Hello 👋

I first learned about this paper when Google Brain released the Tensorflow Recommenders library last month. I focused on it because Google, which operates a massive recommendation system like YouTube, was releasing recommendation system-related code. The overall content is more detailed in the Tensorflow Blog, so please read it. The goals of TFRS (TensorFlow Recommenders) are as follows: Build recommendation candidates quickly and flexibly Structure that freely uses Item, User, Context information Multi-task structure that learns various objectives simultaneously Learned models are efficiently served through TF Serving Actually, the code itself doesn’t have much diverse content, but what was most impressive was the Two Tower Model introduced as the basic model in the code. It’s about training User and Item completely independently and only predicting click/unclick with dot product at the final stage. The more I think about it, the better the structure seems. Although it’s unknown whether it will show tremendous performance since user tower and item tower can’t interact during training, the structure itself has no constraints on input features, so you can freely add possible features, and during inference, you can serve efficiently by having user embeddings and item embeddings and calculating similarity only with dot product, so compatibility with ANN (Approximate Nearest Neighbors) libraries also looks good. ...

Recently, while studying Recommender Systems, I thought I needed to study the field of Multi-armed bandit. I’ve summarized it based on A Survey of Online Experiment Design with the Stochastic Multi-Armed Bandit. Table of Contents 1. Concept 2. Differences between MAB and Existing Statistical Models 1. Concept The background of the term Multi-armed Bandit (hereinafter MAB) is gambling. What is the method for someone to obtain maximum profit through N slot machines with different profit distributions within a given time? If given the opportunity to pull N slot machines within limited time to obtain profit, there should first be a process of exploring which slot machine can earn more money for some time (this is called Exploration), and then there is a process of maximizing profit by pulling slot machines that one judges to be good (this is called Exploitation). ...

This post summarizes what attention is, focusing on several important papers and how it is used in NLP. Table of Contents Problems with Existing Encoder-Decoder Architecture Basic Idea Attention Score Functions What Do We Attend To? Multi-headed Attention Transformer Problems with Existing Encoder-Decoder Architecture The most important part of the Encoder-Decoder architecture is how to vectorize the input sequence. In NLP, input sequences often have dynamic structures, so problems arise when converting them to fixed-length vectors. For example, sentences like “Hello” and “The weather is nice today but the fine dust is severe, so make sure to wear a mask when you go out!” contain very different amounts of information, yet the encoder-decoder structure must convert them to vectors of the same length. Attention was first proposed to reduce information loss and solve problems more intuitively by reflecting which parts should be paid particular attention to in sequence data, as the word suggests. ...