Recently, there has been a renewed interest in using linear RNNs for efficient sequence modeling.

First three authors contributed equally.

However, as we move up the visual hierarchy, the complexity of neuronal computations increases. Consequently, it becomes more challenging to model neuronal activity, requiring more complex models. In this study, we introduce a novel readout architecture inspired by the mechanism of visual attention.

This allows us to include clustering in end-to-end trainable pipelines, with efficient gradients.

The ability to perform causal reasoning is widely considered a core feature of intelligence.

Typically labels for these tasks come from human annotators.