How transferable are features in deep neural networks?
Authors: Jason Yosinski, Jeff Clune, Yoshua Bengio, Hod Lipson
NeurIPS 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this paper we experimentally quantify the generality versus specificity of neurons in each layer of a deep convolutional neural network and report a few surprising results. |
| Researcher Affiliation | Academia | Jason Yosinski,1 Jeff Clune,2 Yoshua Bengio,3 and Hod Lipson4 1 Dept. Computer Science, Cornell University 2 Dept. Computer Science, University of Wyoming 3 Dept. Computer Science & Operations Research, University of Montreal 4 Dept. Mechanical & Aerospace Engineering, Cornell University |
| Pseudocode | No | The paper does not contain any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | Further details of the training setup (learning rates, etc.) are given in the supplementary material, and code and parameter files to reproduce these experiments are available at http://yosinski.com/transfer. |
| Open Datasets | Yes | The Image Net dataset, as released in the Large Scale Visual Recognition Challenge 2012 (ILSVRC2012) (Deng et al., 2009) contains 1,281,167 labeled training images and 50,000 test images, with each image labeled with one of 1000 classes. |
| Dataset Splits | No | Figure 2: The results from this paper s main experiment. Top: Each marker in the figure represents the average accuracy over the validation set for a trained network. The paper mentions the use of a "validation set" but does not provide specific details on its size, proportion, or how it was derived from the 645,000 examples per task (A or B). |
| Hardware Specification | No | Each point is computationally expensive to obtain (9.5 days on a GPU) The paper mentions using "a GPU" but does not specify the model or any other specific hardware details like CPU, memory, or cloud instance types. |
| Software Dependencies | No | We use the reference implementation provided by Caffe (Jia et al., 2014) so that our results will be comparable, extensible, and useful to a large number of researchers. The paper mentions Caffe but does not specify its version or any other software with version numbers. |
| Experiment Setup | No | Further details of the training setup (learning rates, etc.) are given in the supplementary material, and code and parameter files to reproduce these experiments are available at http://yosinski.com/transfer. The paper states that detailed training setup, including hyperparameters like learning rates, are in supplementary material, not the main text. |