SecDVA: A Maliciously Secure Multi-Party Data Valuation Framework
Start here
An arXiv paper proposes SecDVA, which it claims is the first maliciously secure multi-party computation (MPC) data valuation framework, leaking nothing beyond the final result about the seller's data or the buyer's model and test data. It linearizes model updates with the neural tangent kernel and designs a (2+1)-party MPC protocol using function secret sharing; experiments show it outperforms baselines in the plaintext setting and can evaluate the data value of large models under MPC in reasonable time.
Generated by AI from reports · Updated 2 days ago
Attention and developments
Attention, hourly
Now 20, -2% in 24 hours·Peak 22 (Nov 13 18:00)
Related research
This event's paper
- SecDVA: A Maliciously Secure Multi-Party Framework for Data Value Assessment in Modern Machine Learning Data Markets
arXiv · 2026-11-11 · Deep reading
Progress timeline
- arXivSecDVA: A Maliciously Secure Multi-Party Data Valuation Framework for Modern Machine Learning Data Markets
The paper proposes SecDVA, which it claims is the first maliciously secure multi-party computation (MPC) data valuation framework, leaking nothing beyond the final result about the seller's data or the buyer's model and test data. It linearizes model updates with the neural tangent kernel, reduces overhead with a mini-batch approximation, and designs a (2+1)-party MPC protocol using function secret sharing. Experiments show that it outperforms baselines in the plaintext setting and can evaluate the data value of large models under MPC in a reasonable time.
Related discussion
Attention trend
New sources per day
- Nov 13: 1 new sources (1 outlets, 0 accounts)
- Nov 14: 0 new sources (0 outlets, 0 accounts)
- Nov 15: 0 new sources (0 outlets, 0 accounts)