###### tags: `one-offs` `concentration` `martingales` `convexity` # Hoeffding's Inequality by Convex Ordering **Overview**: In this note, I prove Hoeffding's inequality from the perspectives of martingales and convex ordering. ## The Basic Construction Let $-\infty<a<x<b<\infty$, and define a random variable $M$ with law $M\left(x;a,b\right)$ by \begin{align} M=\begin{cases} a & \text{w.p. }\frac{b-x}{b-a}\\ b & \text{w.p. }\frac{x-a}{b-a}. \end{cases} \end{align} One can then check that M has expectation $x$. ## The Hoeffding Martingale Now, consider a random variable $X$ which takes values in a subset of $\left[ a, b\right]$, and conditional on $X$, sample $Y \sim M \left( X; a, b \right)$. Following our previous observation, we see that $\mathbf{E} \left[Y\mid X\right] = X$, i.e. the ordered pair $\left( X, Y\right)$ is a martingale. Note also that marginally, $Y \overset{d}{=} M \left( \mathbf{E} \left[ X \right]; a, b\right)$. Because $\left(X, Y\right)$ form a martingale, by Jensen's inequality, it follows that they are also convex-ordered, i.e. for any convex $F$, it holds that $\mathbf{E} \left[ F\left(X\right) \right] \leqslant \mathbf{E} \left[ F\left(Y\right) \right]$. (Actually, a result of Strassen establishes that this characterisation is tight, i.e. being convex-ordered also implies the martingale property.) In particular, letting $F \left( x \right) = \exp \left( t \cdot x \right)$ for $t \in \mathbf{R}$, it holds that the moment-generating function of $X$ is dominated by the moment-generating function of $Y = M \left( \mathbf{E} \left[ X \right]; a, b \right)$. Because $Y$ is a two-point distribution, its MGF is thus relatively easy to control. Indeed, the MGF of $Y$ is essentially that of a standard biased coin flip (up to shifting and scaling). Using basic arguments from the theory of exponential families, one sees that \begin{align} \left(\log M_{Y}\right)^{''}\left(t\right)\leqslant\frac{\left(b-a\right)^{2}}{4} \end{align} uniformly in $t$, from which one can bound \begin{align} M_{Y}\left(t\right) \leqslant \exp \left(t \cdot \mathbf{E} \left[X\right] + \frac{t^{2}}{8} \cdot \left(b-a\right)^{2} \right). \end{align} One can then see that \begin{align} M_{X} \left(t\right) \leqslant M_{Y} \left(t\right) \leqslant \exp \left(t \cdot \mathbf{E} \left[X\right] + \frac{t^{2}}{8} \cdot \left(b-a\right)^{2}\right), \end{align} and this is essentially the content of Hoeffding's lemma, from which one often deduces Hoeffding's (concentration) inequality. For me, this is nice, because it highlights that the inequality is essentially a comparison which says that "bounded RVs fluctuate less than a (n appropriate) 2-point random RV". The specifics of the MGF and the tail inequality are the practically relevant consequences of this. A potentially interesting variant of this idea would be to consider a construction where $Y$ takes values on e.g. the set $\left[ a, s \right] \cup \left[ t, b \right]$ for some $a<s<t<b$. A similar ordering structure ought to be present, and perhaps there is some nice interpolation result along these lines. For what it's worth, there is also a construction in [this paper](https://arxiv.org/abs/1909.12067) which is not dissimilar to this idea (but is certainly more ambitious in its scope than Hoeffding's lemma!).
Last changed by  
 
sp-monte-carlo
740

Read more

Denoising Diffusions

Overview: In this note, I log some basic observations about diffusion-based generative models.

8/14/2023
Nested Structure in MCMC Algorithms

Overview: In this note, I describe some aspects of hierarchical structure in MCMC algorithms, and how they can be of theoretical and practical relevance.

8/9/2023
The "Approximation to What?" Principle

Overview: In this note, I discuss a recurrent question which can be used to generate research questions about methods of all sorts. I then discuss a specific instance of how this question has proved fruitful in the theory of optimisation algorithms. Methods and Approximations A nice story is that when Brad Efron derived the bootstrap, it was done in service of the question “What is the jackknife an approximation to?”. I can't help but agree that there's something quite exciting about research questions which have this same character of ''What is (this existing thing) an approximation to?''. One bonus tilt on this which I appreciate is that there can be multiple levels of approximation, and hence many answers to the same question. One well-known example is gradient descent, which can be viewed as an approximation to the proximal point method, which can then itself be viewed as an approximation to a gradient flow. There are probably even more stops along the way here. In this case, there is even the perspective that from the perspective of mathematical theory, there may be at least as much to be gained by stopping off at the proximal point interpretation, as there is from the gradient flow perspective. My experience is that generalist applied mathematicians get to grips with the gradient flow quickly, but optimisation theorists can squeeze more out of the PPM formulation. There is thus some hint that using this 'intermediate' approximation can be particularly insightful in its own right. It would be interesting to collect more examples with this character.

5/22/2023
The Advent of Optimisation in Statistics

Overview: I asked the following historical question online: “It is fairly well-documented that the turning point for the Bayesian statistics community towards widespread use of MCMC was in the early 1990s, in light of papers on Gibbs sampling e.g. Gelfand and Smith, Smith and Roberts, and others. Is there a comparable landmark paper (or otherwise) which is historically associated with the advent of { nonlinear, convex, etc. } optimisation in modern statistics?” I received some interesting responses, which I document (with some editing) in this note. Responses “Surely it would be something from the advent of Support Vector Machines. Ideas from $\ell^{1}$ regularisation and similar were of course huge, but tools from convex optimisation (e.g. duality and related notions) were certainly understood and used before the sparsity era.” “Perhaps it occurred somewhere in the transition from linear models to generalised linear models. The initial focus on linear models arguably allowed much of the field to avoid more general optimisation problems, but with the increasing popularity of non-linear link functions, this would have had to become more difficult. Still, GLMs were (and perhaps still are) nearly synonymous with Iteratively-Reweighted Least Squares, and perhaps this weakens the connection to more general convex optimisation.” “The 1977 paper of Dempster, Laird and Rubin on the EM Algorithm might have been one of the earlier works to focus on maximum likelihood, rather than minimum least squares, as an optimisation / estimation principle. However, the EM algorithm is not strictly in the spirit of other procedures from within the optimisation world (at least without a bit of massaging).”

5/22/2023
Read more from sp-monte-carlo
Published on HackMD