A blog containing tutorials, notes, and insights on topics in math, statistics, machine learning, computational biology, and pedagogy. To view posts grouped by topic, click here.
If you’re a practitioner of machine learning, then there is little doubt you have seen or used an algorithm that falls into the general category of kernel methods. The premier example of such methods is the support vector machine. When introduced to these algorithms, one is taught that one must provide the algorithm with a kernel function that, intuitively, computes a degree of “similarity” between the objects you are classifying. In practice, one can get pretty far with only this understanding; however, to understand these methods more deeply, one must understand a mathematical object called a reproducing kernel Hilbert space (RKHS). In this post, I will explain the definition of a RKHS and exactly how they produce the kernels used in kernel methods thereby laying a rigorous foundation for a deeper understanding of these methods.
The dot product is a fundamental operation on two Euclidean vectors that captures a notion of similarity between the vectors. In this post, we’ll define the dot product and offer a number of angles for which to intuit the idea captured by this fundamental operation.
Cells are crowded spaces packed with biomolecules colliding and interacting with one another. Despite this chaotic environment, biologists routinely describe intracellular functions using the clean mathematical language of networks. In this post I will attempt to reconcile these two seemingly contradictory perspectives of the cell. This post will serve as a first part in a series of blog posts I hope to write where I will collect and connect some of the works that have helped me better “intuit” biology as a person coming to biology from the field of computer science.
I am currently reading Sun Tzu’s Art of War and, as cliche as it sounds, am finding much wisdom in it. I have been taking notes during my reading and I thought I’d share them in this post. Here I cover Books 1 and 2.
In Part 1 of this series, we introduced the denoising diffusion probabilistic model for modeling and sampling from complex distributions. We described the diffusion model as a model that can generate new samples by learning how to reverse a diffusion process. In this post, we provide more theoretical justification for the objective function used to fit diffusion models and make connections between the diffusion model and other concepts in statistical inference and probabilistic modeling.
Diffusion models are a family of state-of-the-art probabilistic generative models that have achieved ground breaking results in a number of fields ranging from image generation to protein structure design. In Part 1 of this two-part series, I will walk through the denoising diffusion probabilistic model (DDPM) as presented by Ho, Jain, and Abbeel (2020). Specifically, we will walk through the model definition, the derivation of the objective function, and the training and sampling algorithms. We will conclude by walking through an implementation of a simple diffusion model in PyTorch and apply it to the MNIST dataset of hand-written digits.
We human beings use our vision as our chief sense for understanding the world, and thus when we are confronted with data, we try to understand that data through visualization. Dimensionality reduction methods, such as PCA, t-SNE, and UMAP, are approaches designed to enable the visualization of high-dimensional data. Unfortunately, because these methods inevitably distort aspects of the data, these methods are receiving new scrutiny. In this post, I propose that dimensionality reduction requires a “probabilistic” framework of interpretation rather than a “deterministic” one wherein conclusions one draws from a dimensionality reduction plot have some probability of not actually being true of the data. I will propose that this does not mean these plots are not useful. Rather, to evaluate their utility, I will argue that empirical user studies of these methods will shed light on whether these methods provide more benefit or more harm in practice.
Throughout my blog posts on linear algebra, we have proven various properties about invertible matrices. In this post we bring, all of these statements into a single location and form a set of statements called the “invertible matrix theorem”. Each statement in the invertible matrix theorem proves that the matrix is invertible and implies all of the rest of the statements.
In a previous post, we discussed how RNA-seq provides measurements of relative expression between genes rather than measurements of absolute expression. In this post, we will discuss median-ratio normalization: a procedure that attempts to scale each sample’s read counts so that differences in the read counts between samples better reflects differences in absolute expression. We will start by describing the underlying assumption that must be met for median-ratio normalization to work and then walk through the details of the algorithm.
Graphs are ubiqitous mathematical objects that describe a set of relationships between entities; however, they are challenging to model with traditional machine learning methods, which require that the input be represented as vectors. In this post, we will discuss graph convolutional networks (GCNs): a class of neural network designed to operate on graphs. We will discuss the intution behind the GCN and how it is similar and different to the convolutional neural network (CNN) used in computer vision. We will conclude by presenting a case-study training a GCN to classify molecule toxicity.
The determinant of a matrix is often taught as a function that measures the volume of the parallelepiped formed by that matrix’s columns. In this post, we will go a step further in our understanding of the determinant and discuss what the determinant tells us about the linear transformation that is characterized by the matrix. In short, the determinant tells us how much a matrix’s linear transformation grows or shrinks space. The sign of the determinant tells us whether the matrix also inverts space.
The determinant is a function that maps each square matrix to a value that describes the volume of the parallelepiped formed by that matrix’s columns. While this idea is fairly straightforward conceptually, the formula for the determinant is quite confusing. In this post, we will derive the formula for the determinant in an effort to make it less mysterious. Much of my understanding of this material comes from these lecture notes by Mark Demers re-written in my own words.
Matrices are one of the fundamental objects studied in linear algebra. While on their surface they appear like simple tables of numbers, this simplicity hides deeper mathematical structures that they contain. In this post, we will dive into the deeper structures within matrices by discussing three vector spaces that are induced by every matrix: a column space, a row space, and a null space.
Variational autoencoders (VAEs) are a family of deep generative models with use cases that span many applications, from image processing to bioinformatics. There are two complimentary ways of viewing the VAE: as a probabilistic model that is fit using variational Bayesian inference, or as a type of autoencoding neural network. In this post, we present the mathematical theory behind VAEs, which is rooted in Bayesian inference, and how this theory leads to an emergent autoencoding algorithm. We also discuss the similarities and differences between VAEs and standard autoencoders. Lastly, we present an implementation of a VAE in PyTorch and apply it to the task of modeling the MNIST dataset of hand-written digits.
Variational inference (VI) is a mathematical framework for doing Bayesian inference by approximating the posterior distribution over the latent variables in a latent variable model when the true posterior is intractable. In this post, we will discuss a flexible variational inference algorithm, called blackbox VI via the reparameterization gradient, that works “out of the box” for a wide variety of models with minimal need for the tedious mathematical derivations that deriving VI algorithms usually require. We will then use this method to do Bayesian linear regression.
In this post we discuss the row reduction algorithm for solving a system of linear equations that have exactly one solution. We will then show how the row reduction algorithm can be represented as a process involving a sequence of matrix multiplications involving a special class of matrices called elementary matrices. That is, each elementary matrix represents a single elementary row operation in the row reduction algorithm.
In this blog post, we will discuss the relationship between matrices and systems of linear equations. Specifically, we will show how systems of linear equations can be represented as a single matrix equation. Solutions to the system of linear equations can be reasoned about by examining the characteristics of the matrices and vectors in that matrix equation.
A very important concept linear algebra is that of linear independence. In this blog post we present the definition for the span of a set of vectors. Then, we use this definition to discuss the definition for linear independence. Finally, we discuss some intuition into this fundamental idea.
The calculus of variations is a field of mathematics that deals with the optimization of functions of functions, called functionals. This topic was not taught to me in my computer science education, but it lies at the foundation of a number of important concepts and algorithms in the data sciences such as gradient boosting and variational inference. In this post, I will provide an explanation of the functional derivative and show how it relates to the gradient of an ordinary multivariate function.
When first introduced to Euclidean vectors, one is taught that the length of the vector’s arrow is called the norm of the vector. In this post, we present the more rigorous and abstract definition of a norm and show how it generalizes the notion of “length” to non-Euclidean vector spaces. We also discuss how the norm induces a metric function on pairs of vectors so that one can discuss distances between vectors.
Two of the most important relationships in mathematics, namely equality and definition, are both denoted using the same symbol – namely, the equals sign. The overloading of this symbol confuses students in mathematics and computer programming. In this post, I argue for the use of two different symbols for these two fundamentally different operators.
The concept of a vector space is a foundational concept in mathematics, physics, and the data sciences. In this post, we first present and explain the definition of a vector space and then go on to describe properties of vector spaces. Lastly, we present a few examples of vector spaces that go beyond the usual Euclidean vectors that are often taught in introductory math and science courses.
In this post, we discuss invertible matrices: those matrices that characterize invertible linear transformations. We discuss three different perspectives for intuiting inverse matrices as well as several of their properties.
Like entropy, perplexity is an information theoretic quantity that describes the uncertainty of a random variable. In fact, perplexity is simply a monotonic function of entropy and thus, in some sense, they can be used interchangeabley. So why do we need it? In this post, I’ll discuss why perplexity is a more intuitive measure of uncertainty than entropy.
In this post, I will present a high-level explanation of variational inference: a paradigm for estimating a posterior distribution when computing it explicitly is intractable. Variational inference finds an approximate posterior by solving a specific optimization problem that seeks to minimize the disparity between the true posterior and the approximate posterior.
In my previous post, I outlined a conceptual framework for defining and reasoning about “cell types”. Specifically, I noted that the idea of a “cell type” can be viewed as a human-made partition on the universal cellular state space. In this post, I attempt to distill three strategies for partitioning this state space and agreeing on cell type definitions.
The advent of single-cell genomics has brought about new efforts to characterize and catalog all of the cell types in the human body. Despite these efforts, the very definition of a “cell type” is under debate. In this post, I will discuss a conceptual framework for defining cell types as subsets of states in an underlying cellular state space. Moreover, I will link the cellular state space to biomedical ontologies that attempt to capture biological knowledge regarding cell types.
RNA sequencing (RNA-seq) has become a ubiquitous tool in biomedical research for measuring gene expression in a population of cells, or a single cell, across the genome. Despite its ubiquity, RNA-seq is relatively complex and there exists a large research effort towards developing statistical and computational methods for analyzing the raw data that it produces. In this post, I will provide a high level overview of RNA-seq and describe how to interpret some of the common units in which gene expression is measured from an RNA-seq experiment.
In my formal education, I found that the concept of “intrinsic dimensionality” was never explicitly taught; however, it undergirds so many concepts in linear algebra and the data sciences such as the rank of a matrix and feature selection. In this post I will discuss the difference between the extrinsic dimensionality of a space versus its intrinsic dimensionality.
At first glance, the definition for the product of two matrices can be unintuitive. In this post, we discuss three perspectives for viewing matrix multiplication. It is the third perspective that gives this “unintuitive” definition its power: that matrix multiplication represents the composition of linear transformations.
Linear transformations are functions mapping vectors between two vector spaces that preserve vector addition and scalar multiplication. In this post, we show that there exists a one-to-one corresondence between linear transformations between coordinate vector spaces and matrices. Thus, we can view a matrix as representing a unique linear transformation between coordinate vector spaces.
At the core of linear algebra is the idea that matrices represent functions. In this post, we’ll look at a few common, elementary functions and discuss their corresponding matrices.
Matrix-vector multiplication is an operation between a matrix and a vector that produces a new vector. In this post, I’ll define matrix vector multiplication as well as three angles from which to view this concept. The third angle entails viewing matrices as functions between vector spaces
Here, I will introduce the three main ways of thinking about matrices. This high-level description of the multi-faceted way of thinking about matrices would have helped me better intuit matrices when I was first introduced to them in my undergraduate linear algebra course.
Gaussian mixture models are a very popular method for data clustering. Here I will define the Gaussian mixture model and also derive the EM algorithm for performing maximum likelihood estimation of its paramters.
At the heart of of a number of important machine learning algorithms, such as spectral clustering, lies a matrix called the graph Laplacian. In this post, I’ll walk through the intuition behind the graph Laplacian and describe how it represents the discrete analogue to the Laplacian operator on continuous multivariate functions.
In this post, I will discuss an analogy that I find useful for thinking about what it means to “understand” something: True understanding of a concept is akin to “seeing” the concept in its native three-dimensional space, whereas partial understanding is merely seeing a two-dimensional projection of that inherently three-dimensional concept.
The mathematical field of information theory attempts to mathematically describe the concept of “information”. In the first two posts, we discussed the concepts of self-information and information entropy. In this post, we step through Shannon’s Source Coding Theorem to see how the information entropy of a probability distribution describes the best-achievable efficiency required to communicate samples from the distribution.
The mathematical field of information theory attempts to mathematically describe the concept of “information”. In this series of posts, I will attempt to describe my understanding of how, both philosophically and mathematically, information theory defines the polymorphic, and often amorphous, concept of information. In the first post, we discussed the concept of self-information. In this second post, we will build on this foundation to discuss the concept of information entropy.
The mathematical field of information theory attempts to mathematically describe the concept of “information”. In this series of posts, I will attempt to describe my understanding of how, both philosophically and mathematically, information theory defines the polymorphic, and often amorphous, concept of information. In this first post, I will describe Shannon’s self-information.
The evidence lower bound is an important quantity at the core of a number of important algorithms used in statistical inference including expectation-maximization and variational inference. In this post, I describe its context, definition, and derivation.
Covariance quantifies to what extent two random variables are linearly correlated. In this post, I will outline a visualization of covariance that helped me better intuit this concept.
Expectation-maximization (EM) is a popular algorithm for performing maximum-likelihood estimation of the parameters in a latent variable model. In this post, I discuss the theory behind, and intuition into this algorithm.
In this series of posts, I present my understanding of some basic concepts in measure theory — the mathematical study of objects with “size”— that have enabled me to gain a deeper understanding into the foundations of probability theory.
In this series of posts, I present my understanding of some basic concepts in measure theory — the mathematical study of objects with “size”— that have enabled me to gain a deeper understanding into the foundations of probability theory.
In this series of posts, I will present my understanding of some basic concepts in measure theory — the mathematical study of objects with “size”— that have enabled me to gain a deeper understanding into the foundations of probability theory.
