Research
The lab has always paired rigorous models with real systems. Today that method spans three subjects: quantum, AI, and networks.
Making quantum computation reliable.
As the hardware matures, the hard problem moves from building qubits to keeping computation and communication correct under noise and adversaries. The lab works on the foundations: quantum computation, quantum resilience, and the networking a quantum Internet will need.
A recent thread runs the other way. It applies quantum computing to classical networking, starting with quantum algorithms for network verification.
A theory of how models learn.
Why do large language models work? The lab's answer is Bayesian: in-context learning is inference on a latent manifold, and attention is the geometry that carries it out. The same tools drive the lab's work on causal inference, synthetic control, and AI in medicine and the sciences.
The systems and economics of the Internet.
The founding subject, still active. Fluid models of congestion control, overlay architectures against denial-of-service, peer-to-peer and video delivery, and the economics of who pays for the network: ISP settlement, network neutrality, zero-rating.