Image Classification Formulations

Insights and Analysis - Using Various Deep Learning Architectures on Image Classification Datasets

The Jupyter notebooks contain necessary theory and mathematical formulations

For more on multiclass image classification formulations, check out this repo.

---

Techniques on MNIST

• Eigen Spectrum of Various Subsets
• Extrapolation to the Population Space
• Probabilistic Concepts
  • Maximum Likelihood Estimation
  • Maximum a Posteriori Estimation
  • Bayesian Pairwise Majority Voting
  • Simple Perpendicular Bisector Majority Voting
• Nearest Neighbor based Tasks
  • Classification
  • Outlier Detection
  • Regression

For Performance using 10 One-vs-Rest classifiers and ¹⁰C₂ Binary Classifiers on MNIST, go here.

---

Face Classification on IIIT-CFW, IMFDB, Yale Face Database

• Study of the impact of various feature sets: PCA, k-PCA, LDA, k-LDA, VGGNet, Resnet
• Number of components for reasonable reconstruction
• Classification Formulations: MLP, SVM, D-Tree, Logistic Regression
• Confusion Matrices and t-SNE Visualization
• Task of classifying Cartoon versus Real World Imagery

---