Published inTowards Data Science·Mar 8, 2022Cross-Entropy, Negative Log-Likelihood, and All That JazzTwo closely related mathematical formulations widely used in data science, and notes on their implementations in PyTorch — TL;DR Negative log-likelihood minimization is a proxy problem to the problem of maximum likelihood estimation. Cross-entropy and negative log-likelihood are closely related mathematical formulations. The essential part of computing the negative log-likelihood is to “sum up the correct log probabilities.”Machine Learning9 min readMachine Learning9 min read
Published inTowards Data Science·Jun 27, 2021A unified view of Graph Neural NetworksGraph attention, graph convolution, network propagation are all special cases of message passing in graph neural networks. — Message passing networks (MPN), graph attention networks (GAT), graph convolution networks (GCN), and even network propagation (NP) are closely related methods that fall into the category of graph neural networks (GNN). This post will provide a unified view of these methods, following mainly from chapter 5.3 in [1].Graph Neural Networks4 min readGraph Neural Networks4 min read
Published inTowards Data Science·Jun 20, 2021Network Learning — from Network Propagation to Graph ConvolutionTwo different views of network propagation, a popular method in computational biology, and its connection with graph convolution — You’ve probably heard about graph convolution as it is such a hot topic at the time. Although less well known, network propagation is a dominating method in computational biology for learning on networks. In this post, we’ll dive deep into the theory and intuition behind network propagation, and we’ll also…Network Science8 min readNetwork Science8 min read
Published inTowards Data Science·Jun 13, 2021Explainable Graph Neural NetworksA step forward in explainable AI, and why it is hard to adapt existing explanation methods to GNNs — TL;DR Explainability is a big topic in deep learning as it enables more reliable and trustable predictions. Existing explanation methods can’t be easily adapted to Graph Neural Networks due to the irregularity of graph structure. Quick peek into 5 groups of GNN explanation methods. Explainability increases reliabilityExplainable Ai4 min readExplainable Ai4 min read
Published inTowards Data Science·Jun 7, 2021Node2vec explained graphicallyHow second order random walk on graph works, explained via animations — Node2vec is an embedding method that transforms graphs (or networks) into numerical representations [1]. For example, given a social network where people (nodes) interact via relations (edges), node2vec generates numerical representation, i.e., a list of numbers, to represent each person. This representation preserves the structure of the original network in…Machine Learning8 min readMachine Learning8 min read
Published inTowards Data Science·May 31, 2021Run node2vec ultrafast with less memory using PecanPyAn easy to use implementation of a popular graph embedding method — Node2vec is a node embedding method that generates numerical representation (or embeddings) of nodes in a graph [1]. These embeddings are then used for various down stream tasks such as node classification and link prediction. …Machine Learning6 min readMachine Learning6 min read
Published inThe Citadel·May 12, 2021Towards an Unbiased Development of New Bioinformatics MethodsAdvancements in artificial intelligence and machine learning have led to many new methods in the areas of bioinformatics and computational biology. Common applications include gene classification [1], sample annotation [2], enzyme properties [3], and etc. The problem of optimistic bias Many papers that propose new computational methods conclude with a similar claim that “the proposed…Bioinformatics3 min readBioinformatics3 min read
