Researchers Map Cryptographic Functions to Transformer Networks

Researchers have explored the cryptographic capabilities of transformer networks, investigating whether these models can implement specific cryptographic functions. The study maps cryptographic constructions like Keccak functions, Merkle--Damgard constructions, and Merkle Trees to transformer architectures, deriving scaling laws for circuit width and depth. This work establishes a methodology for assessing transformer computational capacity and provides constructive upper bounds on what transformers of a given size can compute, contributing to principled capability evaluations of AI systems. AI

IMPACT Establishes theoretical limits on transformer computational capacity, aiding in capability evaluations.

RANK_REASON The cluster contains an academic paper detailing research into the capabilities of transformer networks. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• Keccak functions
• Merkle--Damgard constructions
• Merkle Trees
• Merrill et al.
• Transformer Networks

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →