Pith. sign in

REVIEW 1 cited by

Linear Transformer Topological Masking with Graph Random Features

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2410.03462 v2 pith:5ARQ32RT submitted 2024-10-04 cs.LG stat.ML

classification cs.LGstat.ML
keywords graphmaskingtopologicalattentiondatafeatureslinearmathcal
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
abstract

When training transformers on graph-structured data, incorporating information about the underlying topology is crucial for good performance. Topological masking, a type of relative position encoding, achieves this by upweighting or downweighting attention depending on the relationship between the query and keys in a graph. In this paper, we propose to parameterise topological masks as a learnable function of a weighted adjacency matrix -- a novel, flexible approach which incorporates a strong structural inductive bias. By approximating this mask with graph random features (for which we prove the first known concentration bounds), we show how this can be made fully compatible with linear attention, preserving $\mathcal{O}(N)$ time and space complexity with respect to the number of input tokens. The fastest previous alternative was $\mathcal{O}(N \log N)$ and only suitable for specific graphs. Our efficient masking algorithms provide strong performance gains for tasks on image and point cloud data, including with $>30$k nodes.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Learning the RoPEs: Better 2D and 3D Position Encodings with STRING

    cs.LG 2025-02 conditional novelty 6.0 of 10

    STRING parameterizes translation-invariant position encodings as exponentials of commuting skew-symmetric generators, proving they are exactly RoPE in a learnable orthogonal basis, and shows practical gains in 2D/3D v...

Pith tools