# Layer Flexible Adaptive Computational Time

## TL;DR

This paper introduces a layer flexible recurrent neural network with adaptive computation time that dynamically adjusts its depth during sequence processing, improving performance on sequence tasks.

## Contribution

It presents a novel recurrent neural network model that can vary its number of layers dynamically, unlike previous fixed-structure models.

## Key findings

- Achieved 7-12% performance improvement on financial and language modeling tasks.
- Demonstrated the model's ability to adapt the number of layers dynamically during processing.
- Showed that dynamic layering enhances sequence modeling capabilities.

## Abstract

Deep recurrent neural networks perform well on sequence data and are the model of choice. However, it is a daunting task to decide the structure of the networks, i.e. the number of layers, especially considering different computational needs of a sequence. We propose a layer flexible recurrent neural network with adaptive computation time, and expand it to a sequence to sequence model. Different from the adaptive computation time model, our model has a dynamic number of transmission states which vary by step and sequence. We evaluate the model on a financial data set and Wikipedia language modeling. Experimental results show the performance improvement of 7\% to 12\% and indicate the model's ability to dynamically change the number of layers along with the computational steps.