Distributed Monitoring of the R2 Statistic for Linear Regression

DReMo is a distributed algorithm for monitoring the coefficient of determination (R2) in linear regression models across nodes in a dynamic computing environment. The method, developed by NASA, is designed to be communication-efficient and scalable, with theoretical correctness guarantees. It uses both synthetic and real data to trigger model recomputation when R2 falls below a threshold.

Use Cases

• Monitoring model quality degradation in distributed linear regression based on the R2 statistic.
• Triggering model retraining in a communication-efficient fashion based on a collective threshold decision.
• Studying evolving feature-target relationships in large-scale, dynamic data environments.
• Anomaly detection in safety-critical applications where correctness guarantees are required.

Strengths

• Algorithm is described as highly communication-efficient and scalable.
• Method provides theoretical guarantees on correctness.
• Empirical validation uses both synthetic and real data.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

National Aeronautics and Space Administration

Collection Method

Likely contains results from algorithm development and empirical testing.

Time Range

null

Freshness

Last updated 2026-04-10 19:38:54.633303; freshness should be verified.

Geography

null