From Shortcut to Induction Head: How Data Diversity Shapes Algorithm Selection in Transformers

Documenti analoghi

• Transformers as Measure-Theoretic Associative Memory: A Statistical Perspective and Minimax Optimality
  di: Kawata, Ryotaro, et al.
  Pubblicazione: (2026)
• State Space Models are Provably Comparable to Transformers in Dynamic Token Selection
  di: Nishikawa, Naoki, et al.
  Pubblicazione: (2024)
• Mixture of Experts Provably Detect and Learn the Latent Cluster Structure in Gradient-Based Learning
  di: Kawata, Ryotaro, et al.
  Pubblicazione: (2025)
• Learning sum of diverse features: computational hardness and efficient gradient-based training for ridge combinations
  di: Oko, Kazusato, et al.
  Pubblicazione: (2024)
• Pretrained transformer efficiently learns low-dimensional target functions in-context
  di: Oko, Kazusato, et al.
  Pubblicazione: (2024)