Jaegle et al.: iterative cross-attention into a latent perception bottleneck

What changed

Used asymmetric cross-attention to distill high-dimensional multimodal inputs into a fixed-size latent array processed through iterative attention, separating input size from latent network depth.

Scientific scope

Used asymmetric cross-attention to distill high-dimensional multimodal inputs into a fixed-size latent array processed through iterative attention, separating input size from latent network depth. Transformer attention and latent bottlenecks are inherited. Linear input-size scaling assumes fixed latent size/depth and does not guarantee arbitrary-input information retention. Biological analogy is motivation only; acquired contribution is non-biological ML architecture.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Transformer attention and latent bottlenecks are inherited. Linear input-size scaling assumes fixed latent size/depth and does not guarantee arbitrary-input information retention. Biological analogy is motivation only; acquired contribution is non-biological ML architecture.

Contributors

• Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.

• Perceiver: General Perception with Iterative Attention

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction