A Unifying View of Attention Sinks: From Mechanisms to Architectural Interventions

arXiv:2606.08105 · cs.LG · Submitted 2026-06-06 · Read on arXiv

cs.LG

Submitted: 2026-06-06

Updated: 2026-10-05

Code: https://github.com/KempnerInstitute/attention_sink

Terminology

• Transformer — 2× in this paper · explained in 2 episodes
• Trace — 1× in this paper · explained in 2 episodes

Sources

• LeJEPA: Provable and Scalable Self-Supervised Learning Without the Heuristics
• Memory Transformer
• Active-Dormant Attention Heads: Mechanistically Demystifying Extreme-Token Phenomena in LLMs
• Methods of improving LLM training stability
• Rethinking Attention: Polynomial Alternatives to Softmax in Transformers
• Augmenting Self-attention with Persistent Memory
• Replacing softmax with ReLU in Vision Transformers

Related papers

• Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
• AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
• Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
• Convergence issues in Relational Concept Analysis based on AOC-posets
• Beliefs Beyond Posteriors: The Predictive Cost of ELBO Binding
• Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks