Compression & intelligence
Compression is a lens on intelligence
The interesting part is not making data smaller. It is discovering structure that survives beyond the examples.
The notebook of Xiaohan Yuan
Intelligence,
in the making.
Exploring how compact ideas become capable agents—and how those agents meet the constraints of the physical world.
FROM BITS TO MINDFIG. 001
A few starting points
Selected essays
Agent systems
Build the laboratory, not another agent
For domain-specific engineering, the highest-leverage work may be the environment around the agent—not another reasoning loop.
AI × RFIC
At radio frequencies, layout is still design
A netlist is not a complete physical specification. Geometry, coupling, and return paths are part of the design problem.
A working conviction
Not a finished theory
Useful intelligence finds structure.
Good engineering makes it answer
to reality.
I am interested in the space between a model’s capability and an engineering result: the representations, tools, environments, and checks that make the difference.
Why Bits to MindFrom ideas to instruments
Where the questions
become concrete.
Research directions at the intersection of intelligent systems and physical engineering.
Explore the work010203
An RFIC design workbench for agents
Domain tools, native design state, and independent verification—not another general-purpose reasoning loop.
More useful physical queries
Electromagnetic modeling and multifidelity evaluation, organized around the decisions that matter.
Microwave neural computation
Learning in the physical signal path—and understanding the conditions under which it is useful.