Interpolation Learning With Minimum Description Length
arXiv:2302.07263v14 citationsh-index: 73
Originality Synthesis-oriented
AI Analysis
This work addresses theoretical foundations for MDL in machine learning, offering incremental insights into its statistical properties.
The paper tackled the problem of understanding the generalization behavior of Minimum Description Length (MDL) learning rules, proving that they exhibit tempered overfitting and providing finite sample guarantees with random label noise.
We prove that the Minimum Description Length learning rule exhibits tempered overfitting. We obtain tempered agnostic finite sample learning guarantees and characterize the asymptotic behavior in the presence of random label noise.