Mathematics Area - SISSA - Data Science https://www.math.sissa.it/research-group/data-science en Beyond Gaussian models: a mathematical framework for the analysis of real-world data structures https://www.math.sissa.it/seminar/ajs/beyond-gaussian-models-mathematical-framework-analysis-real-world-data-structures <div class="field field-name-field-research-group field-type-taxonomy-term-reference field-label-above"><div class="field-label">Research Group:&nbsp;</div><div class="field-items"><div class="field-item odd" rel=""><a href="/taxonomy/term/2" typeof="skos:Concept" property="rdfs:label skos:prefLabel" datatype="">Mathematical Analysis, Modelling, and Applications</a></div><div class="field-item even" rel=""><a href="/taxonomy/term/4" typeof="skos:Concept" property="rdfs:label skos:prefLabel" datatype="">mathLab</a></div><div class="field-item odd" rel=""><a href="/research-group/data-science" typeof="skos:Concept" property="rdfs:label skos:prefLabel" datatype="">Data Science</a></div></div></div><div class="field field-name-field-speaker field-type-text field-label-hidden"><div class="field-items"><div class="field-item odd" property="schema:name">Lorenzo Bardone</div></div></div><div class="field field-name-field-institution field-type-text field-label-inline clearfix"><div class="field-label">Institution:&nbsp;</div><div class="field-items"><div class="field-item odd" property="schema:affiliation">SISSA</div></div></div><div class="field field-name-field-schedule field-type-datetime field-label-inline clearfix"><div class="field-label">Schedule:&nbsp;</div><div class="field-items"><div class="field-item odd" property="schema:startDate"><span property="schema:startDate" datatype="xsd:dateTime" content="2023-02-10T14:00:00+01:00" class="date-display-single">Friday, February 10, 2023 - 14:00</span></div></div></div><div class="field field-name-field-location field-type-text field-label-inline clearfix"><div class="field-label">Location:&nbsp;</div><div class="field-items"><div class="field-item odd" property="schema:location">Hybrid: in presence and online</div></div></div><div class="field field-name-field-abstract field-type-text-long field-label-above"><div class="field-label">Abstract:&nbsp;</div><div class="field-items"><div class="field-item odd" property="schema:summary"><div class="tex2jax"><p>Neural Network algorithms are very good at extracting information from real-world data. But what makes them perform so well where other algorithms fail? Recent works have linked the success of Neural Networks to their ability to easily access properties of the data distribution that are more complex than the mean vector and the covariance matrix (the high order cumulants-HOC).</p> <p>To investigate the role of HOC for inference tasks, we consider a hypothesis testing problem that compares Gaussian noise with a non-Gaussian whitened distribution. The aim is to bound the sample complexity (SC), i.e. the number of samples necessary to understand which of the two distributions data was drawn from.</p> <p>In this non-Gaussian model the first two cumulants of the data give no information, which is in fact concentrated on HOC. Hence random matrix results, usually employed for similar Gaussian models, cannot be readily applied and new ways to get SC bounds need to be found. In the talk, we will focus on the approach based on the Low-degree Likelihood Ratio.</p></div></div></div></div><span rel="schema:url" resource="/seminar/ajs/beyond-gaussian-models-mathematical-framework-analysis-real-world-data-structures" class="rdf-meta element-hidden"></span><span property="schema:name" content="Beyond Gaussian models: a mathematical framework for the analysis of real-world data structures" class="rdf-meta element-hidden"></span> Thu, 16 Feb 2023 22:49:26 +0000 Ariel Surya Boiardi 14604 at https://www.math.sissa.it SIAM Chapter Colloquia 2021 https://www.math.sissa.it/seminar/siam-chapter-colloquia-2021 <div class="field field-name-field-research-group field-type-taxonomy-term-reference field-label-above"><div class="field-label">Research Group:&nbsp;</div><div class="field-items"><div class="field-item odd" rel=""><a href="/taxonomy/term/2" typeof="skos:Concept" property="rdfs:label skos:prefLabel" datatype="">Mathematical Analysis, Modelling, and Applications</a></div><div class="field-item even" rel=""><a href="/taxonomy/term/4" typeof="skos:Concept" property="rdfs:label skos:prefLabel" datatype="">mathLab</a></div><div class="field-item odd" rel=""><a href="/taxonomy/term/64" typeof="skos:Concept" property="rdfs:label skos:prefLabel" datatype="">MsC Course</a></div><div class="field-item even" rel=""><a href="/research-group/data-science" typeof="skos:Concept" property="rdfs:label skos:prefLabel" datatype="">Data Science</a></div></div></div><div class="field field-name-field-speaker field-type-text field-label-hidden"><div class="field-items"><div class="field-item odd" property="schema:name">Prof. Sara Daneri (GSSI) & Dr. Lorenzo Tamellini (CNR-IMATI Pavia) </div></div></div><div class="field field-name-field-schedule field-type-datetime field-label-inline clearfix"><div class="field-label">Schedule:&nbsp;</div><div class="field-items"><div class="field-item odd" property="schema:startDate"><span property="schema:startDate" datatype="xsd:dateTime" content="2021-07-05T14:00:00+02:00" class="date-display-single">Monday, July 5, 2021 - 14:00</span></div></div></div><div class="field field-name-field-abstract field-type-text-long field-label-above"><div class="field-label">Abstract:&nbsp;</div><div class="field-items"><div class="field-item odd" property="schema:summary"><div class="tex2jax"><p><strong>Prof. Sara Daneri (GSSI)</strong> </p> <p><em>Title: Energy driven pattern formation: an exploration into the mechanisms of self-made order.</em> </p> <p>Abstract:</p> <p>Patterns, intended as ordered regular structures, are ubiquitous in nature. At micro-  and mesoscopic scale ordered structures spontaneously emerge in a surprisingly diverse set of physical and chemical systems and they usually form simple structures with some degree of regularity, such as "bubbles" or "lamellar/striped patterns". One of the leading mechanisms at the base of pattern formation at such scale is since decades recognized to be the competition between short-range attractive forces (favoring pure phases) and long-range repulsive interactions (favoring instead oscillations between different phases). Despite the numerous numerical and experimental studies on the subject, the mathematical mechanisms behind the formation of regular periodic structures is still in most physical cases poorly understood. Two main difficulties reside in the nonlocality of the interactions and in the phenomenon of symmetry breaking, namely the fact that the ground states of such systems have less degrees of symmetry than the energy they minimize.  In this talk we will give an overview of the main open problems in the field and discuss some recent results in which symmetry breaking and pattern formation have been rigorously proved.</p> <p> </p> <p><strong>Dr. Lorenzo Tamellini (CNR-IMATI Pavia)</strong> </p> <p><em>Title: Forward and inverse uncertainty quantification for PDE/ODE with random coefficients</em> </p> <p>Abstract:</p> <p>In this talk we briefly discuss forward and inverse uncertainty quantification (UQ) problems arising in engineering and computational sciences. We focus on two practical examples (one for forward UQ and one for inverse UQ) and illustrate for each problem a) the kind of results that can be obtained, b) the methods used in each case, and c) some of the issues that might come up along the way.</p> <p>The forward UQ problem is a naval engineering problem, that we tackled with a multi-fidelity approach to reduce computational costs. Here, one relevant problem to solve is the impact of the "noise" introduced by the RANS solver in the evaluations of quantities of interest of the analysis (specifically, the hydrodynamic resistance of a ferry). </p> <p>The inverse UQ problem is the tuning of the uncertain parameters of a non-linear system of ODEs (specifically the SIR model, which is the famous prototype for more complex models used in epidemiology), to reliably predict the long-term behavior of the system. Here, the most relevant problem is the so-called "identifiability" of the parameters, which renders the inverse UQ approach a bit trickier than what one might expect. </p> <p> </p> <p><span style="color: #222222; font-family: opensans, Helvetica, Arial, sans-serif; font-size: 14px;">There will be a small break of 10 minutes between the two talks.</span></p> </div></div></div></div><span rel="schema:url" resource="/seminar/siam-chapter-colloquia-2021" class="rdf-meta element-hidden"></span><span property="schema:name" content="SIAM Chapter Colloquia 2021" class="rdf-meta element-hidden"></span> Thu, 24 Jun 2021 12:08:10 +0000 Francesco Romor 14102 at https://www.math.sissa.it Approximation by neural network: a problem of gradient flow https://www.math.sissa.it/seminar/approximation-neural-network-problem-gradient-flow <div class="field field-name-field-research-group field-type-taxonomy-term-reference field-label-above"><div class="field-label">Research Group:&nbsp;</div><div class="field-items"><div class="field-item odd" rel=""><a href="/taxonomy/term/2" typeof="skos:Concept" property="rdfs:label skos:prefLabel" datatype="">Mathematical Analysis, Modelling, and Applications</a></div><div class="field-item even" rel=""><a href="/taxonomy/term/4" typeof="skos:Concept" property="rdfs:label skos:prefLabel" datatype="">mathLab</a></div><div class="field-item odd" rel=""><a href="/taxonomy/term/64" typeof="skos:Concept" property="rdfs:label skos:prefLabel" datatype="">MsC Course</a></div><div class="field-item even" rel=""><a href="/research-group/data-science" typeof="skos:Concept" property="rdfs:label skos:prefLabel" datatype="">Data Science</a></div></div></div><div class="field field-name-field-speaker field-type-text field-label-hidden"><div class="field-items"><div class="field-item odd" property="schema:name">Giovanni Maria Cittaro</div></div></div><div class="field field-name-field-schedule field-type-datetime field-label-inline clearfix"><div class="field-label">Schedule:&nbsp;</div><div class="field-items"><div class="field-item odd" property="schema:startDate"><span property="schema:startDate" datatype="xsd:dateTime" content="2021-05-25T11:00:00+02:00" class="date-display-single">Tuesday, May 25, 2021 - 11:00</span></div></div></div><div class="field field-name-field-abstract field-type-text-long field-label-above"><div class="field-label">Abstract:&nbsp;</div><div class="field-items"><div class="field-item odd" property="schema:summary"><div class="tex2jax"><p>The aim of this seminar is to show good approximation results of neural networks using the Stochastic Gradient Descent method. Approaching the problem in the language of gradient flow permits us to achieve great accuracy even for high dimensional problems. In particular, we will look at the size of the fluctuations and we will cover an example to understand some of the theoretical results.</p> <pre class="moz-quote-pre">Join Zoom Meeting <a class="moz-txt-link-freetext" href="https://sissa-it.zoom.us/j/84672149391?pwd=VWFjajBXTE9Qb1VrdjdpNVFCYzdSdz09">https://sissa-it.zoom.us/j/84672149391?pwd=VWFjajBXTE9Qb1VrdjdpNVFCYzdSdz09</a> Meeting ID: 846 7214 9391 Passcode: 790845</pre></div></div></div></div><span rel="schema:url" resource="/seminar/approximation-neural-network-problem-gradient-flow" class="rdf-meta element-hidden"></span><span property="schema:name" content="Approximation by neural network: a problem of gradient flow" class="rdf-meta element-hidden"></span> Mon, 17 May 2021 17:22:30 +0000 Francesco Romor 14051 at https://www.math.sissa.it Abstraction of Markov Population Dynamics via Generative Adversarial Nets https://www.math.sissa.it/seminar/abstraction-markov-population-dynamics-generative-adversarial-nets <div class="field field-name-field-research-group field-type-taxonomy-term-reference field-label-above"><div class="field-label">Research Group:&nbsp;</div><div class="field-items"><div class="field-item odd" rel=""><a href="/taxonomy/term/2" typeof="skos:Concept" property="rdfs:label skos:prefLabel" datatype="">Mathematical Analysis, Modelling, and Applications</a></div><div class="field-item even" rel=""><a href="/taxonomy/term/4" typeof="skos:Concept" property="rdfs:label skos:prefLabel" datatype="">mathLab</a></div><div class="field-item odd" rel=""><a href="/research-group/data-science" typeof="skos:Concept" property="rdfs:label skos:prefLabel" datatype="">Data Science</a></div></div></div><div class="field field-name-field-speaker field-type-text field-label-hidden"><div class="field-items"><div class="field-item odd" property="schema:name">Francesca Cairoli</div></div></div><div class="field field-name-field-institution field-type-text field-label-inline clearfix"><div class="field-label">Institution:&nbsp;</div><div class="field-items"><div class="field-item odd" property="schema:affiliation">University of Trieste</div></div></div><div class="field field-name-field-schedule field-type-datetime field-label-inline clearfix"><div class="field-label">Schedule:&nbsp;</div><div class="field-items"><div class="field-item odd" property="schema:startDate"><span property="schema:startDate" datatype="xsd:dateTime" content="2021-02-05T11:00:00+01:00" class="date-display-single">Friday, February 5, 2021 - 11:00</span></div></div></div><div class="field field-name-field-location field-type-text field-label-inline clearfix"><div class="field-label">Location:&nbsp;</div><div class="field-items"><div class="field-item odd" property="schema:location">Zoom Meeting</div></div></div><div class="field field-name-field-abstract field-type-text-long field-label-above"><div class="field-label">Abstract:&nbsp;</div><div class="field-items"><div class="field-item odd" property="schema:summary"><div class="tex2jax"><p>Markov Population Models are a widespread formalism used to model the dynamics of complex systems, with applications in Systems Biology and many other fields. The associated Markov stochastic process in continuous time is often analyzed by simulation, which can be costly for large or stiff systems, particularly when a massive number of simulations have to be performed (e.g. in a multi-scale model). A strategy to reduce computational load is to abstract the population model, replacing it with a simpler stochastic model, faster to simulate. Here we pursue this idea constructing a generator capable of producing stochastic trajectories in continuous space and discrete time. This generator is learned automatically from simulations of the original model in a Generative Adversarial setting, meaning we explore the use of Wasserstain GANs, which are state of the art generative models.</p> </div></div></div></div><span rel="schema:url" resource="/seminar/abstraction-markov-population-dynamics-generative-adversarial-nets" class="rdf-meta element-hidden"></span><span property="schema:name" content="Abstraction of Markov Population Dynamics via Generative Adversarial Nets" class="rdf-meta element-hidden"></span> Mon, 01 Feb 2021 08:50:16 +0000 Maria Strazzullo 13982 at https://www.math.sissa.it