Pierre Ablin

Researcher@Apple·法国·OpenReview

研究方向

optimization · invertible neural networks · sparsity · data selection · large language models · large scale learning · small models

5.5

Completed Hyperparameter Transfer across Modules, Width, Depth, Batch and Duration

ICLR 2026Poster

二作

5.0

Multi-View Causal Discovery without Non-Gaussianity: Identifiability and Algorithms

ICLR 2026Rejected

三作

4.5

The data-quality illusion: Rethinking Classifier-based Quality Filtering for LLM Pretraining

ICLR 2026Rejected

通讯

3.5

Balanced Low-Rank Adaptation: Removing Invariance for Fast and Stable Fine-Tuning

合作者 (20)

Angelos Katharopoulos

2 篇

Jason Ramapuram

2 篇

Pierre Ablin

Completed Hyperparameter Transfer across Modules, Width, Depth, Batch and Duration

Multi-View Causal Discovery without Non-Gaussianity: Identifiability and Algorithms

The data-quality illusion: Rethinking Classifier-based Quality Filtering for LLM Pretraining

Balanced Low-Rank Adaptation: Removing Invariance for Fast and Stable Fine-Tuning

Scaling Laws for Optimal Data Mixtures

The AdEMAMix Optimizer: Better, Faster, Older

Task-Adaptive Pretrained Language Models via Clustered-Importance Sampling

Scaling Laws for Forgetting during Finetuning with Pretraining Data Injection

Need a Small Specialized Language Model? Plan Early!

Sparse Repellency for Shielded Generation in Text-to-Image Diffusion Models

Theory, Analysis, and Best Practices for Sigmoid Self-Attention

Soup-of-Experts: Pretraining Specialist Models via Parameters Averaging

Dynamic Gradient Alignment for Online Data Mixing

Shielded Diffusion: Generating Novel and Diverse Images using Sparse Repellency