July 2018 - Talks - lists.usi.ch

Monday 6 August, about 10:00 "Prediction of singular VARs and application to generalized dynamic factor models"

by Announcements of talks＠IDSIA

Title: Prediction of singular VARs and application to generalized dynamic factor models Abstract : Vector autoregressive processes (VARs) with innovations having a singular covariance matrix (in short singular VARs) appear naturally in the context of dynamic factor models. The Yule-Walker estimator of such a VAR is problematic, because the solution of the corresponding equation system tends to be numerically rather unstable. For example, if we overestimate the order of the VAR, then the singularity of the innovations renders the Yule-Walker equation system singular as well. Moreover, even with correctly selected order, the Yule-Walker system tends be close to singular in finite sample. In this paper we are going to show that this has a severe impact on predictions. While the asymptotic rate of the mean square prediction error (MSPE) can be just like in the regular (non-singular) case, the finite sample behaviour is suffering. This effect will turn out to be particularly dramatic in context of dynamic factor models, where we do not directly observe the so-called common components which we aim to predict. Then, when the data are sampled with some additional error, the MSPE often gets severely inflated. We will explain the reason for this phenomenon and show how to overcome the problem. Our numerical results underline that it is very important to adapt prediction algorithms accordingly. Speaker Gilles Nisol, post-doc researcher at ULB (Universite' Libre de Bruxelles) Location Idsia's Meeting Room

6 years, 4 months

1
0
0 / 0

Monday 30 July, 11:30 - CANCELLED

by Announcements of talks＠IDSIA

Dear All, today's talk: Keeping it simple: minimal time series models for critical transitions in complex systems is cancelled. best regards, Giorgio

6 years, 4 months

1
0
0 / 0

Monday 30 July, 11:30 - "Minimal time series models for critical transitions in complex systems"

by Announcements of talks＠IDSIA

::Speaker Flavio Pons ::Title Keeping it simple: minimal time series models for critical transitions in complex systems ::Abstract Many dynamical systems feature bistability, i.e. they are characterized by a double-well potential, where the minima correspond to stable equilibrium points, and the interposed maximum to an unstable equilibrium. Generally, a system in its stable states exhibits a fairly regular behavior, which can be effectively described by simple stochastic models. On the other hand, the passage from one stable point to the other is fast, sudden, and characterized by abrupt changes in the dynamics. Examples can be found both in systems with few degrees of freedom, such as the Lorenz63 model, and in complex natural phaenomena, such as protein folding, turbulence, or the passage from ice ages to non-glacial regimes. Even though the jumps between stable states are generally thought to be governed by the slow variables in the system, growing evidence suggests that an appropriate statistical description of the fast variables enables defining early warning indicators of the upcoming critical transition. The aim of this talk is to show that it is possible to build effective early warning indicators through simple, linear time series models. In particular, two methods are presented, both based on exploiting the misspecification affecting the stochastic model postulated for the stable states, as the system approaches the critical transition. *Bio: Flavio Pons holds a PhD in Statistics from University of Bologna, on the topic: "Statistical modelling of turbulence in a close von Kàrman flow". *Location: IDSIA's meeting room =

6 years, 5 months

1
0
0 / 0

Reminder: PIDSIA Seminar by M. Byshkin, this Wednesday (25.7)

by Announcements of talks＠IDSIA

Reminder of PIDSIA seminar this Wednesday at 12h. ********************* *Speaker: Maksym Byshkin *Title: Fast Maximum Likelihood estimation via Equilibrium Expectation for Large Network Data *Abstract: Network data may be analyzed by constructing statistical models that accurately reproduce the data features that may be of theoretical relevance or empirical interest. Examples of such models are Exponential Random Graph Models for social networks and Markov Random Field for image processing. Typically, Markov chain Monte Carlo (MCMC) methods are used when normalizing constants of statistical models cannot be computed. We propose a new approach for the Maximum Likelihood Estimation (MLE) of parameters of statistical models with intractable normalizing constants. The approach we propose relies on properties of Markov chains at equilibrium and, for this reason, we call it Equilibrium Expectation (EE). Using this approach we design a simple and fast algorithm to find MLE when it exists and is unique. Good scaling properties of the algorithm allow dramatic increase of the size of dependent/network data for which MLE may be computed. Application of the algorithm to large size images, biological and social networks is presented. Application to Restricted Boltzmann machines and similar data-generating models is discussed. *Bio: Maksym Byshkin is a postdoctoral researcher at Università della Svizzera italiana (USI), where he serves also as a lectures and teaches Data Science. His research fields are statistical physics, statistical network analysis, Monte Carlo methods, computational chemistry and interdisciplinary collaborations. The research activity and research interests are focused on developments of empirical models and computational methods for high performance computing. Maksym holds a Master degree in applied mathematics and a Doctoral degree in theoretical physics from the Kharkov Institute of Physics and Technology (Landau school). Before joining USI he spent several years as a postdoctoral research fellow at the Free University of Brussels and the University of Salerno. *When: Wednesday 25th of July 2018, 12:00-13:00 *Location: Manno, Galleria 1, 2nd floor, room G1-204 *Registration: Pizza (or alternative food) and drinks will be offered at the end of the talk. If you plan to attend, please register in a timely fashion at the following link so that we will have no shortage of food: https://doodle.com/poll/vy5cxiibugeznwuw

6 years, 5 months

1
0
0 / 0

Marco Forgione, Tue 24 July, 10:00-10:30

by Announcements of talks＠IDSIA

::Speaker Marco Forgione ::Title Data-driven model Improvement for Model-based Control ::Abstract In this talk, I present a framework for the gradual improvement of model-based control for linear systems operating in closed-loop. The total time of closed-loop operation is divided into a number of consecutive learning intervals. After an interval is executed, the model is refined based on the measured data using system identification techniques. This model is used to synthesize the controller applied during the next interval. Excitation signals can be injected into the control loop during each of the learning intervals. On the one hand, the presence of an excitation signal worsens the control performance in the current interval since it acts as a disturbance. On the other hand, the more informative data generated with the excitation signal applied lead to a more accurate model after re-identification. Therefore, the control performance for the next intervals is likely to improve. In principle, the excitation signals should be designed to maximize the overall control performance taking this dual effect explicitly into account. However, this is in an intractable stochastic optimization problem. For this reason, a convex approximation of the original problem is derived. The approximated problem can be solved efficiently using standard optimization software. The applicability of the method is demonstrated in a simulation study and preliminary extensions to nonlinear systems are discussed in the final part of the presentation. *Bio: Marco Forgione is a researcher in the Control and Automation field. He currently works in the industrial sector. He holds a M.Sc. degree in Computer Engineering from the University of Pavia, Italy. Postdoctoral Researcher at the Ecole Centrale de Lyon, *Location: IDSIA's meeting room

6 years, 5 months

1
0
0 / 0

Reminder: Talk at 11 @ Aula 222

by Announcements of talks＠IDSIA

*Speakers: Susanne Bleisch and Stephan Nebiker *Title: FHNW Geomatics Engineering overview and projects FHNW and IDSIA are currently collaborating on an SNF project: video analytics for crisis management. Susanne and Stephan are going to present us the Geomatics Institute of FHNW (http://www.fhnw.ch/igeo), as well as the ongoing projects in their groups. They are also interested in getting to know people from IDSIA and the projects we are working on. The presentation, together with the follow-up discussion, should take up to 45mins. *Bio: Susanne Bleisch is a professor for Geovisualization and Visual Analytics at FHNW. http://sbleis.ch/ Stephan Nebiker is a professor at the same institute, working on Photogammetry, Remote Sensing, and Geoinformatics. https://www.fhnw.ch/de/personen/stephan-nebiker *When: Thursday 19th of July 2018, 11:00-12:00 *Location: Manno, Galleria 1, 2nd floor, Aula 222

6 years, 5 months

1
0
0 / 0

Susanne Bleisch and Stephan Nebiker, Thu. 19th of July 2018, 11:00-12:00

by Announcements of talks＠IDSIA

*Speakers: Susanne Bleisch and Stephan Nebiker *Title: FHNW Geomatics Engineering overview and projects FHNW and IDSIA are currently collaborating on an SNF project: video analytics for crisis management. Susanne and Stephan are going to present us the Geomatics Institute of FHNW (http://www.fhnw.ch/igeo), as well as the ongoing projects in their groups. They are also interested in getting to know people from IDSIA and the projects we are working on. The presentation, together with the follow-up discussion, should take up to 45mins. *Bio: Susanne Bleisch is a professor for Geovisualization and Visual Analytics at FHNW. http://sbleis.ch/ Stephan Nebiker is a professor at the same institute, working on Photogammetry, Remote Sensing, and Geoinformatics. https://www.fhnw.ch/de/personen/stephan-nebiker *When: Thursday 19th of July 2018, 11:00-12:00 *Location: Manno, Galleria 1, 2nd floor, Aula 222

6 years, 5 months

1
0
0 / 0

PIDSIA Seminar by Maksym Byshkin, Wed. 25th of July 2018, 12:00-13:00

by Announcements of talks＠IDSIA

*Speaker: Maksym Byshkin *Title: Fast Maximum Likelihood estimation via Equilibrium Expectation for Large Network Data *Abstract: Network data may be analyzed by constructing statistical models that accurately reproduce the data features that may be of theoretical relevance or empirical interest. Examples of such models are Exponential Random Graph Models for social networks and Markov Random Field for image processing. Typically, Markov chain Monte Carlo (MCMC) methods are used when normalizing constants of statistical models cannot be computed. We propose a new approach for the Maximum Likelihood Estimation (MLE) of parameters of statistical models with intractable normalizing constants. The approach we propose relies on properties of Markov chains at equilibrium and, for this reason, we call it Equilibrium Expectation (EE). Using this approach we design a simple and fast algorithm to find MLE when it exists and is unique. Good scaling properties of the algorithm allow dramatic increase of the size of dependent/network data for which MLE may be computed. Application of the algorithm to large size images, biological and social networks is presented. Application to Restricted Boltzmann machines and similar data-generating models is discussed. *Bio: Maksym Byshkin is a postdoctoral researcher at Università della Svizzera italiana (USI), where he serves also as a lectures and teaches Data Science. His research fields are statistical physics, statistical network analysis, Monte Carlo methods, computational chemistry and interdisciplinary collaborations. The research activity and research interests are focused on developments of empirical models and computational methods for high performance computing. Maksym holds a Master degree in applied mathematics and a Doctoral degree in theoretical physics from the Kharkov Institute of Physics and Technology (Landau school). Before joining USI he spent several years as a postdoctoral research fellow at the Free University of Brussels and the University of Salerno. *When: Wednesday 25th of July 2018, 12:00-13:00 *Location: Manno, Galleria 1, 2nd floor, room G1-204 *Registration: Pizza (or alternative food) and drinks will be offered at the end of the talk. If you plan to attend, please register in a timely fashion at the following link so that we will have no shortage of food: https://doodle.com/poll/vy5cxiibugeznwuw

6 years, 5 months

1
0
0 / 0

PIDSIA Seminar by Seeun William Umboh, Tuesday 10th July

by Announcements of talks＠IDSIA

Speaker: Seeun William Umboh Title: Nested Convex Bodies are Chaseable Abstract: In the Convex Body Chasing problem, we are given an initial point v_0 in R^d and an online sequence of n convex bodies F_1 , … , F_n. When we receive F_i, we are required to move inside F_i. Our goal is to minimize the total distance travelled. This fundamental online problem was first studied by Friedman and Linial (DCG 1993). They proved an Omega(sqrt{d}) lower bound on the competitive ratio, and conjectured that a competitive ratio depending only on d is possible. However, despite much interest in the problem, the conjecture remains wide open. We consider the setting in which the convex bodies are nested: F_i contains F_{i+1}. The nested setting is closely related to extending the online LP framework of Buchbinder and Naor (ESA 2005) to arbitrary linear constraints. Moreover, this setting retains much of the difficulty of the general setting and captures an essential obstacle in resolving Friedman and Linial's conjecture. In this work, we give the first f(d)-competitive algorithm for chasing nested convex bodies in R^d. Bio: William Umboh is a Lecturer in Algorithms at the University of Sydney. He received his Ph.D. from the University of Wisconsin-Madison in 2015 and subsequently was a postdoc at the Eindhoven University of Technology during 2015-2018. His main research interest is in designing approximation algorithms and online algorithms for combinatorial optimization problems. When: Tuesday 10th of July 2018, 12:00-13:00 Location: Manno, Galleria 1, 2nd floor, room G1-204 Registration: Pizza (or alternative food) and drinks will be offered at the end of the talk. If you plan to attend, please register in a timely fashion at the following link so that we will have no shortage of food: https://doodle.com/poll/5y8d232ykqsdcwus

6 years, 5 months

1
0
0 / 0

PIDSIA Seminar by Seeun William Umboh, 10th July

by Announcements of talks＠IDSIA

Speaker: Seeun William Umboh Title: Nested Convex Bodies are Chaseable Abstract: In the Convex Body Chasing problem, we are given an initial point v_0 in R^d and an online sequence of n convex bodies F_1 , … , F_n. When we receive F_i, we are required to move inside F_i. Our goal is to minimize the total distance travelled. This fundamental online problem was first studied by Friedman and Linial (DCG 1993). They proved an Omega(sqrt{d}) lower bound on the competitive ratio, and conjectured that a competitive ratio depending only on d is possible. However, despite much interest in the problem, the conjecture remains wide open. We consider the setting in which the convex bodies are nested: F_i contains F_{i+1}. The nested setting is closely related to extending the online LP framework of Buchbinder and Naor (ESA 2005) to arbitrary linear constraints. Moreover, this setting retains much of the difficulty of the general setting and captures an essential obstacle in resolving Friedman and Linial's conjecture. In this work, we give the first f(d)-competitive algorithm for chasing nested convex bodies in R^d. Bio: William Umboh is a Lecturer in Algorithms at the University of Sydney. He received his Ph.D. from the University of Wisconsin-Madison in 2015 and subsequently was a postdoc at the Eindhoven University of Technology during 2015-2018. His main research interest is in designing approximation algorithms and online algorithms for combinatorial optimization problems. When: Tuesday 10th of July 2018, 12:00-13:00 Location: Manno, Galleria 1, 2nd floor, room G1-204 Registration: Pizza (or alternative food) and drinks will be offered at the end of the talk. If you plan to attend, please register in a timely fashion at the following link so that we will have no shortage of food: https://doodle.com/poll/5y8d232ykqsdcwus

6 years, 5 months

1
0
0 / 0

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

Talks July 2018