# ISIT 2014: how many samples do we need?

Due to jetlag, my CAREER proposal deadline, and perhaps a bit of general laziness, I didn’t take as many notes at ISIT as I would have, so my posting will be somewhat light (in addition to being almost a month delayed). If someone else took notes on some talks and wants to guest-post on it, let me know!

Strong Large Deviations for Composite Hypothesis Testing
Yen-Wei Huang (Microsoft Corporation, USA); Pierre Moulin (University of Illinois at Urbana-Champaign, USA)
This talk was actually given by Vincent Tan since neither of the authors could make it (this seems to be a theme of talks I’ve attended this summer. The paper was about testing a simple hypothesis $H_1$ versus a composite hypothesis $H_0$ where under $H_0$ the observations are i.i.d. with respect to one of possibly $k$ different distributions. There are therefore $k$ different errors and the goal is to characterize these errors when we ask for the probability of true detection to be greater than $1 - \epsilon$. This is a sort of generalized Neyman-Pearson setup. They look at the vector of log-likelihood ratios and show that a threshold test is nearly optimal. At the time, I understood the idea of the proof, but I think it’s one of things where you need to really read the paper.

Randomized Sketches of Convex Programs with Sharp Guarantees
Mert Pilanci (University of California, Berkeley, USA); Martin J. Wainwright (University of California, Berkeley, USA)
This talk was about using random projections to lower the complexity of solving a convex program. Suppose we want to minimize $\| Ax - y \|^2$ over $x$ given $y$. A sketch would be to solve $\| SAx - Sy \|^2$ where $S$ is a random projection. One question is how to choose $A$. They show that choosing $S$ to be a randomized Hadamard matrix (the paper studies Gaussian matrices), then the objective value of the sketched program is at most $(1 + \epsilon)^2$ times the value of the original program as long as the the number of rows of $S$ is larger than $O( \epsilon^{-2} \mathbb{W}^2(A \mathcal{K}))$, where $\mathbb{W}(A \mathcal{K})$ is the Gaussian width of the tangent cone of the contraint set at the optimum value. For more details look at their preprint on ArXiV.

On Efficiency and Low Sample Complexity in Phase Retrieval
Youssef Mroueh (MIT-IIT, USA); Lorenzo Rosasco (DIBRIS, Unige and LCSL – MIT, IIT, USA)
This was another talk not given by the authors. The problem is recovery of a complex vector $x_0 \in \mathbb{C}^n$ from phaseless measurements of the form $b_i = |\langle a_i, x_0 \rangle|^2$ where $a_i$ are complex spherically symmetric Gaussian vectors. Recovery from such measurements is nonconvex and tricky, but an alternating minimizing algorithm can reach a local optimum, and if you start it in a “good” initial position, it will find a global optimum. The contribution of this paper is provide such a smart initialization. The idea is to “pair” the measurements to create new measurements $y_i = \mathrm{sign}( b_i^{(1)} - b_i^{(2)} )$. This leads to a new problem (with half as many measurements) which is still hard, so they find a convex relaxation of that. I had thought briefly about such sensing setups a long time ago (and by thought, I mean puzzled over it at a coffeshop once), so it was interesting to see what was known about the problem.

Sorting with adversarial comparators and application to density estimation
Jayadev Acharya (University of California, San Diego, USA); Ashkan Jafarpour (University of California, San Diego, USA); Alon Orlitsky (University of California, San Diego, USA); Ananda Theertha Suresh (University of California, San Diego, USA)
Ashkan gave this talk on a problem where you have $m$ samples from an unknown distribution $p$ and a set of distributions $\{q_1, q_2, \ldots, q_n\}$ to compare against. You want to find the distribution that is closest in $\ell_1$. One way to do this is via Scheffe tournament tht compares all pairs of distributions — this runs in time $n^2$ time. They show a method that runs in $O(n)$ time by studying the structure of the comparators used in the sorting method. The motivation is that running comparisons can be expensive (especially if they involve human decisions) so we want to minimize the number of comparisons. The paper is significantly different than the talk, but I think it would definitely be interesting to those interested in discrete algorithms. The density estimation problem is really just a motivator — the sorting problem is far more general.

# ICML 2014: a few more papers

After a long stint of proposal writing, I figured I should catch up on some old languishing posts. So here’s a few quick notes on the remainder of ICML 2014.

• Fast Stochastic Alternating Direction Method of Multipliers (Wenliang Zhong; James Kwok): Most of the talks in the Optimization II session were on ADMM or stochastic optimization, or both. This was int he last category. ADMM can have rather high-complexity update rules, especially on large, complex problems, so the goal is to lower the complexity of the update step by making it stochastic. The hard part seems to be controlling the step size.
• An Asynchronous Parallel Stochastic Coordinate Descent Algorithm (Ji Liu; Steve Wright; Christopher Re; Victor Bittorf; Srikrishna Sridhar): The full version of this paper is on ArXiV. The authors look at a multi-core lock-free stochastic coordinate descent method and characterize how many cores you need to get linear speedups — this depends on the convexity properties of the objective function.
• Communication-Efficient Distributed Optimization using an Approximate Newton-type Method (Ohad Shamir; Nati Srebro; Tong Zhang): This paper looked 1-shot “average at the end” schemes where you divide the data onto multiple machines, have them each train a linear predictor (for example) using stochastic optimization and then average the results. This is just averaging i.i.d. copies of some complicated random variable (the output of an optimization) so you would expect some variance reduction. This method has been studied by a few people int the last few years. While you do get variance reduction, the bias can still be bad. On the other extreme, communicating at every iteration essentially transmits the entire data set (or worse) over the network. They propose a new method for limiting communication by computing an approximate Newton step without approximating the full Hessian. It works pretty well.
• Lower Bounds for the Gibbs Sampler over Mixtures of Gaussians (Christopher Tosh; Sanjoy Dasgupta): This was a great talk about how MCMC can be really slow to converge. The model is a mixture of Gaussians with random weights (Dirichlet) and means (Gaussian I think). Since the posterior on the parameters is hard to compute, you might want to do Gibbs sampling. They use conductance methods to get a lower bound on the mixing time of the chain. The tricky part is that the cluster labels are permutation invariant — I don’t care if you label clusters (1,2) versus (2,1), so they need to construct some equivalence classes. They also have further results on what happens when the number of clusters is misspecified. I really liked this talk because MCMC always seems like black magic to me (and I even used it in a paper!)
• (Near) Dimension Independent Risk Bounds for Differentially Private Learning (Prateek Jain; Abhradeep Guha Thakurta): Abhradeep presented a really nice paper with a tighter analysis of output and objective perturbation methods for differentially private ERM, along with a new algorithm for risk minimization on the simplex. Abhradeep really only talked about the first part. If you focus on scalar regret, they show that essentially the error comes from taking the inner product of a noise vector with a data vector. If the noise is Gaussian then the noise level is dimension-independent for bounded data. This shows that taking $(\epsilon,\delta)$-differential privacy yield better sample complexity results than $(\epsilon,)$-differential privacy. This feels similar in flavor to a recent preprint on ArXiV by Beimel, Nissim, and Stemmer.
• Near-Optimally Teaching the Crowd to Classify (Adish Singla; Ilija Bogunovic; Gabor Bartok; Amin Karbasi; Andreas Krause): This was one of those talks where I would have to go back to look at the paper a bit more. The idea is that you want to train annotators to do better in a crowd system like Mechanical Turk — which examples should you give them to improve their performance? They model the learners as doing some multiplicative weights update. Under that model, the teacher has to optimize to pick a batch of examples to give to the learner. This is hard, so they use a submodular surrogate function and optimize over that.
• Discrete Chebyshev Classifiers (Elad Eban; Elad Mezuman; Amir Globerson): This was an award-winner. The setup is that you have categorical (not numerical) features on $n$ variables and you want to do some classification. They consider taking pairwise inputs and compute for each tuple $(x_i, x_j, y)$ a marginal $\mu_{ij}(x_i, x_j, y)$. If you want to create a rule $f: \mathcal{X} \to \mathcal{Y}$ for classification, you might want to pick one that has best worst-case performance. One approach is to take the one which has best worst-case performance over all joint distributions on all variables that agree with the empirical marginals. This optimization looks hard because of the exponential number of variables, but they in fact show via convex duality and LP relaxations that it can be solved efficiently. To which I say: wow! More details are in the paper, but the proofs seem to be waiting for a journal version.

# Postdoc opportunity at Aalto University

Aalto University is a new university with over a century of experience. Created from a high-profile merger between three leading universities in Finland – the Helsinki School of Economics, Helsinki University of Technology and the University of Art and Design Helsinki – Aalto University opens up new possibilities for strong multidisciplinary education and research. The university has 20 000 students and a staff of 5,000 including 350 professors.

The stochastics research group at the Department of Mathematics and Systems Analysis is currently undergoing a period of regeneration, as new associate and assistant professors have been employed to replace previous long-term faculty, and several new young researchers are being recruited with the aim of significant growth. To strengthen this line of development, we are now seeking to hire a postdoctoral researcher with a PhD in mathematics or a related area.

The postdoctoral researcher will carry out research in collaboration with the stochastics research group, with a small amount of teaching duties included. The salary is competitive, based on experience and qualifications, and includes occupational health and a travel budget for international conferences and workshops. The position is for one year with a possible extension for another year, starting preferably in September 2014 and no later than January 2015.

Further information and instructions for applying:

Please feel free to forward this message to colleagues and potential candidates. The application deadline is 13 June 2014.

# expected time for an optimal game of memory

Via Dan Katz, I learned about a recent problem (warning: JSTOR) from the American Mathematical Monthly (a publication of the Mathematics Association of America, making shaded icosahedrons look cool for almost 100 years):

What to Expect in a Game of Memory
Author(s): Daniel J. Velleman, Gregory S. Warrington
The American Mathematical Monthly, Vol. 120, No. 9 (November), pp. 787-805

The game of memory is played with a deck of $n$ pairs of cards. The cards in each pair are identical. The deck is shuffled and the cards laid face down. A move consists of flipping over first one card and then another. The cards are removed from play if they match. Otherwise, they are flipped back over and the next move commences. A game ends when all pairs have been matched. We determine that, when the game is played optimally, as $n \rightarrow \infty$
• The expected number of moves is $(3 - 2 \ln 2) n + 7/8 - 2 \ln 2 \approx 1.61 n$.
• The expected number of times two matching cards are unwittingly flipped over is $\ln 2$.
• The expected number of flips until two matching cards have been seen is $2^{2n} / \binom{2n}{n} \approx \sqrt{\pi n}$.

This is not a competitive game of memory, but the singe player version. It’s a kind of explore-exploit tradeoff with a simple structure — if you know how to exploit, do it. Note that one could do $2 n$ moves by flipping every card over once (there are $2n$ cards) to learn all of their identities and then removing all of the pairs one by one. The better strategy is

1. Remove any known pair.
2. If no known pair is known, flip a random unknown card and match it if you can.
3. If the first card is not matchable, flip another random unknown card to learn its value (and remove the pair if it matches.

This strategy exploits optimally when it can and explores optimally when it can’t. The second bullet point in the abstract is the gain from getting lucky, i.e. two randomly drawn cards matching.

The paper is an interesting read, but the arguments are all combinatorial. Since the argument is a limiting one as $n \to \infty$, I wonder if there is a more “probabilistic” argument (this is perhaps a bit fuzzy) for the results.

# estimating probability metrics from samples

I took a look at this interesting paper by Sriperumbudur et al., On the empirical estimation of integral probability metrics (Electronic Journal of Statistics Vol. 6 (2012) pp.1550–1599). The goal of the paper is to estimate a distance or divergence between two distributions $P$ and $Q$ based on samples from each distribution. This sounds pretty vague at first… what kind of distributions? How many samples? This paper looks at integral probability metrics, which have the form

$\gamma(P,Q) = \sup_{f \in \mathcal{F}} \left| \int_{S} f dP - \int_{S} f dQ \right|$

where $S$ is a measurable space on which $P$ and $Q$ are defined, and $\mathcal{F}$ is a class of real-valued bounded measurable functions on $S$. This class doesn’t contain Csiszár $\phi$-divergences (also known as Ali-Silvey distances), but does contain the total variational distance.

Different choices of the function class give rise to different measures of difference used in so-called two-sample tests, such as the Kolmogorov-Smirnov test. The challenge in practically using these tests is that it’s hard to get bounds on how fast an estimator of $\gamma(P,Q)$ converges if we have to estimate it from samples of $P$ and $Q$. The main result of the paper is to provide estimators with consistency and convergence guarantees. In particular, they estimators are based on either linear programming or (in the case of kernel tests) in closed form.

The second section of the paper connects tests based on IPMs with the risk associated to classification rules for separating $P$ and $Q$ when the separation rule is restricted to come from the function class $\mathcal{F}$ associated to the rule. This is a nice interpretation of these two-sample tests — they are actually doing similar things for restricted classes of classifiers/estimators.

Getting back to KL divergence and non-IPM measures, since total variation gives a lower bound on the KL divergence, they also provide lower bounds on the total variation distance in terms of other IPM metrics. This is important since the total variation distance can’t be estimated itself in a strongly consistent way. This could be useful for algorithms which need to estimate the total variation distance for continuous data. In general, estimating distances between multivariate continuous distributions can become a bit of a mess when you have to use real data — doing a plug-in estimate using, e.g. a kernel density estimator is not always the best way to go, and instead attacking the distance you want to measure directly could yield better results.

# Postdoc at INRIA-ENS Paris on Graphs, Algorithms and Probability

Applications are invited for a Postdoc position (full-time, up to 2 years) at INRIA-ENS in Paris. The position is funded by the ANR GAP grant “Graphs, Algorithms and Probability.”

Requirements are a PhD degree in Computer Science or Mathematics and a strong background in some of the following topics:

• discrete probability
• statistical learning
• combinatorial optimization
• stochastic networks

Applications must include a research statement, a CV and the names and contacts of references. All material should be sent by email to Marc Lelarge. Please indicate in the subject POSTDOC GAP.

Important dates:

• Intention of application (short email): as soon as possible
• Deadline for application: December 1st, 2013
• Suggested starting dates: Jan.-Feb. 2014