# LogitBoost autoregressive networks

## TL;DR

This paper demonstrates that training separate LogitBoost estimators for each dimension of multivariate binary data achieves competitive performance, enabling parallel training and simplified complexity control compared to neural network-based autoregressive models.

## Contribution

It introduces a novel approach of using independent LogitBoost estimators for each conditional distribution in autoregressive models, simplifying training and complexity management.

## Key findings

- Parallel training over data dimensions is feasible and effective.
- LogitBoost-based conditionals achieve competitive benchmark results.
- Different complexity control methods are experimentally compared.

## Abstract

Multivariate binary distributions can be decomposed into products of univariate conditional distributions. Recently popular approaches have modeled these conditionals through neural networks with sophisticated weight-sharing structures. It is shown that state-of-the-art performance on several standard benchmark datasets can actually be achieved by training separate probability estimators for each dimension. In that case, model training can be trivially parallelized over data dimensions. On the other hand, complexity control has to be performed for each learned conditional distribution. Three possible methods are considered and experimentally compared. The estimator that is employed for each conditional is LogitBoost. Similarities and differences between the proposed approach and autoregressive models based on neural networks are discussed in detail.