Fundamental Limits of Neural Network Sparsification: Evidence from Catastrophic Interpretability Collapse

Documenti analoghi

• A Multi-Level Causal Intervention Framework for Mechanistic Interpretability in Variational Autoencoders
  di: Roy, Dip, et al.
  Pubblicazione: (2025)
• Posterior-Calibrated Causal Circuits in Variational Autoencoders: Why Image-Domain Interpretability Fails on Tabular Data
  di: Roy, Dip, et al.
  Pubblicazione: (2026)
• Detection Without Correction: A Robust Asymmetry in Activation-Based Hallucination Probing
  di: Roy, Dip, et al.
  Pubblicazione: (2026)
• ARDDQN: Attention Recurrent Double Deep Q-Network for UAV Coverage Path Planning and Data Harvesting
  di: Kumar, Praveen, et al.
  Pubblicazione: (2024)
• MemGuard-Alpha: Detecting and Filtering Memorization-Contaminated Signals in LLM-Based Financial Forecasting via Membership Inference and Cross-Model Disagreement
  di: Roy, Anisha, et al.
  Pubblicazione: (2026)