Autoregressive Text Generation beyond Feedback Loops
EMNLP/IJCNLP (1), pp. 3398-3404, 2019.
Autoregressive state transitions, where predictions are conditioned on past predictions, are the predominant choice for both deterministic and stochastic sequential models. However, autoregressive feedback exposes the evolution of the hidden state trajectory to potential biases from well-known train-test discrepancies. In this paper, we...More
PPT (Upload PPT)