PPaperPicks

Thomas Hofmann

26 papers at tracked venues · 21 at CORE A* · active 20242025

Venues

Frequent coauthors

Papers

  1. Causal Estimation of Tokenisation Bias
  2. Emergence of Globally Attracting Fixed Points in Deep Neural Networks With Nonlinear Activations
  3. Generalized Interpolating Discrete Diffusion
  4. LIME: Localized Image Editing via Attention Regularization in Diffusion Models
  5. LoRACLR: Contrastive Adaptation for Customization of Diffusion Models
  6. On the Expressiveness and Length Generalization of Selective State Space Models on Regular Languages
  7. Scalable Non-Equivariant 3D Molecule Generation via Rotational Alignment
  8. Structured Sparse Transition Matrices to Enable State Tracking in State-Space Models
  9. The Directionality of Optimization Trajectories in Neural Networks
  10. The Importance of Being Lazy: Scaling Limits of Continual Learning
  11. The Non-Linear Representation Dilemma: Is Causal Abstraction Enough for Mechanistic Interpretability?
  12. UIP2P: Unsupervised Instruction-Based Image Editing via Edit Reversibility Constraint
  13. A Language Model's Guide Through Latent Space
  14. An Art-Centric Perspective on AI-Based Content Moderation of Nudity
  15. Causal Estimation of Memorisation Profiles
  16. Exposed or Erased: Algorithmic Censorship of Nudity in Art
  17. How Good is a Single Basin?
  18. Local and Global Decoding in Text Generation
  19. Navigating Scaling Laws: Compute Optimality in Adaptive Model Training
  20. On the Effect of (Near) Duplicate Subwords in Language Modelling
  21. Recurrent Distance Filtering for Graph Representation Learning
  22. Simplifying Transformer Blocks
  23. Super Consistency of Neural Network Landscapes and Learning Rate Transfer
  24. Towards Meta-Pruning via Optimal Transport
  25. Transformer Fusion with Optimal Transport
  26. Understanding and Minimising Outlier Features in Transformer Training