Details

Autor(en) / Beteiligte

• Pintea, Silvia L.
• Liu, Yue
• van Gemert, Jan C.

Titel

Recurrent Knowledge Distillation

Ist Teil von

• 2018 25th IEEE International Conference on Image Processing (ICIP), 2018, p.3393-3397

Ort / Verlag

IEEE

Erscheinungsjahr

2018

Link zum Volltext

• Volltext

Quelle

IEEE Xplore Digital Library

Beschreibungen/Notizen

• Knowledge distillation compacts deep networks by letting a small student network learn from a large teacher network. The accuracy of knowledge distillation recently benefited from adding residual layers. We propose to reduce the size of the student network even further by recasting multiple residual layers in the teacher network into a single recurrent student layer. We propose three variants of adding recurrent connections into the student network, and show experimentally on CIFAR-10, Scenes and MiniPlaces, that we can reduce the number of parameters at little loss in accuracy.