Wenyu Wang (Kelley School of Business, Indiana University)
The Incentives of SPAC Sponsors abstract
Abstract:
The market of Special Purpose Acquisition Companies (SPACs) has exploded in recent years, yet its volatile performance calls into question the implications of this unique business model and particularly the incentives of the SPAC sponsors on the welfare of SPAC shareholders. This paper quantitatively studies these questions by estimating a model featuring the strategic interactions between SPAC sponsors, targets, and investors. The estimation uses a comprehensive hand-collected dataset of SPACs that completed acquisitions between 2009 and April 2022 with rich information such as sponsor concessions, earnouts, redemptions, etc. Agency costs appear pervasive: the inter-quintile range of returns to non-redeeming shareholders reaches 19% in deals sorted by their agency conflicts. Average SPAC investors make sizeable mistakes in inferring deal quality, leading them to earn a 7% lower return. Tying more of the sponsor\'s promote shares to earnouts significantly reduces the agency cost and improves investors\' expected return, while cutting back the issuance of warrants has a limited impact on the average SPAC investors\' welfare.
10:30 • EPF Lausanne, UniL campus, Extranef 126
Christophe Geuzaine (Université de Liège)
Non-reflecting boundary conditions and domain decomposition methods for industrial flow acoustics abstract
Abstract:
This talk is devoted to non-overlapping Schwarz domain decomposition methods for the resolution of high frequency flow acoustics problems of industrial relevance. First, we will present recent advances on non-reflecting boundary techniques that provide local approximations to the Dirichlet-to-Neumann operator for convected and heterogeneous time-harmonic wave propagation problems [1]. Then we will show how to adapt a generic domain decomposition framework to flow acoustics, based on these newly designed transmission conditions, and highlight the benefit of the approach on the simulation of three-dimensional noise radiation of a high by-pass ratio turbofan engine intake [2].
[1] Marchner, P., Antoine, X., Geuzaine, C., & Bériot, H. (2022). Construction and numerical assessment of local absorbing boundary conditions for heterogeneous time-harmonic acoustic problems. SIAM Journal on Applied Mathematics, 82(2), 476-501.
[2] Lieu, A., Marchner, P., Gabard, G., Beriot, H., Antoine, X., & Geuzaine, C. (2020). A non-overlapping Schwarz domain decomposition method with high-order finite elements for flow acoustics. Computer Methods in Applied Mechanics and Engineering, 369, 113223.
11:00 • Universität Basel, Spiegelgasse 5, room 05.001
Sergey Dolgov (University of Bath)
Tensor train approximation of deep transport maps for Bayesian inverse problems abstract
Abstract:
Characterising intractable high-dimensional random variables is one of the fundamental challenges in stochastic computation. We develop a deep transport map that is suitable for sampling concentrated distributions defined by an unnormalised density function. We approximate the target distribution as the pushforward of a reference distribution under a composition of order-preserving transformations, in which each transformation is formed by a tensor train decomposition. The use of composition of maps moving along a sequence of bridging densities alleviates the difficulty of directly approximating concentrated density functions. We propose two bridging strategies suitable for wide use: tempering the target density with a sequence of increasing powers, and smoothing of an indicator function with a sequence of sigmoids of increasing scales. The latter strategy opens the door to efficient computation of rare event probabilities in Bayesian inference problems. Numerical experiments on problems constrained by differential equations show little to no increase in the computational complexity with the event probability going to zero, and allow to compute hitherto unattainable estimates of rare event probabilities for complex, high-dimensional posterior densities.
14:00 • Université de Genève, Conseil Général 7-9, Room 1-15
Weijie Su (Wharton, University of Pennsylvania)
Some Geometric Patterns of Real-World Deep Neural Networks abstract
Abstract:
In this talk, we will investigate the emergence of geometric patterns in well-trained deep learning models by making use of the layer-peeled model and the law of equi-separation. The former is a nonconvex optimization program that models the last-layer features and weights. We use the model to shed light on the neural collapse phenomenon of Papyan, Han, and Donoho, and to predict a hitherto-unknown phenomenon that we term minority collapse in imbalanced training. This is based on joint work with Cong Fang, Hangfeng He, and Qi Long (arXiv:2101.12699).In the second part, we study how real-world deep neural networks process data in the interior layers. Our finding is a simple and quantitative law that governs how deep neural networks separate data according to class membership throughout all layers for classification. This law shows that each layer improves data separation at a constant geometric rate, and its emergence is observed in an authoritative collection of network architectures and datasets during training. This law offers practical guidelines for designing architectures, improving model robustness and out-of-sample performance, as well as interpreting the predictions. This is based on joint work with Hangfeng He (arXiv:2210.17020).
15:15 • ETH Zentrum, Rämistrasse 101, Zürich, Building HG, Room G 19.1
Sara Salvador (Universität Bern)
Abstract:
This thesis deals with directional distributions, hence distributions defined over Rp. We will consider in details the case of circular distributions, i.e. directional distributions in R2. We will focus on the study of an extension of the von Mises distribution (vM), namely the generalized von Mises distribution of order two (GvM2, or simply GvM). This distribution allows higher flexibility in terms of asymmetry and bimodality than the vM distribution. Two Bayesian tests are computed on the GvM. The first test concerns the symmetry of the GvM. Inference on symmetry is made via Bayes factors. A real circular data case is considered. The second test concerns the bimodality of the GvM. The problem is reduced to the study of the real roots of a quartic whose coefficients depend on the parameters of the model. A detailed analysis of the quartic is given and a region W of parameters that are associated to bimodality is obtained. Then, inference on bimodality is made via Bayes factors and via highest posterior density (HPD) credible sets. We use Markov Chain Monte Carlo methods (MCMC) to compute posterior probabilities of bimodality. Conclusions confirm that, when data are in accordance with the null hypotheses, Bayes factors are typically large. On the other hand, the HPD credible set is entirely contained inside W and bimodality is confirmed.Moreover, in this thesis we consider directional distributions over the unit sphere Sp−1 of Rp. These are called spherical distributions and here we manly focus on the generalized von Mises-Fisher (GvMF) distribution. This is an extension of the von Mises- Fisher distribution (vMF). We will give two methods to generate random variables from the GvMF using a conditional acceptance-rejection method and the Metropolis Hastings algorithm.
16:00 • Universität Bern, ExWi, Sidlerstrasse 5, 3012 Bern, Hörsaal B7