Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Parallel Recursive LSTM

About

Transformers have become the dominant architecture for sequence modeling by using self-attention to enable expressive and highly parallel processing. However, the resulting quadratic time and memory costs limit efficiency in long-context settings. Recurrent models such as LSTMs provide explicit nonlinear state updates and strong state-tracking capabilities, yet their strictly sequential computation limits parallelism. We introduce the Parallel Recursive LSTM (PR-LSTM), a hierarchical recurrent architecture that replaces left-to-right recurrence with recursive nonlinear state composition over a balanced computation tree. Tokens are first mapped independently to latent states, which are then recursively merged by a learned gated composition block. This structure uses the reduction pattern underlying parallel scans as a fixed execution schedule, rather than assuming an associative recurrence. As a result, PR-LSTM retains nonlinear gated state representations while reducing recurrent parallel depth from linear to logarithmic. Empirically, PR-LSTM achieves strong sequence-length generalization on formal-language benchmarks, solving more tasks than standard RNN, LSTM, and Transformer baselines, while avoiding the quadratic scaling of attention. These results suggest that recurrent computation can be reorganized hierarchically to expose parallelism without restricting the transition dynamics to linear or associative forms.

Tristan Gaudreault, Yongyi Mao• 2026

Related benchmarks

TaskDatasetResultRank
Bucket SortFormal-language benchmark lengths 41-500 (test)
Accuracy99.4
6
Missing DuplicateFormal-language benchmark lengths 41-500 (test)
Accuracy100
6
Binary MultiplicationFormal-language benchmark lengths 41-500 (test)
Accuracy (%)52.9
6
Compute SqrtFormal-language benchmark lengths 41-500 (test)
Accuracy (%)56.8
6
Duplicate StringFormal-language benchmark lengths 41-500 (test)
Accuracy54.7
6
Odds FirstFormal-language benchmark lengths 41-500 (test)
Accuracy55
6
Binary AdditionFormal-language benchmark lengths 41-500 (test)
Accuracy51.8
6
Cycle NavigationFormal-language benchmark lengths 41-500 (test)
Accuracy100
6
Even PairsFormal-language benchmark lengths 41-500 (test)
Accuracy (%)100
6
Modular ArithmeticFormal-language benchmark lengths 41-500 (test)
Accuracy36.4
6
Showing 10 of 15 rows

Other info

Follow for update