Tony Wang

AI researchers are fascinated by Large Language Models. But Large Tabular Models? Nope, they don't get anything like as much research love. Our PhD student Xiangjian Jiang wants to change that. He has just been awarded a prestigious 2025 Global Google PhD Fellowship and hopes the accolade will help him highlight an under-explored area of AI research that has applications in realms from finance to health. https://lnkd.in/gryW5wVc

59

If you are trained in AB testing and linear models, how do you move into deep learning? Linear models are such an important foundation in data science, and especially for AB testing. They power variance reduction, heterogeneous effects, and control for confounders in cases when randomization broke. Inference from a linear model is largely a series of matrix multiplications. Deep learning is too. It's a composition of linear functions stacked with nonlinear activations, and can be thought of as a way to form nonlinear basis functions for regression. These days it is more important than ever to truly understand what your model is doing, so I've written an educational piece on how to pivot a mental model around regression to a mental model around neural networks. Check it out here https://lnkd.in/g24Wr_kw

73

1 Comment

In a major step toward practical quantum computers, Princeton engineers have built a superconducting qubit that lasts three times longer than today’s best versions. “The real challenge, the thing that stops us from having useful quantum computers today, is that you build a qubit and the information just doesn’t last very long,” said PMI's Andrew Houck, leader of a federally funded national quantum research center, Princeton’s dean of engineering and co-principal investigator on the paper. “This is the next big jump forward.” In a Nov. 5 article in the journal Nature, the Princeton team reported their new qubit lasts for over 1 millisecond. This is three times longer than the best ever reported in a lab setting, and nearly 15 times longer than the industry standard for large-scale processors. The researchers built a fully functioning quantum chip based on this qubit to validate its performance, clearing one of the key obstacles to efficient error correction and scalability for industrial systems. Read more: https://lnkd.in/e3EGtPeE

4

🤖 The continuum robotics challenge just got solved. Researchers have cracked a unified mathematical model for tendon-actuated concentric tube robots—a breakthrough that combines the best of two worlds. For years, engineers faced a trade-off: tendon-driven robots lacked flexibility in their degrees of freedom, while concentric tube designs suffered from instability. This new Cosserat rod-based framework handles multiple tubes with multiple tendons actuating each one, validated through experiments achieving <4% tip prediction error. The model is so robust it accurately predicts behavior across existing robots in the field. • Why it matters: Accurate modeling means better shape estimation and control—unlocking precision surgical applications, delicate manipulation tasks, and next-gen minimally invasive procedures. Follow for more breakthroughs in robotics and AI innovation! 🚀 #Robotics #ContinuumRobotics #Biomimetics #ControlSystems #RoboticSurgery #Innovation Source: https://lnkd.in/gQtDFW7Z

2

Quantum magic 🪄 Sandia researchers and partners at Quantinuum and University of California, Davis validated a new algorithm for error-resistant quantum computation. In a paper published by American Physical Society, the team details a new and more efficient method for preparing “magic states,” enabling future quantum computers to run programs even if one component fails, a concept called fault-tolerance. “A magic state is a particular superposition state into which a qubit can be initialized,” explained Sandia researcher Robin Blume-Kohout. “It can be difficult to prepare a magic state, but once accomplished, it becomes very useful in calculations. It’s like a $100 gift card that you can redeem for cool stuff whenever you want. Specifically, the quantum computer’s programmer can ‘redeem’ it for a special logic operation, the T gate, that is necessary from time to time in any useful quantum program.” Read more: https://bit.ly/4rEaGWf

97

3 Comments

I returned to the #UniversityOfMinnesota on Thursday to talk about Spectral Properties of Graph Neural Networks (GNNs). The message here is that graph convolutions have trouble handling high frequencies, defined as signals that change quickly over the graph. This observation has implications on the stability of GNN outputs to graph deformations and the transferability of GNNs across scales. Stability and transferability properties of GNNs are analogous to stability and transferability properties of CNNs. This is because GNNs and CNNs are algebraically equivalent objects. (See references below). A recording of this talk delivered at the #SimonsFoundation: https://lnkd.in/e4Rxw-gX My graph neural networks course at #Penn: https://gnn.seas.upenn.edu Stability properties of GNNs: https://lnkd.in/en-UH8jj Transferability properties of GNNs: algehttps://lnkd.in/efrnDJFM Algebraic neural networks: manifold https://lnkd.in/eDpxp-tj Manifold convolutions: https://lnkd.in/eraqxeDP Learning by transference: https://lnkd.in/etaPVQVS Learnable perception action communication loops: chttps://lnkd.in/ejZ5qKkC Covariance Neural Networks: https://lnkd.in/ee2CBweA For those of you that have never lived in #Minnesota, it is always a treat to return to the best state in the union.

143

1 Comment

Imagine quantum key distribution (QKD) systems so small they fit on a single chip. That's the future. Researchers are already dedicating years to this, creating compact, efficient systems. Soon, this technology will be as ubiquitous as your GPU or CPU. While the exact applications and costs are still evolving, one thing is clear: quantum tech is on the verge of a major breakthrough. #QuantumTech #QKD #Cybersecurity #Innovation

4

A new feature from Nature titled "Quantum computers will finally be useful: what’s behind the revolution" highlights research from Princeton ECE professors Andrew Houck and Nathalie de Leon. The piece describes how their work improving the performance of qubits has contributed to dramatically shortening the projected timeline for practical quantum computers. "Just a few years ago, many researchers in quantum computing thought it would take several decades to develop machines that could solve complex tasks, such as predicting how chemicals react or cracking encrypted text," according to the article. "But now, there is growing hope that such machines could arrive in the next ten years." Check it out 👉 https://lnkd.in/eC5cSDs4

19

1 Comment