PulseAugur
EN
LIVE 04:42:23

HASTE framework enables training-free compression of CNNs

By PulseAugur Editorial · [2 sources] ·

Researchers have developed HASTE, a novel framework designed to compress large pre-trained convolutional neural networks (CNNs) without requiring additional training or data access. This plug-and-play module utilizes locality-sensitive hashing to dynamically merge redundant channels during inference, thereby reducing computational costs. Experiments on datasets like CIFAR-10 and ImageNet show significant reductions in FLOPs, such as a 46.2% decrease in ResNet34 with a minimal accuracy drop. AI

IMPACT Enables more efficient deployment of large CNNs on resource-constrained devices without retraining.

RANK_REASON The cluster describes a new research paper detailing a novel framework for CNN compression.

Read on arXiv cs.CV →

AI-generated summary · Google Gemini · from 2 sources. How we write summaries →

HASTE framework enables training-free compression of CNNs

COVERAGE [2]

  1. arXiv cs.CV TIER_1 English(EN) · Lukas Meiner, Jens Mehnert, Alexandru Paul Condurache ·

    HASTE: A Framework for Training-Free, Dynamic, and Steerable Compression of Pre-Trained Convolutional Neural Networks

    arXiv:2606.30516v1 Announce Type: new Abstract: Deploying large convolutional neural networks (CNNs) on resource-constrained devices is challenging due to their high computational cost. While dynamic execution methods are promising, existing approaches for CNNs typically require …

  2. arXiv cs.CV TIER_1 English(EN) · Alexandru Paul Condurache ·

    HASTE: A Framework for Training-Free, Dynamic, and Steerable Compression of Pre-Trained Convolutional Neural Networks

    Deploying large convolutional neural networks (CNNs) on resource-constrained devices is challenging due to their high computational cost. While dynamic execution methods are promising, existing approaches for CNNs typically require specialized training or fine-tuning, limiting th…

RELATED ENTITIES

RELATED TOPICS