aiqbench analysis
This research proposes an attention-free architecture for masked language models that utilizes low-rank bottleneck autoencoders to mix contextual information more efficiently than standard Transformers. By introducing an iterative refinement procedure involving manifold projection, the model achieves performance comparable to BERT baselines while reducing computational cost by approximately 1.9x FLOPs. The approach offers a significant pathway toward optimizing large-scale language model training through novel architectural constraints rather than just scaling parameters.
Manifold Projection and Iterative Autoencoder Refinement for Masked Language Modeling
arXiv:2609.30288v1 Announce Type: new Abstract: In Transformer-based masked language models, attention is the primary mechanism for context mixing, but there are other ways to mix data across tokens. Recent attention-free mixers replace attention with fixed or hypernetwork-generated MLPs, alternating their dynamic, content-dependent weighting for computational simplicity. We build an alternative that gets the same property from a low-rank bottleneck autoencoder. We replace attention with a stack of autoencoder-based mixing modules, one operating over local neighborhoods, one over the full sequence, and one across attention heads, each compressing and reconstructing its input through a bottleneck, and its width is a hyperparameter rather than a training effect. In masked positions, we introduce an iterative refinement procedure that has two distinct steps. A pulling step that pulls an embedding representation toward a weighted average of its neighbors, and a correcting step that projects the result back to the learned manifold via an autoencoder. Our architecture achieves a significant portion of attention's performance at about $1.9 \times$ fewer FLOPs when pretrained on C4 and evaluated with parameter-matched BERT baselines. Our model equals parameter-matched BERT and TinyBERT baselines on the rarest-token frequency bucket using a frequency-aware training schedule that samples rare tokens more than uniformly for the masking tasks.
Abstract:In Transformer-based masked language models, attention is the primary mechanism for context mixing, but there are other ways to mix data across tokens. Recent attention-free mixers replace attention with fixed or hypernetwork-generated MLPs, alternating their dynamic, content-dependent weighting for computational simplicity. We build an alternative that gets the same property from a low-rank bottleneck autoencoder. We replace attention with a stack of autoencoder-based mixing modules, one operating over local neighborhoods, one over the full sequence, and one across attention heads, each compressing and reconstructing its input through a bottleneck, and its width is a hyperparameter rather than a training effect. In masked positions, we introduce an iterative refinement procedure that has two distinct steps. A pulling step that pulls an embedding representation toward a weighted average of its neighbors, and a correcting step that projects the result back to the learned manifold via an autoencoder. Our architecture achieves a significant portion of attention's performance at about $1.9 \times$ fewer FLOPs when pretrained on C4 and evaluated with parameter-matched BERT baselines. Our model equals parameter-matched BERT and TinyBERT baselines on the rarest-token frequency bucket using a frequency-aware training schedule that samples rare tokens more than uniformly for the masking tasks.
Computation and Language (cs.CL); Machine Learning (cs.LG)
arXiv:2609.30288 [cs.CL]
(or arXiv:2609.30288v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2609.30288 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 295 words
Related coverage
Learning Through Energy Refinement and Manifold Projection: A Cooperative EBM-AE Framework
arXiv · Sep 15, 2026
J++ Lens: Jacobian Filtering Enables More Faithful Workspace Lenses
LessWrong · Oct 8, 2026
Multi-Turn Reflective Masking Elicits Reasoning in Mask Diffusion Models
arXiv · Jun 17, 2026
PAIR: Bridging Perception and Action in Vision-Language-Action Models
arXiv · Oct 8, 2026