How to design data representation techniques with applications to fast retrieval tasks — Hashing is one of the most fundamental operations in data management. It allows fast retrieval of data items using a small amount of memory. Hashing is also a fundamental algorithmic operation with rigorously understood theoretical properties. …

A simple reduction that allows using libraries for nearest neighbor search for the efficient detection of vectors with large inner product — Motivation Nearest neighbor search is one of the most fundamental problems in data science. Applications range from user segmentation to near-duplicate detection. Not surprisingly, most programming languages used for data science applications offer libraries with highly scalable algorithms for nearest neighbor search.

Advanced neural network architectures work by learning feature interactions Bilinear pooling originated in the computer vision community as a method for fine-grained visual recognition. Or in less fancy language, a method that looks for specific details when recognizing and classifying visual objects. At a high level, the approach works as…

Graph kernels methods that are easy to implement and yield models that can be efficiently trained — In the third post of these series I will discuss graph kernels. Graph kernels are an approach that generalizes kernel machines such as support-vector machines to working with graphs. In a nutshell, the idea of graph kernels is to compare two graphs by a function that has as input the…

Scalable data summarization techniques for one of the most fundamental problems in data analysis Computing the variance of a data feature is probably one of the most fundamental operations in data analysis. …

Node embeddings. In the previous blog post in these series I presented basic graph statistics that should be collected before starting to develop machine learning models for your graph data. In the second post I will present the first proper machine learning approach. Namely, learning to represent discrete objects like…

Collecting basic statistics — In a series of posts, I will provide an overview of several machine learning approaches to learning from graph data. Starting with basic statistics that are used to describe graphs, I will go deeper into the subject by discussing node embeddings, graph kernels, graph signal processing, and eventually graph neural…

And how to train embeddings from similarity functions — Representing discrete objects by continuous vectors, the so-called embeddings, has been at the heart of many successful machine learning solutions. The superiority comes from the fact that, unlike the original discrete objects, the embedding vectors offer a compact representation that captures the similarity between the original objects. In this article…

It is all about finding the right space At a high level, many machine learning algorithms can be thought of as consisting of the following steps. We are given an input dataset where examples are described by n-dimensional feature vectors. In supervised problems we are also given a target label. …