Výsledky vyhledávání

Report

Emergence of Abstractions: Concept Encoding and Decoding Mechanism for In-Context Learning in Transformers

Autor: Han, Seungwook, Song, Jinyeop, Gore, Jeff, Agrawal, Pulkit

Humans distill complex experiences into fundamental abstractions that enable rapid learning and adaptation. Similarly, autoregressive transformers exhibit adaptive learning through in-context learning (ICL), which begs the question of how. In this pa

Externí odkaz: http://arxiv.org/abs/2412.12276

Zobrazit plný text záznamu

Report

A Resource Model For Neural Scaling Law

Autor: Song, Jinyeop, Liu, Ziming, Tegmark, Max, Gore, Jeff

Neural scaling laws characterize how model performance improves as the model size scales up. Inspired by empirical observations, we introduce a resource model of neural scaling. A task is usually composite hence can be decomposed into many subtasks,

Externí odkaz: http://arxiv.org/abs/2402.05164

Zobrazit plný text záznamu

Report

Dynamic clustering of passive colloids in dense suspensions of motile bacteria

Autor: Gokhale, Shreyas, Li, Junang, Solon, Alexandre, Gore, Jeff, Fakhri, Nikta

Mixtures of active and passive particles are predicted to exhibit a variety of nonequilibrium phases. Here we report a dynamic clustering phase in mixtures of colloids and motile bacteria. We show that colloidal clustering results from a balance betw

Externí odkaz: http://arxiv.org/abs/2110.02294

Zobrazit plný text záznamu

Report

Slow expanders invade by forming dented fronts in microbial colonies

Autor: Lee, Hyunseok, Gore, Jeff, Korolev, Kirill S.

Most organisms grow in space, whether they are viruses spreading within a host tissue or invasive species colonizing a new continent. Evolution typically selects for higher expansion rates during spatial growth, but it has been suggested that slower

Externí odkaz: http://arxiv.org/abs/2108.04008

Zobrazit plný text záznamu