In this blog, Alberto Bernacchia, Principal AI Research Lead at MediaTek, explores how MediaTek Research is developing new AI breakthroughs for a variety of applications.
AI is transforming almost every industry imaginable. For businesses, AI is massively increasing productivity and efficiency on a scale that we’ve never seen before. For consumers, AI opens up new possibilities for the way we interact with the world around us and get things done. All of our devices are getting smarter, more intuitive, and personalized, elevating the user experience. It’s also powerful to consider how AI is helping to bridge the digital divide, which goes hand in hand with MediaTek’s vision to make great technology more accessible to everyone.
To help accelerate the pace of AI innovation, MediaTek has a dedicated team called MediaTek Research, which is working to develop fundamental AI breakthroughs along with practical AI applications. We’re committed to empowering people, while also creating systems that are intelligent, ethical, secure, and sustainable. By pushing the boundaries of what AI can do, we can unlock new opportunities and develop new innovations that will shape our future.
The MediaTek Research team works with state-of-the-art research centers located in Cambridge and Taiwan, along with universities across the globe. We regularly present our research at the top AI conferences to share our findings with industry and academia. Some of our recent projects include improving the speed and reliability of AI model training, developing the world's first Large Language Model (LLM) in Traditional Chinese, and creating an innovative approach to wireless channel modelling.
We’re leveraging these AI developments in a variety of ways. Where possible, we’re integrating our AI innovations across our product portfolio, from the mobile segment to smart home devices, automotive, the healthcare industry, datacenters, industrial applications, and beyond. One example is the wireless channel modelling approach mentioned above, which will shape the way we design 6G communication systems. Additionally, we’re using machine learning to design chipsets more quickly and efficiently. Finding new ways to optimize circuit block layouts is more important than ever as performance demands continue to rise. MediaTek is using innovative AI algorithms that can predict the location and shape of the optimal circuit block, while also reducing errors. This allows our engineers to produce better results in less time, and helps our customers get their products to market faster.
We are aware of some of the limitations of current AI solutions. Besides the issues of privacy, safety, bias, and reliability, we are particularly focused on eliminating inequalities in the landscape of AI development. Training AI models is costly, which can be fully sustained only by a small number of companies. We would like to make AI training more accessible to everyone by significantly reducing the costs of training AI. That is a daunting task we are undertaking with our team of incredibly talented, experienced, and diverse individuals, along with our collaboration with the top universities worldwide.
We’re also laser focused on the impact of generative AI. It’s incredible to think that computers are no longer just following our instructions; they can anticipate what we need and automate our tasks in a personalized way. My colleague Federica Freddi, senior deep learning researcher at MediaTek Research, hit the nail on the head when she said that “generative AI is like giving a paintbrush to a computer and allowing it to create its own masterpiece.”
A lot of the attention around AI has been focused on artistic endeavors such as creating images and videos, and routine tasks like writing emails. However, some of the most compelling use cases are the ones that will help us to live longer, happier, and healthier lives. For example, AI can be used in medicine to develop personalized treatment plans for patients. Or think about how AI can improve climate modeling to provide better guidance on environmental policies.