Contents
- Introduction
- Hard Code Word
- Results
- Locality-constrained Linear
- Results
- Grid Search
- Sequential Hierarchy Classifier
- Manually assigned clusters
- Results
- Clusters from K-means
- Results
- Other Dataset Evaluation
- Birds
- Butterflies
- Other Experiments
- Results
- Scene Datasets
- Code
- Git Logs
- References
|
Different Classifiers
SVM Modifications
We experimented with some different classifiers as well as modifications of the SVM classifier
with different kernels. We found the SVM to give the best results when using the
histogram intersection kernel. Using a radial basis kernel, we retrieved extremely poor results that were near the level of chance.
k-Nearest Neighbor
We also built a simple k-nearest neighbor classifier. This performed extremely fast and gave us decent results with a mean accuracy of about 50% for using k=5 nearest neighbors. However, as expected, that 50% does not compare to our results using SVM.
Neural Networks
Finally, we used the machine learning toolbox of Matlab and tried classifying using a neural network. However, we could not get any meaningful results from this due to the problem of having too many features compared to the number of training examples.
|