CONNECTED RESEARCH

Knowledge map

Explore research through connected companies, technologies and questions. Every relationship links to its evidence.

55 / 55 articles with reviewed reader guides · 202 reviewed relationships

Per-step policy-learning gradients

Per-step policy-learning gradientsPer-step policy-learning graTRACETRACE

Select a connected node to continue exploring. Open the evidence below to distinguish reporting, analysis and hypotheses.

Reported evidence

TRACE → uses → Per-step policy-learning gradients

The article describes TRACE as a proposed temporal gradient-inversion attack using per-step policy-learning gradients. The relationship concerns the method’s input, not evidence that an attacker has access to those inputs in a deployed system.

Read the evidence
The paper proposes Temporal Reconstruction Attack on Consecutive Encodings, abbreviated TRACE. As described in the abstract, it exploits temporal structure to autoregressively reconstruct private observation–action sequences from per-step policy-learning gradients.[S2]

[S2] arXiv

Read the full research

Connected research