All Years Seminars
[INMA] 2020-11-03 (14h) : The quest to schedule event-based controllers. An approach via formal abstractions.
At MS Teams
Speaker :
Manuel Mazo Jr. (TU Delft)
Abstract :
Modern control systems are often implemented as networked systems, in which bandwidth of a communication medium is shared among a number of applications. Furthermore, often the act of communicating also has an energy impact on the system infrastructure, e.g. in the case of wireless control systems. In the last decade the study of controller implementation strategies aiming at reducing the communication footprint, to save bandwidth (and energy), has thus attracted great attention. Arguably, one of the most promising implementation paradigms in the literature is that of event-triggered control (ETC), in which communications are triggered by the state of the plant itself, which makes the system communicate only when necessary to retain stability of the closed-loop.
ETC implementations, while reducing drastically the amount of required bandwidth, result in aperiodic transmissions whose time of occurrence are highly unpredictable, leading to difficulties for scheduling such systems. Such difficulties usually imply that the freed bandwidth may not be reusable by other (real-time) applications, and that energy savings may be completely lost, thus rendering ETC schemes inefficient.
In this talk I describe a plan to the ETC scheduling problem through the construction of timed-automata models for the ETC systems' traffic. I start introducing ETC and describing the difficulty of scheduling such systems employing a control theoretic perspective. Next, I introduce the notion of model abstraction we employ to make the scheduling problem solvable. Two approaches to the construction of this abstractions are described and how the resulting models can be employed to solve scheduling problems. The talk finalizes going briefly over our most recent advances, the difficulties, and possible opportunities for future research in this field.
[INMA] 2020-10-27 (14:00) : Welcome Seminar for new PhD students (#1)
At MS Teams
Section 1:Controlling COVID-19: Labor structure more important than lockdown policy
Speaker :
Isabella Loaiza
Abstract : The government response to the COVID-19 pandemic in Colombia has been characterised by the closure of international borders and strict lockdown measures. In spite of a dictated nationwide lockdown, the implementation of this lockdown has been administered by municipal governments. Thus, lockdown measures vary across municipalities. Different measures allow residents to go out for essential shopping based on the last digit of their national ID card, their gender or a mix of both. This creates an opportunity to gain insight into the effects of different lockdown policies on urban mobility via this natural experiment.
We investigate the extent to which different policies have affected residents' mobility and compare pre and post lockdown disruption of commute patterns. While we find that residents of municipalities with more severe policies had, on average, greater reductions in trip frequency during lockdown, the effect of policy severity is only about half that of city size - residents of larger cities. We also find that policy severity had no bearing on the extent to which residents reduced their daily travel distance during lockdown - nor on the likelihood that residents abandoned their established commuting patterns. However, we do find that municipalities with higher GDP, larger populations, and higher labor formality have experienced disproportionately greater reductions in mobility and greater reductions in commuting patterns. Interestingly, this difference in mobility reduction across cities of varying size/wealth is more pronounced than the difference within cities based on neighborhood socioeconomic stratum. To validate our results derived from CDR data we use google mobility data and find strong correlation with our original results.
Section 2:Multisensory Integration in Dynamic Environments
Speaker :
Hoffmann, Anne
Abstract : In my project I am investigating how the brain integrates information from different sensory modalities in a dynamic environment using a combination of reaching experiments and computational modeling. Human motor proficiency results from an intricate interaction between sensory and motor systems in the brain. The nervous system plans and controls movements using vision, proprioception, and tactile sensors. Being able to combine sensory information about the body and the environment in real-time is essential for executing fast, adaptable, and accurate movements in the presence of uncertainty and external disturbances. Today, most of our knowledge about multisensory integration is based on experiments in which the sensory input is constant over time. These static cue-combination tasks have shown that humans optimally integrate information from different sensory modalities to form a coherent percept of their body and the environment. However, when we move, the position of our body is constantly changing and sensory organs provide the brain with disparate and asynchronous information. How may the human brain resolve these computational challenges? To address this important question, I am conducting a series of psychophysical experiments on healthy adults in which they perform arm movements while encountering various visual and force perturbations. My empirical results will be interpreted in the context of Optimal Feedback Control theory. Developing a computational explanation of dynamic multisensory integration will have important implications for our fundamental understanding of the sensorimotor system, as well as for clinical applications. Indeed, most of our everyday behavior requires coordinated motor actions and deficits in multisensory integration may lead to biased or unreliable perception of body and environment, which ultimately impairs motor performance.
Section 3: Automatic Performance Estimation for Decentralized Optimization
Speaker :
Colla, Sébastien
Abstract : Decentralized optimization consists in having interconnected computers optimizing a function f collaboratively, by iteratively updating estimates of the optimal solution. It is particularly relevant in machine learning applications where f depends on datasets too large to be stored on a single computer, or in sensor networks, multi-robot and micro-grid applications.
We take the view that progress in this promising field is hindered by the high complexity of analyzing these decentralized algorithms, due to the interplay between the effects of optimization and of the interconnection network; Most performance bounds are indeed complex and conservative, making it difficult to compare algorithms and understand the reasons limiting their performance.
I will create a mathematical formulation allowing to automatically compute the exact performance of any decentralized optimization algorithm and identify the bottleneck instances, providing insight on what limits the performance. This will allow for rapid exploration of ideas and iterative improvement and help develop a better understanding of the decentralized optimization algorithms. I intend to achieve this by extending to the decentralized case the Performance Estimation (PEP) approach. It relies on formulating the evaluation of an algorithm worst-case performance as an optimization problem itself, whose variables are the objective functions and initial conditions. Such problems are very complex, but have been shown to be solvable exactly. One of my main challenges will be to embed description of classes of interconnection networks in the PEP formulations.
[INMA] 2020-10-20 (14:00) : Welcome Seminar for new postdocs (#1)
At MS Teams
Section 1:Oracle Complexity Separation in Convex Optimization
Speaker :
Evgeniia Vorontsova
Abstract : Empirical risk minimization problems in machine learning are considered. Such kind of problems are often consisted of several blocks that can be treated
with different types of oracles, e.g., full gradients, stochastic gradients or coordinate derivatives. Optimal oracle complexity is known and achievable
separately for each type of the oracles. We propose a generic framework to combine several optimal algorithms for different types of oracles in order to
achieve separate optimal oracle complexity for every block. As a particular example, we demonstrate that in a case of full and stochastic gradient oracles' combination
our approach leads to the optimal number of oracle calls separately for the full gradient part and the stochastic part.
Section 2:Accelerated first-order methods for large-scale convex optimization: nearly optimal complexity under strong convexity
Speaker :
Masoud Ahookhosh
Abstract : In this study, four accelerated (sub)gradient algorithms (ASGA) are introduced and analyzed for solving several classes of convex optimization problems with a large number of variables. The objective function is the sum of two convex functions in which the first one is weakly smooth (either nonsmooth with the bounded variation of subgradients or smooth with Hölder continuous gradients). Specifically, two novel estimation sequences are presented to majorize the objective function, and accordingly two iterative schemes are developed with respect to these upper estimations. In both cases, the first scheme requires the smoothness parameter and a Hölder constant, while the second scheme is fully parameter-free (except for the strong convexity parameter which we set zero if it is not available) at the price of applying a finitely terminated backtracking line search. Notably, the proposed algorithms attain the optimal complexity for smooth problems with Lipschitz continuous gradients, nonsmooth problems with the bounded variation of subgradients, and weakly smooth problems with Hölder continuous gradients. In addition, for strongly convex problems, they are optimal for smooth problems while nearly optimal for nonsmooth and weakly smooth problems. Finally, numerical experiments for applications in sparse optimization and machine learning are reported, which support our theoretical foundations.
[INMA] 2020-10-13 (14h) : Nonconvex and nonsmooth separable objectives: what if we use piecewise linearization?
At Euler building (room A.002) + MS Teams
Speaker :
Van Hoorebeeck, Loïc
Abstract :
In this presentation, we detail an algorithm that aims at the global solution of a structured nonsmooth and nonconvex problem. The algorithm consists in solving a sequence of surrogate optimization problems where the original objective function is replaced by a carefully constructed piecewise underapproximation.
The method is applied to the economic dispatch problem, an important task in power systems which consists in the optimal allocation of power among committed units. In particular, the consideration of the valve-point effect in gas power plants makes this problem a challenging nonsmooth and nonconvex optimization problem, which fits the framework developed here.
[INMA] 2020-10-06 (14:00) : Scaling Monte Carlo inference for State-Space Models
At MS Teams
Speaker :
Alexander Shestopaloff (School of Mathematical Sciences, Queen Mary University of London, UK )
Abstract :
The iterated conditional Sequential Monte Carlo (cSMC) method is a particle MCMC method commonly used for state inference in non-linear, non-Gaussian state space models. Standard implementations of iterated cSMC provide an efficient way to sample state sequences in low-dimensional state space models. However, efficiently scaling iterated cSMC methods to perform well in models with a high-dimensional state remains a challenge. One reason for this is the use of a global proposal, without reference to the current state sequence. In high dimensions, such a proposal will typically not be well-matched to the posterior and impede efficient sampling. I will describe a technique based on the embedded HMM (Hidden Markov Model) framework to construct efficient proposals in high dimensions that are local relative to the current state sequence. A second obstacle to scalability of iterated cSMC is not using the entire observed sequence to construct the proposal. Typical implementations of iterated cSMC use a proposal at time t that relies only on data up to time t, even when data after time t is available. In high dimensions and in the presence of informative data, such proposals can become inefficient and slow down sampling. I will introduce a principled approach to incorporating all data in the cSMC proposal at time t. By considering several examples, I will demonstrate that both strategies improve the performance of iterated cSMC for state sequence sampling in high-dimensional state space models.
Seminars
Subscribe (ICS)
INMA Contact Info
Mathematical Engineering (INMA)
L4.05.01
Avenue Georges Lemaître, 4
+32 10 47 80 36
secretaire-inma@uclouvain.be
Mon – Fri 9:00A.M. – 5:00P.M.
JOIN US