human avatars within 30 minutes."
HUGS employs 3D Gaussian splatting to represent both the human figure and the background scene, utilizing an efficient rendering technique. Compared to previous avatar generation methods, HUGS achieves a 100x improvement in training and rendering speed. After 30 minutes of optimization on a typical gaming GPU, it demonstrates realistic results and surpasses existing technologies in 3D reconstruction quality.
Expanding Memory Space in AI Inference
In the second paper, Apple researchers tackle the critical challenge of deploying large language models (LLMs) on devices with limited memory. Modern natural language models, such as GPT-4, contain hundreds of billions of parameters, making inference on consumer hardware expensive.
Their proposed system minimizes the amount of data transferred from flash storage to limited DRAM during inference by constructing a cost model aligned with flash memory behavior, optimizing two key areas: reducing data transfer volume and reading data in larger, more contiguous chunks. On Apple M1 Max CPUs, these methods improved inference latency by 4-5 times, while achieving 20-25 times acceleration on GPUs. This breakthrough is crucial for deploying advanced LLMs in resource-constrained environments, promising smooth operation of complex AI assistants and chatbots on iPhones, iPads, and other mobile devices.
These two papers demonstrate Apple's growing leadership in AI research and application. Despite the promising prospects of these innovations, experts still caution Apple to proceed carefully when integrating these technologies into consumer products. From privacy protection to preventing misuse, social impacts must be fully considered. By incorporating these innovations into its product line, Apple is not only enhancing device capabilities but also anticipating future demands for AI-powered services. Enabling more sophisticated AI models to run on memory-limited devices, Apple has the potential to create new possibilities for previously unattainable applications and services.
Additionally, by publishing these studies, Apple is contributing to the broader AI community, potentially inspiring further developments in the field.
Apple's latest research achievements mark substantial progress in artificial intelligence, empowering devices such as iPhones and iPads with more robust functionalities. Through HUGS technology and memory optimization approaches, Apple provides users with realistic 3D avatar experiences and more efficient AI inference. However, as these innovations gradually integrate into products, Apple must remain highly vigilant about societal impacts including privacy concerns and potential misuse. Overall, these innovations open new doors for the future of AI, delivering richer and more advanced experiences to users.