The trade-off between time and space is a central issue in reverse mode automatic differentiation. The optimal checkpointing allows people to access computational states in a reversed order with only logarithmic overheads in time and space. Bennett's reversible computing scheme also allows efficient time and space trade-offs. This talk is about the pros and cons of two time and space trade-off schemes within the simple pebble game setup that even a gorilla can understand.
Youtube Video: https://youtu.be/Qf10_ngKMRw
References
- Reversible programming
https://github.com/GiggleLiu/NiLang.jl - Optimal checkpointing
https://www.tandfonline.com/doi/abs/10.1080/10556789208805505