AI research uses multi-armed bandits to prune neural networks

Researchers have developed a novel method for pruning feature maps in convolutional neural networks (CNNs) to reduce computational costs and storage requirements. This approach utilizes multi-armed bandit algorithms, specifically UCB1 and Thompson Sampling, to identify and remove redundant feature maps while minimizing accuracy loss. The study demonstrates that these bandit-based methods significantly outperform traditional greedy and magnitude-based pruning techniques, achieving accuracy comparable to unpruned models on various datasets including MNIST, CIFAR-10, and SVHN. AI

IMPACT This research could lead to more efficient AI models by reducing computational load and storage needs, making them more accessible for deployment.

RANK_REASON The item is an academic paper detailing a new method for optimizing neural networks. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• CIFAR-10
• CIFAR-100
• convolutional neural network
• CUB-200-2011
• Friedman test
• LeNet-5
• MNIST database
• Multi-armed bandits for adjudicating documents in pooling-based evaluation of information retrieval systems
• Nemenyi test
• Oxford Flowers 102
• The Street View House Numbers Dataset
• Thompson sampling
• Upper Confidence Bound 1

• paper
• other

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