Omid Saremi

Researcher@Apple·美国·OpenReview

研究方向

Multimodal Self Supervised Learning · Generalization in Transformers and state space model · Reasoning in transformers · Training dynamics in Neural networks

7.0

To Infinity and Beyond: Tool-Use Unlocks Length Generalization in State Space Models

ICLR 2026Oral

二作

7.6

When can transformers reason with abstract symbols?

ICLR 2024Poster

二作

7.0

How JEPA Avoids Noisy Features: The Implicit Bias of Deep Linear Self Distillation Networks

合作者 (20)

Omid Saremi

To Infinity and Beyond: Tool-Use Unlocks Length Generalization in State Space Models

When can transformers reason with abstract symbols?

How JEPA Avoids Noisy Features: The Implicit Bias of Deep Linear Self Distillation Networks

What Algorithms can Transformers Learn? A Study in Length Generalization

LiDAR: Sensing Linear Probing Performance in Joint Embedding SSL Architectures

Vanishing Gradients in Reinforcement Finetuning of Language Models

How Far Can Transformers Reason? The Globality Barrier and Inductive Scratchpad

Adaptivity and Modularity for Efficient Generalization Over Task Complexity