Pretrained transformers and RNNs linearly encode the Bayesian-updated belief geometry of the minimal classical, quantum, or post-quantum generator of their training data.
The process is defined by two parameters,α and x, with dependent quantitiesβ = (1−α)/2 and y = 1− 2x
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Neural networks leverage nominally quantum and post-quantum representations
Pretrained transformers and RNNs linearly encode the Bayesian-updated belief geometry of the minimal classical, quantum, or post-quantum generator of their training data.