RNNs trained on short parity sequences can suddenly generalize to arbitrary length by merging hidden states that agree on future outputs, forming a finite automaton.
ISSN: 2640-3498
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
-
Algorithm Development in Neural Networks: Insights from the Streaming Parity Task
RNNs trained on short parity sequences can suddenly generalize to arbitrary length by merging hidden states that agree on future outputs, forming a finite automaton.