Continual learning streaming experience replay buffer with Vitter's reservoir sampling.
Discover more agent tools at GenPark and the GenPark MCP Catalog.
graph LR
A[Continuous Agent Trajectories] -->|P = C / N| B[Reservoir Sampling Buffer]
B -->|Uniform Sample Batch| C[Continual Replay Optimizer]
C --> D[Interleaved Experience Gradient Updates]
- Bounded
$O(C)$ capacity memory footprint. - Uniform probability guarantee across arbitrarily long streams.
- Zero external dependencies.