Unsupervised Learning on Neural Network Outputs: With Application in Zero-Shot Learning
Authors: Yao Lu
IJCAI 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our zero-shot learning method achieves the state-of-the-art results on the Image Net of over 20000 classes. In Table 4, we show the results of different methods on Image Net 2011fall. Our method performs better when using PCA or ICA for the visual features than random features. And our method, especially with PCA or ICA features, achieves superior results on this zero-shot learning task. |
| Researcher Affiliation | Academia | Yao Lu Aalto University University of Helsinki Helsinki Institute for Information Technology |
| Pseudocode | No | The paper provides mathematical equations for the algorithms used (e.g., Equation 4 for the SGD-based ICA algorithm) but does not include structured pseudocode or an algorithm block explicitly labeled as such. |
| Open Source Code | Yes | The code for reproducing the experiments is in 3. https://github.com/yaolubrain/ULNNO |
| Open Datasets | Yes | We used all the images in the Image Net ILSVRC2012 training set to compute the ICA matrix... and Following the zero-shot learning experimental settings of De Vi SE and con SE, we used a CNN trained on Image Net ILSVRC2012 (1000 seen classes), and test our method to classify images in Image Net 2011fall (20842 unseen classes... |
| Dataset Splits | No | The paper mentions using 'Image Net ILSVRC2012 training set' and 'Image Net 2011fall' for testing but does not provide specific training/validation/test dataset splits (e.g., percentages, sample counts, or predefined split citations). |
| Hardware Specification | No | The paper mentions using CNN models like GoogLeNet and AlexNet and software like Caffe and Theano, but does not provide specific hardware details (e.g., CPU/GPU models, memory amounts) used for running the experiments. |
| Software Dependencies | No | The paper states that 'The computation of CNN outputs was done with Caffe [Jia et al., 2014]. The ICA algorithm was ran with Theano [Bergstra et al., 2010].' but does not specify the version numbers for these software components. |
| Experiment Setup | Yes | The learning rate was set to 0.005 and was halved every 10 epochs. The computation of CNN outputs was done with Caffe [Jia et al., 2014]. The ICA algorithm was ran with Theano [Bergstra et al., 2010]. V was initialized as a random orthogonal matrix. We used k = 100, 500, 900 in our experiments. The ICA algorithm was ran with Theano [Bergstra et al., 2010]. ... with mini-batch size 500. |