Physics-inspired KSSE model achieves 88.93% ImageNet accuracy with reduced parameters

Researchers have developed a novel physics-inspired energy-based model called Kohn-Sham Spectral Embedding (KSSE) for image classification. KSSE replaces traditional dense Convolutional Neural Networks (CNNs) with a sparse-graph spectral embedding evaluated at a specific temperature. This approach leverages concepts from random-bond Ising models and random-matrix theory to optimize graph topology and analyze fractal learning landscapes. In experiments on ImageNet-1000, KSSE achieved a Top-1 accuracy of 88.93% with significantly fewer parameters than comparable models like Swin-L and ViT-H/14. AI

IMPACT This novel approach could lead to more parameter-efficient image classification models, potentially reducing computational costs and enabling deployment on resource-constrained devices.

RANK_REASON The cluster contains a research paper detailing a new model and its performance on a benchmark dataset. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• EfficientNet-B4
• ImageNet-1000
• Kohn-Sham Spectral Embedding
• Swin-L
• ViT-H/14

• paper
• model release

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