Showing papers by "Richard Cole published in 2016"

Convex Program Duality, Fisher Markets, and Nash Social Welfare

Abstract: We study Fisher markets and the problem of maximizing the Nash social welfare (NSW), and show several closely related new results. In particular, we obtain: -- A new integer program for the NSW maximization problem whose fractional relaxation has a bounded integrality gap. In contrast, the natural integer program has an unbounded integrality gap. -- An improved, and tight, factor 2 analysis of the algorithm of [7]; in turn showing that the integrality gap of the above relaxation is at most 2. The approximation factor shown by [7] was $2e^{1/e} \approx 2.89$. -- A lower bound of $e^{1/e}\approx 1.44$ on the integrality gap of this relaxation. -- New convex programs for natural generalizations of linear Fisher markets and proofs that these markets admit rational equilibria. These results were obtained by establishing connections between previously known disparate results, and they help uncover their mathematical underpinnings. We show a formal connection between the convex programs of Eisenberg and Gale and that of Shmyrev, namely that their duals are equivalent up to a change of variables. Both programs capture equilibria of linear Fisher markets. By adding suitable constraints to Shmyrev's program, we obtain a convex program that captures equilibria of the spending-restricted market model defined by [7] in the context of the NSW maximization problem. Further, adding certain integral constraints to this program we get the integer program for the NSW mentioned above. The basic tool we use is convex programming duality. In the special case of convex programs with linear constraints (but convex objectives), we show a particularly simple way of obtaining dual programs, putting it almost at par with linear program duality. This simple way of finding duals has been used subsequently for many other applications.

Large Market Games with Near Optimal Efficiency

Abstract: As is well known, many classes of markets have efficient equilibria, but this depends on agents being non-strategic, i.e. that they declare their true demands when offered goods at particular prices, or in other words, that they are price-takers. An important question is how much the equilibria degrade in the face of strategic behavior, i.e. what is the Price of Anarchy (PoA) of the market viewed as a mechanism? Often, PoA bounds are modest constants such as 4/3 or 2. Nonetheless, in practice a guarantee that no more than 25% or 50% of the economic value is lost may be unappealing. This paper asks whether significantly better bounds are possible under plausible assumptions. In particular, we look at how these worst case guarantees improve in the following large settings. -Large Walrasian auctions: These are auctions with many copies of each item and many agents. We show that the PoA tends to 1 as the market size increases, under suitable conditions, mainly that there is some uncertainty about the numbers of copies of each good and demands obey the gross substitutes condition. We also note that some such assumption is unavoidable. -Large Fisher markets: Fisher markets are a class of economies that has received considerable attention in the computer science literature. A large market is one in which at equilibrium, each buyer makes only a small fraction of the total purchases; the smaller the fraction, the larger the market. Here the main condition is that demands are based on homogeneous monotone utility functions that satisfy the gross substitutes condition. Again, the PoA tends to 1 as the market size increases. Furthermore, in each setting, we quantify the tradeoff between market size and the PoA.

A Unified Approach to Analyzing Asynchronous Coordinate Descent and Tatonnement

Abstract: This paper concerns asynchrony in iterative processes, focusing on gradient descent and tatonnement, a fundamental price dynamic. Gradient descent is an important class of iterative algorithms for minimizing convex functions. Classically, gradient descent has been a sequential and synchronous process, although distributed and asynchronous variants have been studied since the 1980s. Coordinate descent is a commonly studied version of gradient descent. In this paper, we focus on asynchronous coordinate descent on convex functions $F:\mathbb{R}^n\rightarrow\mathbb{R}$ of the form $F(x) = f(x) + \sum_{k=1}^n \Psi_k(x_k)$, where $f:\mathbb{R}^n\rightarrow\mathbb{R}$ is a smooth convex function, and each $\Psi_k:\mathbb{R}\rightarrow\mathbb{R}$ is a univariate and possibly non-smooth convex function. Such functions occur in many data analysis and machine learning problems. We give new analyses of cyclic coordinate descent, a parallel asynchronous stochastic coordinate descent, and a rather general worst-case parallel asynchronous coordinate descent. For all of these, we either obtain sharply improved bounds, or provide the first analyses. Our analyses all use a common amortized framework. The application of this framework to the asynchronous stochastic version requires some new ideas, for it is not obvious how to ensure a uniform distribution where it is needed in the face of asynchronous actions that may undo uniformity. We believe that our approach may well be applicable to the analysis of other iterative asynchronous stochastic processes. We extend the framework to show that an asynchronous version of tatonnement, a fundamental price dynamic widely studied in general equilibrium theory, converges toward a market equilibrium for Fisher markets with CES utilities or Leontief utilities, for which tatonnement is equivalent to coordinate descent.