AMSB is a themed puzzle set where modern ML building blocks are recast as precise algorithmic tasks inside a living “city of signals.” Streets are graphs that hum in the spectral domain; laboratories distill high-dimensional statistics into tiny matrix operations; studios rearrange textures on rings with one-step transports; control rooms run autoregressive blocks without ever simulating a network step; vaults preserve noisy time-negatives for closed-form reconstruction; and a spiking district fires once—only when algebra says it must. Each problem hides a familiar idea behind clean I/O and demands that you recognize structure, precompute the right invariants, and answer at scale with careful numerics.
What you’ll navigate
- Spectral filtering on graphs via Chebyshev recurrences and sparse–dense blocks.
- Low-rank covariance interactions disguised in linear projections.
- Distribution comparison from minimal sufficient statistics (Fréchet/Bures-style metrics).
- Large-scale convolutions—direct, FFT, or circulant—chosen by structure.
- Autoregressive, causal blocks solved with single scans and range prefixes.
- Diffusion-style forward chains turned into exact linear functionals with fixed RNG.
- One-step mean transports on a ring through circular normalization.
- Time-to-First-Spike reasoning using event times and first-hit formulas.
AMSB expects individuals or teams that see through the veneer: spot the transform, batch the heavy work, clamp the edge cases, and let offline preparation turn vast query loads into instant answers. The contest will last for seven days from Oct 1st. And we consider offering the answer after 10 or more members requiring it.
Contest is now VISIBLE on SZUEA OJ Site. Please register and have fun!
Note: This contest is not the second assessment of the Shenzhen University Electronics Association.
注意：此比赛并非深圳大学电子协会第二次考核范围。当然，我们欢迎老登和有能力的小登参与，题目难度较大，请酌情考虑。（老登请勿参与新生招新二面测试，如想小试牛刀，请完成AMSB的题目，感谢您的理解）
