🤝 Work with Us

If interested in working with us, have a read of this page, to see if you

• 🌈 buy our research belief,
• 📚 are interested in the research topics,
• 🌟 consider yourself a good fit for the expected qualities,
• 🎯 and love what outcome you can expect in such an experience.

We welcome talented students and researchers at all levels (undergraduate, master, PhD, post-doc) and from all backgrounds (computer science, neuroscience, physics, mathematics, etc.).

🌈 Research Belief

We work across intersections of some of the most exciting topics related to intelligence (biological neurons & brains, new artificial neural networks, dissected transformers, next-generation GPUs, neuromorphic devices & systems, emerging memories, and etc.). Most importantly, we look at things in the most first-principled way (from a most fundamental level). Thus explains why the above seemingly diverse topics form a coherent whole in our research:

We aim to discover and understand the fundamental elements of intelligence, so as to build the next-generation learning systems (for the future 10+ years 🚀).

📚 Research Topics

Below lists the topics that we are currently most excited about and actively working on (being updated regularly):

• ⚛️ The “atom” of learning. We have been working on a learning rule that internally we called “3Q”, because the learning rule can be described by three simple equations of basic operations like addition and multiplication. It has the best biological plausibility we’ve ever seen (thus potentially the best candidate for designing next-generation AI acceleration hardware). However, we have not yet quite figured out its learning capacity. What is the learning dynamic of 3Q? What are the “learning guarantees” of 3Q? How would 3Q (or can it) implement key computation motifs like gating, attention, etc.? (The work is not published yet, so apologies for not more details here.)
• 🌀 The “essence” of self-attention. Self-attention has demonstrated to be the most expressive architecture that essentially opens up the era of “large models”. We want to understand the “essence” of self-attention. For example, we see it is essentially an “associative memory” (see our ICLR paper), and it is related to the simplest problem in machine learning, linear regression (see here, particularly, Mufeng spotted this). So what is the “essence” of self-attention? Would understanding this unlock better self-attention, with simpler operators that is more biological plausible or/and hardware friendly? Or would understanding this inspire alternative mechanisms of self-attention?
• 🤔 Better self-attention from better biological associative memory? Since we know self-attention can be seen as an “associative memory” from our ICLR paper, this nature paper talks about a new type of associative memory, would it inform better alternative of self-attention?

More to come… But you should get a taste of our research here.

🌟 Expected Qualities

🌌 “Belief-wise”: first principle believer. You should be a strong believer in first principles: meaning you have a strong curiosity in asking “but fundamentally why and how? 🤓”. You enjoy such a process of deep thinking and exploring to understand the essence of things (in science, technology, even also in life). You enjoy this because you believe that’s the only way to make sense of things and create unique value! 💡 Linked to this, you should be an honest person: honest with people, with science and technology, and most importantly, honest with yourself.

🧩 “Intelligence-wise”: you should either be able to “handle complexity” or can “make things simple”. I see broadly two kinds of talents, and you can be either one of them (i.e., don’t have to be both):

• Able to “handle complexity”. You’re just good at working with symbols, and you’re good at math or physics (demonstrated by your transcripts). In this case, you don’t even have to be good at coding for now. But of course, you should consider coding as a tool and have a passion for learning it (at least some basics).
• Able to “make things simple”. You’re eager to understand things in the deepest way, which often means the simplest way of explaining. You can demonstrate this by explaining a new concept/idea to a general audience. You enjoy explaining things and can make the explanation so simple and intuitive that it becomes a pleasant experience for both you and your audience. You’re normally good at coding in this case (or should be able to get good at coding soon) due to the particular way your brain approaches things.

🌟 I myself am unfortunately not the first type but fortunate to be the second type.

🌍 We will have intensive intentional collaboration, so you should be confident in your English or ready to improve in a short time.

🌱 You expect the time to be a pleasant intellectual exercise; you are curious, eager to explore, and learn.

🧑‍🏫 Advising Style

My advising style is highly engaged — deeply shaped by my own PhD advisor, who led by example. I write code, derive equations, and prepare slides to explain ideas when needed.

That is, I don’t just give high-level guidance — I work with you side by side through the thinking and building. I see research as a collaborative journey, where both of us are actively engaged, intellectually invested, and driven by curiosity.

This works best when you are also highly motivated: proactive in exploring ideas, open to feedback, and excited about learning and creating together. If that sounds like you, we’ll likely have a great time working together.

🎯 Expected Outcomes

Just as this research agenda carried me to fulfilling roles in academia, industrial research, and entrepreneurship, it can launch you toward success along whichever of these paths you embark on. (I will add more details here later along each line.)

🎉 Congrats in making it to the end! You are now ready to reach out to us, including a brief self-introduction and why you’re interested in working with us.