Researchers have developed a new categorical framework called Learning in Infinitesimal Non-Compositional Sketches (LINCS) to address non-compositionality in machine learning. This framework defines non-compositionality as a failure in universal factorization problems and proposes using tangent lifts to perturb models and preserve compositionality. The paper introduces Tangent Learning Sketches and the INC endofunctor, formulating machine learning as a search for a coalgebraic fixed point. An experimental evaluation of LINCS is currently underway for deep learning, large language models, and reinforcement learning. AI
IMPACT Introduces a novel theoretical framework for addressing non-compositionality in machine learning models.
RANK_REASON The cluster contains an academic paper detailing a new theoretical framework for machine learning.
- Aczel--Mendler theorem
- deep learning
- INC endofunctor
- large-language models
- Learning in Infinitesimal Non-Compositional Sketches
- reinforcement learning
- Tangent Learning Sketches
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
