Home

Hello, and welcome to my website! I will be hosting all of my blogs here. I will initially start with blog posts concerning Learning-to-Rank, which is the application of machine learning to Information Retrieval/Search, and hope to write about a variety of topics such as Information Retrieval, Mathematics, etc...


Recent posts
Recent Posts Widget

Popular posts from this blog

Learning-to-Rank Workshop/Seminar #1: Summary

Image
Learning-to-Rank Workshop/Seminar #1: Summary Last month, a Learning-to-Rank (LTR) seminar/workshop was held at RONDHUIT Co, Ltd. , at which I gave an overview of the implemented neural network algorithms in the Learning-to-Rank for Lucene ( LTR4L ) project. More details about the event, including the slides presented by the speakers, can be found here . The neural network algorithms presented at the workshop were:  - PRank (Perceptron Ranking)  - OAP-BPM  - RankNet  - FRankNet (Factored RankNet, not RankNet with a fidelity loss function.)  - LambdaRank  - SortNet  - ListNet In addition to the neural network algorithms, I also gave a brief overview of the following:  - Loss functions  - LTR evaluation metrics  - Gradient Descent/ Stochastic Gradient Descent  - Back Propagation Although each of the algorithms have been summarized in the README of LTR4L, as we discussed the algorithms in more detail at the workshop, I would like to elaborate a bit more in this post.
Read more

Introduction to Boosting

Image
Introduction to Boosting Boosting is a type of machine learning algorithm in which a strong learner (a classifier with stron predictions) is made by adding together many weak learners (classifiers with weak predictions). The idea of strong learners and weak learners are fundamental to understanding boosting, and so I will attempt to very simply summarize them below. A weak learner is classifier which predicts with slightly higher accuracy than just randomly predicting. Generally, weak learners have a very simple structure. For example, for simplicity let us consider two categories: male and female. Let us attempt to find a classifier which can predict a person's biological sex (not gender, though correlations may or may not exist), based on the following information: Features of a person: (age, height, social security number, the number of times a dress or skirt was worn in the past five years) Obviously, none of the information above will directly tell you the biological se
Read more