Temporal superposition and feature geometry of RNNs under memory demands

ICLROral2026

Authors: Pratyaksh Sharma, Alexandra Maria Proca, Lucas Prieto, Pedro A. M. Mediano
Affiliation: G-Research
Venue: ICLR 2026
Track: Oral

TL;DR

We study the feature geometry of RNNs under memory demands and characterize their representational strategies using a novel framework of temporal superposition.

Opening excerpt from the authors’ abstract. source

Read the paper

OpenReview

Topics

memory

← All ICLR 2026 Oral papers · Browse the whole archive