Reference library
Training Deep Nets with Sublinear Memory Cost
Paper summary
Reduces neural-training memory by storing selected activations and recomputing others during backpropagation. Graph analysis and checkpoint schedules expose a controlled tradeoff between memory and extra computation.
Relevance to the project
Relevant if external training is later used. Distinguish training activation checkpointing from full simulation checkpoints required for exact continuation.
Recomputation requires deterministic, correctly restored inputs and is not free.
Source record
- Reference ID
- activation-checkpointing
- Archived version
- 1604.06174v2
- Summary provenance
- reference-library/papers/activation-checkpointing/metadata.json · summary
- Review status
- initial orientation summary; not a critical full-paper review
- Reading scope
- Selected abstract/opening and available closing discussion/conclusion passages in the preserved text extract; broader project notes provide context. No reproduction of experiments, exhaustive methods review, or equation/figure audit.