Saved in:
Bibliographic Details
Main Author: Berezkin, Igor
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2603.00812
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917302487744512
author Berezkin, Igor
author_facet Berezkin, Igor
contents Work introduces a hierarchical binary tree-based reduction that replaces standard self-attention. The core idea is to use a recursive Gated Linear Unit merge operation, achieving O(n) total merge operations O(log n) parallel depth O(n d^2) total work and O(n) space complexity. In these experiments, the model significantly outperforms standard Transformers in both convergence speed and accuracy on long-range structural dependencies, specifically where hierarchical inductive bias is critical.
format Preprint
id arxiv_https___arxiv_org_abs_2603_00812
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Wave-Attractor-Tree: A Hierarchical Binary Tree Reduction Architecture for Efficient Sequence Modeling
Berezkin, Igor
Machine Learning
68T07
Work introduces a hierarchical binary tree-based reduction that replaces standard self-attention. The core idea is to use a recursive Gated Linear Unit merge operation, achieving O(n) total merge operations O(log n) parallel depth O(n d^2) total work and O(n) space complexity. In these experiments, the model significantly outperforms standard Transformers in both convergence speed and accuracy on long-range structural dependencies, specifically where hierarchical inductive bias is critical.
title Wave-Attractor-Tree: A Hierarchical Binary Tree Reduction Architecture for Efficient Sequence Modeling
topic Machine Learning
68T07
url https://arxiv.org/abs/2603.00812