The Topological Trouble With Transformers
arxiv:2604.17121
LLMs: architectures, context, training · 2026-04-18
Complete cleaned Markdown · Source Markdown including bibliography
Read in parts
Contents
- The Topological Trouble With Transformers
- Abstract
- 1 Introduction
- 2 State tracking
- 3 Recurrent architectures
- 4 Architectural limitations and workarounds
- 5 Promising directions
- 5.1 Enhanced State-Space Models
- 5.2 Approximating state tracking in feedforward transformers
- 5.3 Coarse recurrence
- 5.4 Leveraging representational alignment
- 5.5 Efficient training of recurrence
- 6 Conclusions
- Acknowledgments and Disclosure of Funding