Blog Post Ideas
Published on 18 Mar 2017
1. Why bias term in neural networks? Illustrate with a simple 2 class problem in two dimensions. 2. Loss functions - use a squared error loss such that it is hinged (set to 0) if \$$ty > 1 \$$. What kind of classifier it leads to. (Less robust to outliers). Demonstrate with a sample dataset. 3. Ho Kashyap algorithm 4. Spectral clustering and tip on number of isolated components 5. EM algorithm for multiple regressions (already have the ipython notebook) 6. Backpropagation - vectorized implementation 7. SVM - solving dual vs primal 8. Proof of chain rule in backprop 9. Influence of \$$\eta \$$, the learning rate on neural network learning 10. What is stochastic gradient descent? 11. sklearn custom estimator - also include my codes for perceptron, neural net & naive bayes 12. ReLU activation function - implement a neural network with ReLU 13. word2vec 14. MRF simple image denoising 15. Implement violat jones + adaboost 16. A new perspective on PCA - \$$min ||x - UU^Tx||^2 \$$ with the constraint \$$UU^T=I \$$ 17. Brown Clustering vs Word embeddings 18. SVM Dual Visualization - https://www.youtube.com/watch?v=ztCXSAUAe38