Research

Questions that connect generalization with real conditions.

The focus is on methods and evaluations that remain meaningful when data, environments, or operational constraints change.

01

Domain generalization

How can learned representations preserve useful structure across domains without access to the future target distribution?

Current questions

  • Evaluation under realistic distribution shift
  • Representation stability and failure analysis
  • Reproducible comparison protocols
Deep Learning notes ↗
02

Time series learning

How should models combine temporal structure, uncertainty, and changing data regimes?

Current questions

  • Reliable forecasting under drift
  • Temporal representation learning
  • Evaluation beyond aggregate error
Research notes ↗
03

Research infrastructure

How can experiment and deployment systems make results easier to reproduce, audit, and extend?

Representative work

  • Deterministic static publishing
  • Versioned artifacts and atomic releases
  • Clear operational verification
Engineering notes ↗