首页/数据集/Impact of training approaches on LKD loss and student model accuracy.

Impact of training approaches on LKD loss and student model accuracy.

Figshare2025-06-11 更新2026-04-28 收录

数据链接：

https://figshare.com/articles/dataset/Impact_of_training_approaches_on_LKD_loss_and_student_model_accuracy_/29296468 数据链接链接失效反馈

官方服务：

资源简介：

Impact of training approaches on LKD loss and student model accuracy.

应用场景：

创建时间：

2025-06-11

相关数据集

Distill-and-Select (DnS)

知识蒸馏优化

视频检索效率

在这项工作中，我们提出了一个称为提取和选择 (DnS) 的知识蒸馏框架，该框架从性能良好的细粒度教师网络开始学习: a) 在不同的检索性能和计算效率权衡下的学生网络; b) 在测试时快速将样本引导到适当的学生以保持高检索性能和高计算效率的选择器网络。我们用不同的体系结构训练几个学生，并得出性能和效率的不同权衡，即速度和存储要求，包括使用二进制表示存储/索引视频的细粒度学生

OpenXLab50

Accuracy with different kernel functions in the Cora dataset.

图分类

知识蒸馏优化

Note that KL divergence-based kernel provides the best accuracy, and the student ‘None’ without the embedding distillation shows a poor performance.

Figshare2021-08-13 更新20

Carbon footprints and performance for different KD approaches.

知识蒸馏优化

模型碳足迹评估

Here, for image recognition, we use ResNet18 (teacher) and MobileNetV2 (student) models. For object detection, we use VGG16 (teacher) and MobileNetV2 (student). We report average results after running

Figshare2023-05-15 更新20

Carbon footprints of Teacher (ResNet18) and Student (MobileNetV2) model before KD (top two rows), traditional KD [26] and Ours approach on three different datasets.

知识蒸馏优化

AI碳排放评估

Carbon footprints of Teacher (ResNet18) and Student (MobileNetV2) model before KD (top two rows), traditional KD [26] and Ours approach on three different datasets.

NIAID Data Ecosystem40

Impact of parameter sampling strategies on KD from ResNet-34 to ResNet-18.

知识蒸馏优化

模型压缩

Impact of parameter sampling strategies on KD from ResNet-34 to ResNet-18.

NIAID Data Ecosystem20