paper implementation tracker #818
Description
over the years, a bunch of papers have been published at MLO and around which might help DISCO one way or the other. here goes a list of the ones we spoke about. it is still up for debate which we actually need the most.
- 🌪️ communication efficiency
- 🔒 privacy-preserving
- secure aggregation for privacy-preserving ML
- somewhat done, only honest-but-curious server, need to be validated
- secure byzantine-robust ML
- secure aggregation for privacy-preserving ML
- 🛠️ fault-tolerant and dynamic over time
- buffered async aggregation
- done, need to document it in code
- async SGD
- buffered async aggregation
- 🥷 robust to malicious actors and data poisoning
- byzantine robust optimisation
- partially done via norm clipping
- byzantine robust learning on heterogeneous data
- byzantine robust optimisation
- 🍎 🍌 interpretable in imperfectly interoperable data distributions
- 🪞 personalizable