I came to find out that I can no longer listen to QQ music since I moved here to Singapore. That is definitely something I need to take care of since we all need some of the medoly to boost our work. After a quick seach, I have found this perticular approach that can help with that. What’s more, new Apple music subscriber would have a long period of 3 months free service.

Neighbour Components Analysis (NCA) learns a linear transfromation for the input space such that in the transformed space, KNN performs well. Let $Q = A^TA$, we have $$ \begin{aligned}d(x,y) &= (x-y)^T Q (x-y) \\ &= (x-y)^TA^TA(x-y) \\&= (Ax-Ay)^T(Ax-Ay)\end{aligned} $$ NCA proposes to use a softmax value to represent the probability that for a given sample $i$, sample $j$ is selected as its neighbour. $$ p_{ij} = \frac{\exp(-||Ax_i - Ax_j||^2)}{\sum_{k \neq i} \exp(-||Ax_i - Ax_k||^2)} $$ which is the pair similarity over all posible pairs, looking for the nearnest sample in the new space.

After few hours of template searching, I have found myself this particular template that I love. So, it’s official now! I have a new blog running. As the first post here, I would like to record the process of creating and deploying this blog so people who wants to do the same can have some reference. Hope this post can be helpful! Prerequisites This blog is built with Hugo and one of the popular themes LoveIt.