Yubei Chen is co-founder of Aizip inc., a company that builds the world’s smallest and most efficient AI models. He is also an assistant professor in the ECE Department at University of California, Davis. Chen’s research is at the intersection of computational neuroscience and deep unsupervised (self-supervised) learning, enhancing our understanding of the computational principles governing unsupervised representation learning in both brains and machines, and reshaping our insights into natural signal statistics.
Prior to joining UC Davis, Chen did his postdoc study with Prof. Yann LeCun at NYU Center for Data Science (CDS) and Meta Fundamental AI Research (FAIR). He completed his Ph.D. at Redwood Center for Theoretical Neuroscience and Berkeley AI Research (BAIR), UC Berkeley, advised by Prof. Bruno Olshausen.
Aizip develops ultra-efficient AI solutions optimized for edge devices, offering compact models for vision, audio, time-series, language, and sensor fusion applications. Its products enable tasks like face and object recognition, keyword spotting, ECG/EEG analysis, and on-device chatbots, all powered by TinyML. Through its AI nanofactory platform, Aizipline, the company accelerates model development using foundation and generative models to push toward full AI design automation. Aizip’s Gizmo series of small language models (300M–2B parameters) supports a wide range of devices, bringing intelligent capabilities to the edge.
You did your postdoc with Yann LeCun at NYU and Meta FAIR. How did working with him and your research at UC Berkeley shape your approach to building real-world AI solutions?
At Berkeley, my work was deeply rooted in scientific inquiry and mathematical rigor. My PhD research, which combined electrical engineering, computer science, and computational neuroscience, focused on understanding AI systems from a “white-box” perspective, or developing methods to reveal the underlying structures of data and learning models. I worked on building interpretable, high-performance AI models and visualization techniques that helped open up black-box AI systems.
At Meta FAIR, the focus was on engineering AI systems to achieve state-of-the-art performance at scale. With access to world-class computational resources, I explored the limits of self-supervised learning and contributed to what we now call “world models” — AI systems that learn from data and imagine possible environments. This dual experience — scientific understanding at Berkeley and engineering-driven scaling at Meta — has given me a comprehensive perspective on AI development. It highlighted the importance that both theoretical insight and practical implementation have when you’re developing AI solutions for real-world applications
Your work combines computational neuroscience with AI. How do insights from neuroscience influence the way you develop AI models?
In computational neuroscience, we study how the brain processes information by measuring its responses to various stimuli, much like how we probe AI models to understand their internal mechanisms. Early in my career, I developed visualization techniques to analyze word embeddings — breaking down words like “apple” into their constituent semantic elements, such as “fruit” and “technology.” Later on, this approach expanded to more complex AI models like transformers and large language models which helped reveal how they process and store knowledge.
These methods actually parallel techniques in neuroscience, such as using electrodes or fMRI to study brain activity. Probing an AI model’s internal representations allows us to understand its reasoning strategies and detect emergent properties, like concept neurons that activate for specific ideas (such as the Golden Gate Bridge feature Anthropic found when mapping Claude). This line of research is now widely adopted in the industry because it’s proven to enable both interpretability and practical interventions, removing biases from models. So neuroscience-inspired approaches essentially help us make AI more explainable, trustworthy, and efficient.
What inspired you to co-found Aizip? Can you share the journey from concept to company launch?
As a fundamental AI researcher, much of my work was theoretical, but I wanted to bridge the gap between research and real-world applications. I co-founded Aizip to bring cutting-edge AI innovations into practical use, particularly in resource-constrained environments. Instead of building large foundation models, we focused on developing the world’s smallest and most efficient AI models which would be optimized for edge devices.
The journey basically began with a key observation: While AI advancements were rapidly scaling up, real-world applications often required lightweight and highly efficient models. We then saw an opportunity to pioneer a new direction that balanced scientific rigor with practical deployment. By leveraging insights from self-supervised learning and compact model architectures, Aizip has been able to deliver AI solutions that operate efficiently at the edge and open up new possibilities for AI in embedded systems, IoT, and beyond.
Aizip specializes in small AI models for edge devices. What gap in the market did you see that led to this focus?
The AI industry has largely focused on scaling models up, but real-world applications often demand the opposite — high efficiency, low power consumption, and minimal latency. Many AI models today are too computationally expensive for deployment on small, embedded devices. We saw a gap in the market for AI solutions that could deliver strong performance while operating within extreme resource constraints.
We recognized that it is not only unnecessary for every AI application to run on massive models, but that it also wouldn’t be scalable to rely on models of that size for everything either. Instead, we focus on optimizing algorithms to achieve maximum efficiency while maintaining accuracy. By designing AI models tailored for edge applications — whether in smart sensors, wearables, or industrial automation — we enable AI to run in places where traditional models would be impractical. Our approach makes AI more accessible, scalable, and energy-efficient, unlocking new possibilities for AI-driven innovation beyond the cloud.
Aizip has been at the forefront of developing Small Language Models (SLMs). How do you see SLMs competing or complementing larger models like GPT-4?
SLMs and larger models like GPT-4 are not necessarily in direct competition because they serve different needs. Larger models are powerful in terms of generalization and deep reasoning but require substantial computational resources. SLMs are designed for efficiency and deployment on low-power edge devices. They complement large models by enabling AI capabilities in real-world applications where compute power, latency, and cost constraints matter — such as in IoT devices, wearables, and industrial automation. As AI adoption grows, we see a hybrid approach emerging, where large, cloud-based models handle complex queries while SLMs provide real-time, localized intelligence at the edge.
What are the biggest technical challenges in making AI models efficient enough for low-power edge devices?
One of the fundamental challenges is the lack of a complete theoretical understanding of how AI models work. Without a clear theoretical foundation, optimization efforts are often empirical, limiting efficiency gains. Additionally, human learning happens in diverse ways that current machine learning paradigms don’t fully capture, making it difficult to design models that mimic human efficiency.
From an engineering perspective, pushing AI to work within extreme constraints requires innovative solutions in model compression, quantization, and architecture design. Another challenge is creating AI models that can adapt to a variety of devices and environments while maintaining robustness. As AI increasingly interacts with the physical world through IoT and sensors, the need for natural and efficient interfaces — such as voice, gesture, and other non-traditional inputs — becomes critical. AI at the edge is about redefining how users interact with the digital world seamlessly.
Can you share some details about Aizip’s work with companies like Softbank?
We recently partnered with SoftBank on an aquaculture project that earned a CES Innovation Award — one we’re especially proud of. We developed an efficient, edge-based AI model for a fish counting application that can be used by aquaculture operators for fish farms. This solution addresses a critical challenge in fish farming which can ultimately create sustainability, food waste, and profitability issues. The industry has been slow to adopt AI as a solution due to unreliable power and connectivity at sea, making cloud-based AI solutions impractical.
To solve this, we developed a solution based on-device. We combined SoftBank’s computer graphics simulations for training data with our compact AI models and created a highly accurate system that runs on smartphones. In underwater field tests, it achieved a 95% recognition rate, dramatically improving fish counting accuracy. This allowed farmers to optimize storage conditions, determine whether fish should be transported live or frozen, and detect potential diseases or other health issues in the fish.
That breakthrough improves efficiency, lowers costs, and reduces reliance on manual labor. More broadly, it shows how AI can make a tangible impact on real-world problems.
Aizip has introduced an “AI Nanofactory” concept. Could you explain what that means and how it automates AI model development?
The AI Nanofactory is our internal AI Design Automation pipeline, inspired by Electronic Design Automation (EDA) in semiconductor manufacturing. Early development in any emerging technology field involves a lot of manual effort, so automation becomes key to accelerating progress and scaling solutions as the field matures.
Instead of simply using AI to accelerate other industries, we asked, can AI accelerate its own development? The AI Nanofactory automates every stage of AI model development from data processing to architecture design, model selection, training, quantization, deployment, and debugging. By leveraging AI to optimize itself, we’ve been able to reduce the development time for new models by an average factor of 10. In some cases, by over 1,000 times. This means a model that once took over a year to develop can now be created in just a few hours.
Another benefit is that this automation also ensures that AI solutions are economically viable for a wide range of applications, making real-world AI deployment more accessible and scalable.
How do you see the role of edge AI evolving in the next five years?
Edge AI promises to transform how we interact with technology, similar to how smartphones revolutionized internet access. Most AI applications today are cloud-based, but this is starting to shift as AI moves closer to the sensors and devices that interact with the physical world. This shift emphasizes a critical need for efficient, real-time processing at the edge.
In the next five years we expect edge AI to enable more natural human-computer interactions, such as voice and gesture recognition and other intuitive interfaces, which would remove reliance on traditional barriers like keyboards and touchscreens. AI is also expected to become more embedded in everyday environments like smart homes or industrial automation to enable real-time decision-making with minimal latency.
Another key trend will be the increasing autonomy of edge AI systems. AI models will become more self-optimizing and adaptive thanks to advancements in AI Nanofactory-style automation, so they will be able to reduce the need for human intervention in deployment and maintenance. That will open new opportunities across a number of industries like healthcare, automotive, and agriculture.
What are some upcoming AI-powered devices from Aizip that you’re most excited about?
We’re working to expand use cases for our models in new industries, and one we’re especially excited about is an AI Agent for the automotive sector. There’s growing momentum, particularly among Chinese automakers, to develop voice assistants powered by language models that feel more like ChatGPT inside the cabin. The challenge is that most current assistants still rely on the cloud, especially for natural, flexible dialogue. Only basic command-and-control tasks (like “turn on the AC” or “open the trunk”) typically run locally on the vehicle, and the rigid nature of those commands can become a distraction for drivers if they do not have them memorized with total accuracy.
We’ve developed a series of ultra-efficient, SLM-powered AI agents called Gizmo that are currently used in a number of applications for different industries, and we’re working to deploy them as in-cabin “co-pilots” for vehicles too. Gizmo is trained to understand intent in a more nuanced way, and when serving as a vehicle’s AI Agent, could execute commands through conversational, freeform language. For example, the agent could adjust the cabin’s temperature if a driver simply said, “I’m cold,” or respond to a prompt like, “I’m driving to Boston tomorrow, what should I wear?” by checking the weather and offering a suggestion.
Because they run locally and don’t depend on the cloud, these agents continue functioning in dead zones or areas with poor connectivity, like tunnels, mountains, or rural roads. They also enhance safety by giving drivers complete voice-based control without taking their attention off the road. And, on a separate and lighter note, I thought I’d also mention that we’re also currently in the process of putting an AI-powered karaoke model for vehicles and bluetooth speakers into production, which runs locally like the co-pilot. Basically, it takes any input audio and removes human voices from it, which allows you to create a karaoke version of any song in real-time. So aside from helping customers more safely manage controls in the car, we’re also looking for ways to make the experience more fun.
These kinds of solutions, the ones that make a meaningful difference in people’s everyday lives, are the ones we’re most proud of.
Aizip develops ultra-efficient AI solutions optimized for edge devices, offering compact models for vision, audio, time-series, language, and sensor fusion applications. Its products enable tasks like face and object recognition, keyword spotting, ECG/EEG analysis, and on-device chatbots, all powered by TinyML. Through its AI nanofactory platform, Aizipline, the company accelerates model development using foundation and generative models to push toward full AI design automation. Aizip’s Gizmo series of small language models (300M–2B parameters) supports a wide range of devices, bringing intelligent capabilities to the edge.
Thank you for the great interview, readers who wish to learn more should visit Aizip.