Dr. Guillaume Obozinski (Swiss Data Science Center, EPF Lausanne)
Convex unmixing and learning the effect of latent variables in Gaussian Graphical models with unobserved variables abstract
Abstract:
The edge structure of the graph defining an undirected graphical model describes precisely the structure of dependence between the variables in the graph. In many applications, the dependence structure is unknown and it is desirable to learn it from data, often because it is a preliminary step to be able to ascertain causal effects. This problem, known as structure learning, is hard in general, but for Gaussian graphical models it is slightly easier because the structure of the graph is given by the sparsity pattern of the precision matrix of the joint distribution, and because independence coincides with decorrelation. A major difficulty too often ignored in structure learning is the fact that if some variables are not observed, the marginal dependence graph over the observed variables will possibly be significantly more complex and no longer reflect the direct dependencies that are potentially associated with causal effects. In this work, we consider a family of latent variable Gaussian graphical models in which the graph of the joint distribution between observed and unobserved variables is sparse, and the unobserved variables are conditionally independent given the others. Prior work was able to recover the connectivity between observed variables, but could only identify the subspace spanned by unobserved variables, whereas we propose a convex optimization formulation based on structured matrix sparsity to estimate the complete connectivity of the complete graph including unobserved variables, given the knowledge of the number of missing variables, and a priori knowledge of their level of connectivity. Our formulation is supported by a theoretical result of identifiability of the latent dependence structure for sparse graphs in the infinite data limit, which is a particular instance of a more general result we prove for unmixing with convex norms. We propose an algorithm leveraging recent active set methods, which performs well in the experiments on synthetic data.
• EPF Lausanne, Salle CM 1 113
Ron Kaniel (Simon School of Business, University of Rochester)
Relative Pay for Non-Relative Performance: Keeping up with the Joneses with Optimal Contracts abstract
Abstract:
We consider multi-agent contracting when agents have "keeping up with the Joneses" (KUJ) preferences. Although contracts appear inefficient, as performance seems inadequately benchmarked, when a single principal can commit to a public contract, the optimal contract hedges agents’ relative wage risk without sacrificing efficiency. With multiple principals, or a principal that is unable to commit, a "rat race" emerges in which agents are more productive, but wages increase even more, reducing profits and undermining efficiency. Renegotiation reduces or enhances efficiency depending on whether KUJ effects are weak or strong. Public disclosure of contracts across firms can cause output to collapse.
10:30 • EPF Lausanne, Quartier UNIL - Dorigny, Salle Extranef 126
Chris Stolk (University of Amsterdam)
Recent methods for solving the high-frequency Helmholtz equation on a regular mesh abstract
Abstract:
In this talk I discuss a recently developed finite difference discretisation of the Helmholtz equation and some solution methods for the resulting linear systems. In high-frequency Helmholtz problems, pollution errors, due to numerical dispersion, are a main source of error. We will show that such errors can be strongly reduced compared to other schemes, including high-order finite elements, by selecting coefficients for the discrete system that maximise the accuracy of geometrical optics phases and amplitudes. Such low dispersion schemes are of interest by themselves, but can also be used to improve the efficiency of multigrid schemes. Computation times for a solver combining a multigrid method with domain decomposition compare well to those of alternative methods.
11:00 • Universität Basel, Mathematisches Institut, Spiegelgasse 5, 4051 Basel, Raum 05.001
Prof. Dr. Raf Cluckers (Université de Lille)
Igusa\'s conjecture on exponential sums abstract
Abstract:
I will present the recent progress with M. Mustata and K. H. Nguyen on Igusa\'s conjecture on exponential sums, which is about bounds for finite exponential sums with index sets consisting of integers modulo positive powers of primes, and where uniformity is key. The presented proof relies on recent results from the minimal model program in algebraic geometry, and is optimal in the case of non-rational singularities. Igusa intended to use his conjecture on exponential sums to advance the study of local-global principles, but these ideas remain mostly unachieved up to date.
14:15 • ETH Zentrum, Rämistrasse 101, Zürich, Building HG, Room G 43
Dr. Federico Stra (EPF Lausanne)
Linear Lipschitz and C^1 Extension Operators through Random Projection abstract
Abstract:
I present the construction of a regular random projection of a metric space onto a closed doubling subset and use it to linearly extend Lipschitz functions. This tool provides a way to prove more directly a result by Lee and Naor and to generalize the classical extension theorem by Whitney to Banach spaces.
14:15 • EPF Lausanne, Salle MA B1 11
Paulo Rodrigues (Bank of Portugal)
Testing for Episodic Predictability in Stock Returns abstract
Abstract:
Standard tests based on predictive regressions estimated over the full available sample data have tended to nd little evidence of predictability in stock returns. Recent approaches based on the analysis of subsamples of the data have been considered, suggesting that predictability where it occurs might exist only within so-called \\pockets of predictability" rather than across the entire sample. However, these methods are prone to the criticism that the sub-sample dates are endogenously determined such that the use of standard critical values appropriate for full sample tests will result in incorrectly sized tests leading to spurious ndings of stock returns predictability. To avoid the problem of endogenously-determined sample splits, we propose new tests derived from sequences of predictability statistics systematically calculated over sub-samples of the data. Specically, we will base tests on the maximum of such statistics from sequences of forward and backward recursive, rolling, and double-recursive predictive sub-sample regressions. We develop our approach using the over-identied instrumental variable-based predictability test statistics of Breitung and Demetrescu (2015). This approach is based on partial-sum asymptotics and so, unlike many other popular approaches including, for example, those based on Bonferroni corrections, can be readily adapted to implementation over sequences of subsamples. We show that the limiting distributions of our proposed tests are robust to both the degree of persistence and endogeneity of the regressors in the predictive regression, but not to any heteroskedasticity present even if the sub-sample statistics are based on heteroskedasticity-robust standard errors. We therefore develop xed regressor wild bootstrap implementations of the tests which we demonstrate to be rst-order asymptotically valid. Finite sample behaviour against a variety of temporarily predictable processes is considered. An empirical application to US stock returns illustrates the usefulness of the new predictability testing methods we propose.
15:15 • ETH Zentrum, Rämistrasse 101, Zürich, Building HG, Room G 26.3
Tim Buelles (ETH Zürich)
Gromov-Witten theory of K3 surfaces and tautological classes
16:00 • ETH Zentrum, Rämistrasse 101, Zürich, Building HG, Room G 43
Prof. Dr. Christian Cachin (Universität Bern)
Abstract:
Practical Byzantine consensus protocols have been available for decades, but they were never deployed in production until very recently. The widespread interest in crypto-currencies and blockchains has changed this. Blockchains offer resilience and security based on the collective trust placed in the nodes maintaining it. This presentation will discuss protocols for Byzantine consensus and explore older and newer protocols that power blockchains. The talk also introduces Hyperledger Fabric, a modular and extensible blockchain platform that is developed open-source under the Hyperledger Project.
16:15 • Universität Bern, Alpeneggstrasse 22, 3012 Bern, Hörsaal 203
Ingrid Irmer (Technion, Haifa)
Stable commutation length in 3-manifolds abstract
Abstract:
Commutator length is a group theoretical analogue of genus. By taking a limit, stable commutator length , scl, is obtained. This is a group invariant for studying metric properties of bounded group cohomology, and detects the existence of surface subgroups. There are open questions regarding computability of scl and the structure of its unit norm ball. This talk will give some background on scl in low dimensional topology, and will outline work in progress of the speaker towards resolving these questions for closed surface and 3-manifold groups.
16:15 • Université de Genève, Section Mathématiques, Centre Acacias II, Salle 17
Thomas Frédéric Hilke (EPF Lausanne)
Méthodes numériques liées à la distribution d\'alumine dans une cuve d\'électrolyse d\'aluminium
17:00 • EPF Lausanne, Salle CM 1 5