From Condensation to Rank Collapse: A Two-Stage Analysis of Transformer Training Dynamics

Documenti analoghi

• On Multi-Stage Loss Dynamics in Neural Networks: Mechanisms of Plateau and Descent Stages
  di: Chen, Zheng-An, et al.
  Pubblicazione: (2024)
• Disentangling Feature Structure: A Mathematically Provable Two-Stage Training Dynamics in Transformers
  di: Gong, Zixuan, et al.
  Pubblicazione: (2025)
• A Two-Stage Data Selection Framework for Data-Efficient Model Training on Edge Devices
  di: Gong, Chen, et al.
  Pubblicazione: (2025)
• Rank, Head-Channel Non-Identifiability, and Symmetry Breaking: A Precise Analysis of Representational Collapse in Transformers
  di: Cirrincione, Giansalvo
  Pubblicazione: (2026)
• Dynamic Graph Condensation
  di: Chen, Dong, et al.
  Pubblicazione: (2025)