# Training a Subsampling Mechanism in Expectation

## TL;DR

This paper introduces a subsampling mechanism for sequences that can be trained via backpropagation, tested on a toy problem, and discusses its limitations.

## Contribution

It presents a novel subsampling method with a way to compute its expected output for training purposes.

## Key findings

- Effective on toy problems
- Allows training via backpropagation
- Highlights limitations of the approach

## Abstract

We describe a mechanism for subsampling sequences and show how to compute its expected output so that it can be trained with standard backpropagation. We test this approach on a simple toy problem and discuss its shortcomings.