In Defense of Fully Connected Layers in Visual Representation Transfer

85 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Maintaining Discrimination and Fairness in Class Incremental Learning (2019, background): “Actually, the FC layer plays an important role in the visual representation transfer [36].”
RIFLE: Backpropagation in Depth for Deep Transfer Learning through Re-Initializing the Fully-connected LayEr (2020, background): “Although the fully-connected (FC) layer would be ﬁne tuned with superior performance to classify the training datasets (Zhang et al., 2017), the features learned for classiﬁcations might still be largely based on the source datasets of transfer learning rather than the target one (Yosinski et al.,…”
How Case-Based Reasoning Explains Neural Networks: A Theoretical Analysis of XAI Using Post-Hoc Explanation-by-Example from a Survey of ANN-CBR Twin-Systems (2019, background): “…is meant to behave similarly to the black-box system but is simpler for explanatory purposes (so, the CBR in ANN-CBR twin is one type of proxy model). systems (notably, with multiple fully connected dense layers using image transfer learning [88]) in comparative tests (see [4] and also Fig 2).”