In-context learning decomposes into concept-coordinate regression plus off-subspace leakage, with recoverable task information concentrating in a 68-73 dimensional task-aligned subspace of the residual stream that restores 78.8% of the accuracy gap in Llama-3-8B experiments.
Springer, 2 edition
3 Pith papers cite this work. Polarity classification is still indexing.
3
Pith papers citing it
citation-role summary
background 1
citation-polarity summary
years
2026 3verdicts
UNVERDICTED 3roles
background 1polarities
background 1representative citing papers
MC² corrects low-budget Monte Carlo solutions for elliptic PDEs with a single-pass neural network to match the accuracy of 1000× more Monte Carlo samples while outperforming classical and learned baselines.
Multi-zone unified statistical model for DART spike prediction combined with closed-form impact-aware INC/DEC trading strategy improves risk-return profile over unit-size trades in NYISO.
citing papers explorer
-
In-Context Learning Operates as Concept Subspace Learning
In-context learning decomposes into concept-coordinate regression plus off-subspace leakage, with recoverable task information concentrating in a 68-73 dimensional task-aligned subspace of the residual stream that restores 78.8% of the accuracy gap in Llama-3-8B experiments.
-
MC$^2$: Monte Carlo Correction for Fast Elliptic PDE Solving
MC² corrects low-budget Monte Carlo solutions for elliptic PDEs with a single-pass neural network to match the accuracy of 1000× more Monte Carlo samples while outperforming classical and learned baselines.
-
Trading Electrons: Predicting DART Spread Spikes in ISO Electricity Markets
Multi-zone unified statistical model for DART spike prediction combined with closed-form impact-aware INC/DEC trading strategy improves risk-return profile over unit-size trades in NYISO.