Universal Quantum Transformer Achieves Exact Generalization on 5-Qubit NISQ Hardware
UQT paper demonstrates exact algebraic learning via quantum inductive bias on NISQ hardware.
The arXiv preprint 2606.00045 presents the Universal Quantum Transformer as a quantum-native architecture relying on parameterized geometric phase embedding and SU(2) wave-interference. The 5-qubit implementation learns Z_11 modular arithmetic and S_4 non-Abelian permutations with deterministic exactness. Authors report crystallization, defined as mathematically precise generalization beyond classical grokking. The framework claims logarithmic compression of representation dimension and elimination of quadratic self-attention scaling, with deployment shown on IBM Quantum NISQ devices per the abstract. Submission metadata lists author Alireza Talebpour with v1 posted 29 Apr 2026. Related quantum machine learning results on symmetry enforcement appear in arXiv:2206.08351, which examines variational circuits for group representations but does not address transformer attention. Primary measurements remain limited to the single 5-qubit substrate described.
AXIOM: Quantum phase embedding may enable exact discrete reasoning on current hardware where classical over-parameterization produces only stochastic convergence.
Sources (2)
- [1]Primary Source(https://arxiv.org/abs/2606.00045)
- [2]Related Source(https://arxiv.org/abs/2206.08351)