### NIPS 2016 Summaries

I did not attend this year’s NIPS, but I’ve gathered many summaries published online by those who did attend the conference.

I did not attend this year’s NIPS, but I’ve gathered many summaries published online by those who did attend the conference.

Previously we covered Variational Autoencoders (VAE) — popular inference tool based on neural networks. In this post we’ll consider, a followup work from Torronto by Y. Burda, R. Grosse and R. Salakhutdinov, Importance Weighted Autoencoders (IWAE). The crucial contribution of this work is introduction of a new lower-bound on the marginal log-likelihood \(\log p(x)\) which generalizes ELBO, but also allows one to use less accurate approximate posteriors \(q(z \mid x, \Lambda)\).

On a dessert we’ll discuss another paper, Variational inference for Monte Carlo objectives by A. Mnih and D. Rezende which aims to broaden the applicability of this approach to models where reparametrization trick can not be used (e.g. for discrete variables).

So far we had a little of “neural” in our VI methods. Now it’s time to fix it, as we’re going to consider Variational Autoencoders (VAE), a paper by D. Kingma and M. Welling, which made a lot of buzz in ML community. It has 2 main contributions: a new approach (AEVB) to large-scale inference in non-conjugate models with continuous latent variables, and a probabilistic model of autoencoders as an example of this approach. We then discuss connections to Helmholtz machines — a predecessor of VAEs.

In the previous post we covered Stochastic VI: an efficient and scalable variational inference method for exponential family models. However, there’re many more distributions than those belonging to the exponential family. Inference in these cases requires significant amount of model analysis. In this post we consider Black Box Variational Inference by Ranganath et al. This work just as the previous one comes from David Blei lab — one of the leading researchers in VI. And, just for the dessert, we’ll touch upon another paper, which will finally introduce some neural networks in VI.

In the previous post I covered well-established classical theory developed in early 2000-s. Since then technology has made huge progress: now we have much more data, and a great need to process it and process it fast. In big data era we have huge datasets, and can not afford too many full passes over it, which might render classical VI methods impractical. Recently M. Hoffman et al. dissected classical Mean-Field VI to introduce stochasticity right into its heart, which resulted in Stochastic Variational Inference.

As a member of Bayesian methods research group I’m heavily interested in Bayesian approach to machine learning. One of the strengths of this approach is ability to work with hidden (unobserved) variables which are interpretable. This power however comes at a cost of generally intractable exact inference, which limits the scope of solvable problems.

Another topic which gained lots of momentum in Machine Learning recently is Deep Learning, of course. With Deep Learning we can now build big and complex models that outperform most hand-engineered approaches given lots of data and computational power. The fact that Deep Learning needs a considerable amount of data also requires these methods to be scalable — a really nice property for any algorithm to have, especially in a Big Data epoch.

Given how appealing both topics are it’s not a surprise there’s been some work to marry these two recently. In this series of blogsposts I’d like to summarize recent advances, particularly in variational inference. This is not meant to be an introductory discussion as prior familiarity with classical topics (Latent variable models, Variational Inference, Mean-field approximation) is required, though I’ll introduce these ideas anyway just to remind it and setup the notation.

During work on my machine learning project I was needed to perform some quite computation-heavy calculations several times — each time with a bit different inputs. These calculations were CPU and memory bound, so just spawning them all at once would just slow down overall running time because of increased amount of context switches. Yet running 4 (=number of cores in my CPU) of them at a time (actually, 3 since other applications need CPU, too) should speed it up.

Fortunately, I have an old laptop with 2 cores as well as an access to somewhat more modern machine with 4 cores. That results in 10 cores spread across 3 machines (all of`em have some version of GNU Linux installed). The question was how to exploit such a treasury.

It’s well known that lower bound for sorting problem (in general case) is \(\Omega(n \log n)\). The proof I was taught is somewhat involved and is based on paths in “decision” trees. Recently I’ve discovered an information-theoretic approach (or reformulation) to that proof.

Once upon a time I was asked (well, actually a question wasn’t for me only, but for whole habrahabr’s community) is it possible to implement namespaced methods in JavaScript for built-in types like:

```
5..rubish.times(function() { // this function will be called 5 times
console.log("Hi there!");
});
"some string".hask.map(function(c) { return c.hask.code(); });
// equivalent to
"some string".split('').map(function(c) { return c.charCodeAt(); });
"another string".algo.lcp("annotation");
// returns longest common prefix of two strings
```

As you can see at the link, it’s possible using ECMAScript 5 features. And that’s how:

Suppose we have an expression like `(5+5 * (x^x-5 | y && 3))`

and we’d like to get some computer-understandable representation of that expression, like:

`ADD Token[5] (MUL Token[5] (AND (BIT_OR (XOR Token[x] (SUB Token[x] Token[5])) Token[y]) Token[3])`

In case if you don’t know how to do that or are looking for the solutin right now, you should know that I’m not going to present a correct solution. This post is just a joke. You should use either a Shunting-yard algorithm or a recursive descent parser.

So if you’re ready for madness… Let’s go!