.png){kind=icon}
.png){kind=icon}
Return
KD2M: A Unifying Framework for Feature Knowledge Distillation
.png){kind=icon}
PRE
AI
E
DOI：10.1007/978-3-032-03921-7_13.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Knowledge Distillation (KD) seeks to transfer the knowledge of a teacher, towards a student neural net. This process is often done by matching the networks' predictions (i.e., their output), but, recently several works have proposed to match the distributions of neural nets' activations (i.e., their features), a process known as distribution matching. In this paper, we propose an unifying framework, Knowledge Distillation through Distribution Matching ((KDM)-M-2), which formalizes this strategy. Our contributions are threefold. We i) provide an overview of distribution metrics used in distribution matching, ii) benchmark on computer vision datasets, and iii) derive new theoretical results for KD.
Keywords:
Knowledge Distillation
Optimal Transport
Computational Information Geometry
Deep Learning
Journal
G
IF:
0
Papers:
41
Citations:
0
Organization
No organization information available
Cited Papers