New associative neural networks boost sparse pattern storage capacity

Researchers have developed new associative neural network models that significantly increase storage capacity for sparse patterns. These models combine higher-order interaction terms with sparse pattern mechanisms, achieving storage scales of order N^n / (log N)^n for fixed interaction orders. When the interaction order grows logarithmically with the number of neurons, the storage scale becomes super-polynomial, and for block-structured architectures, it can reach c^n. AI

IMPACT This research could lead to more efficient memory systems for AI models, enabling them to handle larger and more complex datasets.

RANK_REASON Academic paper detailing a new theoretical model. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• Amari models
• Gripon--Berrou architecture
• Matthias Löwe
• Willshaw models

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →