Papers/2609.30288
🧪 Test?View on arXiv

Manifold Projection and Iterative Autoencoder Refinement for Masked Language Modeling

Not provided

masked language modelingautoencoderstransformer alternativesefficiency
2609.30288
Builder Relevance
70%
1h ago

Abstract

This paper presents an alternative to attention mechanisms in masked language models using low-rank bottleneck autoencoders for context mixing.

Reality Card

Core Claim

The proposed architecture achieves a significant portion of attention's performance while using approximately 1.9 times fewer FLOPs compared to parameter-matched BERT baselines.

Method / Result

Achieves performance comparable to BERT and TinyBERT on rare token frequency tasks with a frequency-aware training schedule.

Limitations

The paper does not provide detailed information on the reproducibility of the results or the specific implementation details of the autoencoder modules.

Paper to code

Verified implementation resources so builders can test the paper’s claims instead of stopping at the abstract.

No verified implementation link has been attached yet. AIBuzzHub will keep this panel separate from unverified search results.
← Back to all papers