Spotlight Talks
summary
Proposes a fair exchange contract that pairs agents one-for-one and proves the induced game is supermodular, so pure Nash equilibria exist, form a lattice, and the maximal equilibrium is computable in O(n^2) and truthfully implementable under simple enforcement; extends to graph-restricted and discrete collections with efficiently computable Pareto-optimal equilibria. Separately, operationalizes explanations with ExpProof/FairProof using ZKPs, adapts LIME to ZKP-friendly variants (uniform sampling, no kernel), implements in Gnark and reports proof-generation times down to about a minute, showing practical feasibility.
AI-written summary · about
chapters (20)
- 0:00Fair data exchange research
- 2:10No-money data sharing models
- 3:15Fair exchange contract model
- 6:33Exchange contract equilibrium issues
- 7:14Supermodularity and Nash properties
- 8:23Maximal equilibrium algorithm
- 10:50Truthful reporting in contracts
- 13:42Graph and discrete data extensions
- 15:10Research directions and open problems
- 18:53Q&A on data costs
- 20:18Trust verification in AI systems
- 21:03Limitations of trustworthy AI tools
- 23:10Why explanations fail
- 26:21Zero-knowledge proofs for AI
- 28:05ExpProof protocol guarantees
- 29:57LIME with zero-knowledge proofs
- 33:14ZKP-friendly LIME sampling
- 35:52Experimental results and trade-offs
- 36:54Future scalability challenges
- 37:14Q&A on proof and verification
related talks
29mAccountable AI with ZKPs: Certifying Fairness and Explanations under model Confidentiality
26mPractical, Efficient ZKPs, Economic and Social Processes
01h 15mTo interact or not to interact: Cryptographic proof systems and the Fiat-Shamir heuristic
01h 08mTutorial on Proof Systems
General Game-Theoretic Multiagent Reinforcement Learning
28mZero Knowledge Proofs and Their Future Applications
01h 09mProofs - Practice 1
36mPanel Discussion