Saved in:
Bibliographic Details
Main Authors: Zerihun, Liyu, Plashchinsky, Alexandr
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2602.00535
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917238431285248
author Zerihun, Liyu
Plashchinsky, Alexandr
author_facet Zerihun, Liyu
Plashchinsky, Alexandr
contents Long sequence neural memory remains a challenging problem. RNNs and their variants suffer from vanishing gradients, and Transformers suffer from quadratic scaling. Furthermore, compressing long sequences into a finite fixed representation remains an intractable problem due to the difficult optimization landscape. Invertible Memory Flow Networks (IMFN) make long sequence compression tractable through factorization: instead of learning end-to-end compression, we decompose the problem into pairwise merges using a binary tree of "sweeper" modules. Rather than learning to compress long sequences, each sweeper learns a much simpler 2-to-1 compression task, achieving O(log N) depth with sublinear error accumulation in sequence length. For online inference, we distilled into a constant-cost recurrent student achieving O(1) sequential steps. Empirical results validate IMFN on long MNIST sequences and UCF-101 videos, demonstrating compression of high-dimensional data over long sequences.
format Preprint
id arxiv_https___arxiv_org_abs_2602_00535
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Invertible Memory Flow Networks
Zerihun, Liyu
Plashchinsky, Alexandr
Machine Learning
Long sequence neural memory remains a challenging problem. RNNs and their variants suffer from vanishing gradients, and Transformers suffer from quadratic scaling. Furthermore, compressing long sequences into a finite fixed representation remains an intractable problem due to the difficult optimization landscape. Invertible Memory Flow Networks (IMFN) make long sequence compression tractable through factorization: instead of learning end-to-end compression, we decompose the problem into pairwise merges using a binary tree of "sweeper" modules. Rather than learning to compress long sequences, each sweeper learns a much simpler 2-to-1 compression task, achieving O(log N) depth with sublinear error accumulation in sequence length. For online inference, we distilled into a constant-cost recurrent student achieving O(1) sequential steps. Empirical results validate IMFN on long MNIST sequences and UCF-101 videos, demonstrating compression of high-dimensional data over long sequences.
title Invertible Memory Flow Networks
topic Machine Learning
url https://arxiv.org/abs/2602.00535