Towards a Theoretical Understanding of Synthetic Data in LLM Post-Training: A Reverse-Bottleneck Perspectiveby Zeyu Gan,…
Extending Contextual Self-Modulation: Meta-Learning Across Modalities, Task Dimensionalities, and Data Regimesby Roussel Desmond Nzoyem, David…
Reducing Variance in Meta-Learning via Laplace Approximation for Regression Tasksby Alfredo Reichlin, Gustaf Tegnér, Miguel…
Mitigating Copy Bias in In-Context Learning through Neuron Pruningby Ameen Ali, Lior Wolf, Ivan TitovFirst…
PhyMPGN: Physics-encoded Message Passing Graph Network for spatiotemporal PDE systemsby Bocheng Zeng, Qi Wang, Mengtao…
On Expressive Power of Looped Transformers: Theoretical Analysis and Enhancement via Timestep Encodingby Kevin Xu,…
Equivariant score-based generative models provably learn distributions with symmetries efficientlyby Ziyu Chen, Markos A. Katsoulakis,…
Investigating the Synergistic Effects of Dropout and Residual Connections on Language Model Trainingby Qingyang Li,…
On the Generalization and Causal Explanation in Self-Supervised Learningby Wenwen Qiang, Zeen Song, Ziyin Gu,…
Fine-Grained Gradient Restriction: A Simple Approach for Mitigating Catastrophic Forgettingby Bo Liu, Mao Ye, Peter…