A determinant-free method to simulate the parameters of large Gaussian fields

Together with Louis Ellam, Iain Murray, and Mark Girolami, we just published / arXived a new article on dealing with large Gaussian models. This is slightly related to the open problem around the GMRF model in our Russian Roulette paper back a while ago.

We propose a determinant-free approach for simulation-based Bayesian inference in high-dimensional Gaussian models. We introduce auxiliary variables with covariance equal to the inverse covariance of the model. The joint probability of the auxiliary model can be computed without evaluating determinants, which are often hard to compute in high dimensions. We develop a Markov chain Monte Carlo sampling scheme for the auxiliary model that requires no more than the application of inverse-matrix-square-roots and the solution of linear systems. These operations can be performed at large scales with rational approximations. We provide an empirical study on both synthetic and real-world data for sparse Gaussian processes and for large-scale Gaussian Markov random fields.

Article is here. Unfortunately, the journal is not open-access, but the arXiv version is.

Unbiased Big Bayes without sub-sampling bias: Paths of Partial Posteriors

Together with Dino Sejdinovic and Mark Girolami, I recently arxived our second draft of a paper on the popular topic of how to scaleup Bayesian inference when facing large datasets.

This project is yet another method to scale up Bayesian inference via sub-sampling. But unlike most other work on this field, we do not use sub-sampling within an MCMC algorithm to simulate from the posterior distribution of interest. Rather, we use sub-sampling to create a series of easier inference problems, whose results we then combine to get the answer to the question we were originally asking — in an estimation sense, without full posterior simulation. I’ll describe the high-level ideas now.

Updates on the project:

Let’s assume the following desiderata for large-scale Bayesian estimators:

  • Average computational costs sub-linear in the data size $N$.
  • Bounded variance (even for growing $N$) and can be controlled.
  • No sub-sampling bias is introduced. The method inherits the usual finite time bias from MCMC though.
  • Not limited to certain models or inference schemes.
  • Trade off redundancy in the data against computational savings.
  • Perform competitive in practice.

Bayesian inference usually involves integrating a functional $\varphi:\Theta\rightarrow\mathbb{R}$ over a posterior distribution $$\phi=\int d\theta\varphi(\theta)\underbrace{p(\theta|{\cal D})}_{\text{posterior}},$$ where $$p(\theta|{\cal D})\propto\underbrace{p({\cal D}|\theta)}_{\text{likelihood data}}\underbrace{p(\theta)}_{\text{prior}}.$$ For example, to compute the predictive posterior mean for a linear regression model. This is often done by constructing an MCMC algorithm to simulate $m$ approximately iid samples from approximately $\theta^{(j)}\sim p(\theta|{\cal D})$ (the second “approximately” here refers to the fact that MCMC is biased for any finite length of the chain, and the same is true for our method) and then performing Monte Carlo integration $$\phi\approx\frac{1}{m}\sum_{j=1}^{m}\varphi(\theta^{(j)}).$$ Constructing such Markov chains can quickly become infeasible as usually, all data needs to be touched in every iteration. Take for example the standard Metropolis-Hastings transition kernel to simulate from $\pi(\theta)\propto p(\theta|{\cal D})$, where at a current point $\theta^{(j)}$, a proposed new point $\theta’\sim q(\theta|\theta^{(j)})$ is accepted with probability $$\min\left(\frac{\pi(\theta’)}{\pi(\theta^{(j)})}\times\frac{q(\theta^{(j)}|\theta’)}{q(\theta’|\theta^{(j)})},1\right).$$ Evaluating $\pi$ requires to iterate through the whole dataset. A natural question therefore is: is it possible to only use subsets of $\mathcal{D}$?

Existing work. There has been a number of papers that tried to use sub-sampling within the transition kernel of MCMC:

All these methods have in common that they either introduce bias (in addition to the bias caused by MCMC), have no convergence guarantees, or mix badly. This comes from the fact that it is extremely difficult to modify the Markov transition kernel and maintain its asymptotic correctness.

In our paper, we therefore take a different perspective. If the goal is to solve an estimation problem as the one above, we should not make our life hard trying to simulate. We construct a method that directly estimates posterior expectations — without simulating from the full posterior, and without introducing sub-sampling bias.

The idea is very simple:

  1. Construct a series of partial posterior distributions $\tilde{\pi}_{l}:=p({\theta|\cal D}_{l})\propto p({\cal D}_{l}|\theta)p(\theta)$ from subsets of sizes $$\vert\mathcal{D}_1\vert=a,\vert\mathcal{D}_2\vert=2^{1}a,\vert\mathcal{D}_3\vert=2^{2}a,\dots,\vert\mathcal{D}_L\vert=2^{L-1}a$$ of the data. This gives $$p(\theta)=\tilde{\pi}_{0}\rightarrow\tilde{\pi}_{1}\rightarrow\tilde{\pi}_{2}\rightarrow\dots\rightarrow\tilde{\pi}_{L}=\pi_{N}=p({\theta|\cal D}).$$
  2. Compute posterior expectations $\phi_{t}:=\hat{\mathbb{E}}_{\tilde{\pi}_{t}}\{\varphi(\theta)\}$ for each of the partial posteriors. This can be done for example MCMC, or other inference methods. This gives a sequence of real valued partial posterior expectations — that converges to the true posterior.
  3. Use the debiasing framework, by Glynn & Rhee. This is a way to unbiasedly estimate the limit of a sequence without evaluating all elements. Instead, one randomly truncates the sequence at term $T$ (which should be “early” in some sense, $T$ is a random variable), and then computes $$\phi^*=\sum_{t=1}^T \frac{\phi_{t}-\phi_{t-1}}{\mathbb{P}\left[T\geq t\right]}.$$ This is now an unbiased estimator for the full posterior expectation. The intuition is that one can rewrite any sequence limit as a telescopic sum $$\lim_{t\to\infty} \phi_t=\sum_{t=1}^\infty (\phi_{t}-\phi_{t-1})$$ and then randomly truncate the infinite sum and divide through truncation probabilities to leave the expectation untouched, i.e. $$\mathbb{E}[\phi^*]=\phi,$$where $\phi$ is the sequence limit.
  4. Repeat $R$ times for reducing the estimator variance as $1/R$.

In the paper, we choose the random truncation time as $$\mathbb{P}(T=t)\propto 2^{-\alpha t}.$$ From here, given a certain choice of $\alpha$, it is straight-forward to show the average computational cost of this estimator is $\mathcal{O} (N^{1-\alpha}),$ which is sub-linear in $N$. We also show that variance is finite and can be bounded even when $N\to\infty$. This statement is modulo certain conditions on the $\phi_t$. In addition, the choice of the $\alpha$-parameter of the random truncation time $T$ is crucial for performance. See the paper for details how to set it.

Experimental results are very encouraging. On certain models, we are able to estimate model parameters accurately before plain MCMC even has finished a single iteration. Take for example a log-normal model $$p(\mu,\sigma|{\cal D})\propto\prod_{i=1}^{N}\log{\cal N}(x_{i}|\mu,\sigma^{2})\times\text{flat prior}$$ and functional $$\varphi((\mu,\sigma))=\sigma.$$ We chose the true parameter $\sigma=\sqrt{2}$ and ran our method in $N=2^{26}$ data. The pictures below contain some more details.

Comparison to stochastic variational inference is something we only thought of quite late in the project. As we state in the paper, as mentioned in my talks, and as pointed out in feedback from the community, we cannot do uniformly better than MCMC. Therefore, comparing to full MCMC is tricky — even though there exist trade-offs of data redundancy and model complexity that we can exploit as showed in the above example. Comparing to other “limited data” schemes really showcases the strength (and generality) of our method.

We chose to compare against “Gaussian Processes for Big Data“, which puts together a decade of engineering of variational lower bounds, the inducing features framework for GPs, and stochastic variational inference to solve Gaussian Process regression on roughly a million data points. In contrast, our scheme is very general and not specifically designed for GPs. However, we are able to perform similar to this state of the art method. On predicting airtime delays, we reach a slightly lower RMSE at comparable computing effort. The paper contains details on our GP, which is slightly different (but comparable) to the one used in the GP for Big Data paper. Another neat detail of the GP example is that the likelihood does not factorise, which is in contrast to most other Bayesian sub-sampling schemes.

In summary, we arrive at

  • Computational costs sub-linear in $N$
  • No introduced sub-sampling bias (in addition to MCMC, if used)
  • Finite, and controllable variance
  • A general scheme that is not limited to MCMC nor to factorising likelihoods
  • A scheme that trades off redundancy for computational savings
  • A practical scheme that performs competitive to state-of-the-art methods and that has no problems with transition kernel design.
  • See the paper for a list of weaknesses and open problems.

Russian Roulette for intractable Likelihoods

Updates:
December 2015
: The journal version of the paper finally got published — after just three years.

While I was working at UCL’s Statistics Department in winter, I got involved into a very exciting project in the group of Mark Girolami. It is based around the Pseudo-Marginal Metropolis-Hastings algorithm. In 2003, a Genetics paper [1] described an approach to sample a distribution using the standard Metropolis-Hastings algorithm when the density function is not available by simply replacing it with an unbiased estimate.

For a standard Bayesian inference problem with likelihood $\pi(y|\theta) $, prior $\pi(\theta)$, and a proposal $Q$, rather than using the standard M-H ratio $$\frac{\pi(y|\theta^{\text{new}})}{pi(y|\theta)}\times\frac{\pi(\theta^{\text{new}})}{\pi(\theta)}\times \frac{Q(\theta|\theta^{\text{new}})}{Q(\theta^{\text{new}}|\theta)},$$ the likelihood is replaced by an unbiased estimator as

$$\frac{\hat{\pi}(y|\theta^{\text{new}})}{\hat{\pi}(y|\theta)}\times\frac{\pi(\theta^{\text{new}})}{\pi(\theta)}\times \frac{Q(\theta|\theta^{\text{new}})}{Q(\theta^{\text{new}}|\theta)}.$$ Remarkably  the resulting Markov chain converges to the same posterior distribution as the exact algorithm. The approach was later formalised and popularised in [2].

In our project, we exploited this idea to perform inference over models whose likelihood functions are intractable. Example of such intractable likelihoods are for example Ising models or, even simpler, very large Gaussian models. Both of those models’ normalising constants are very hard to compute. We came up with a way of producing unbiased estimators for the likelihoods, which are based on writing likelihoods as an infinite sum, and then truncating it stochastically.

Producing unbiased estimators for the Pseudo-Marginal approach is a very challenging task. Estimates have to be strictly positive. This can be achieved via pulling out the sign of the estimates in the final Monte-Carlo integral estimate and add a correction term (which increases the variance of the estimator). This problem is studied under the term Sign problem. The next step is to write the likelihood function as an infinite sum. In our paper, we do this for a geometrically titled correction of a biased estimator obtained by an approximation such as importance sampling estates, upper bounds, or deterministic approximations, and for likelihoods based on the exponential function.

I in particular worked on the exponential function estimate. We took a very nice example from spatial statistics: a worldwide grid of ozone measurements from a satellite that consists of a about 173,405 measurements. We fitted a simple Gaussian model whose covariance matrices are massive (and sparse). In such models of the form $$ \log \mathcal{N}_x(\mu,\Sigma))=-\log(\det(\Sigma)) – (\mu-x)^T \Sigma^{-1}(\mu-x) + C, $$ the normalising constant involves a log-determinant of such a large matrix. This is impossible using classical methods such as Cholesky factorisation $$\Sigma=LL^T \Rightarrow \log(\det(\Sigma))=2\sum_i\log(L_{ii}),$$ due to memory shortcomings: It is not possible to store the Cholesky factor $L$ since it is not in general sparse. We therefore constructed an unbiased estimator using a very neat method based on graph colourings and Krylov methods from [3].

This unbiased estimator of the log-likelihood is then turned into a (positive) unbiased estimator of the likelihood itself via writing the exponential function as an infinite series $$\exp(\log(\det(\Sigma)))=1+\sum_{i=1}^\infty \frac{\log(\det(\Sigma))^i}{i!}. $$

We then construct an unbiased estimator of this series by playing Russian Roulette: We evaluate the terms in the series and plug in a different estimator for $\log(\det(\Sigma))$ for every $i$; once those values are small, we start flipping a coin every whether we continue the series or not. If we do continue, we add some weights that ensure unbiasedness. We also ensure that it is less likely to continue in every iteration so that the procedure eventually stops. This basic idea (borrowed from Physics papers from some 20 years ago) and some technical details and computational tricks then give an unbiased estimator of the likelihood of the log-determinant of our Gaussian model and can therefore be plugged into Pseudo-Marginal M-H. This allows to perform Bayesian inference over models of sizes where it has been impossible before.

More details can be found on our project page (link, see ozone link), and in our paper draft on arXiv (link). One of my this year’s Google summer of Code projects for the Shogun Machine-Learning toolbox is about producing a sophisticated implementation of log-determinant estimators (link). Pretty exciting!

[1]: Beaumont, M. A. (2003). Estimation of population growth or decline in genetically monitored populations. Genetics 164 1139–1160.
[2]: Andrieu, C., & Roberts, G. O. (2009). The pseudo-marginal approach for efficient Monte Carlo computations. The Annals of Statistics, 37(2), 697–725.
[3]: Aune, E., Simpson, D., & Eidsvik, J. (2012). Parameter Estimation in High Dimensional Gaussian Distributions.