10 February 2023 - 10 February 2023

Range functions are an important tool for interval computations, and they can be employed for the problem of root isolation. In this talk, I will first introduce two new classes of range functions for real functions. They are based on the remainder form by Cornelius and Lohner (1984) and provide different improvements for the remainder part of this form. On the one hand, I will show how centered Taylor expansions can be used to derive a generalization of the classical Taylor form with higher than quadratic convergence. On the other hand, I will discuss a recursive interpolation procedure, in particular based on quadratic Lagrange interpolation, leading to recursive Lagrange forms with cubic and quartic convergence. These forms can be used for isolating the real roots of square-free polynomials with the algorithm EVAL, a relatively recent algorithm that has been shown to be effective and practical. Finally, a comparison of the performance of these new range functions against the standard Taylor form will be given. Specifically, EVAL can exploit features of the recursive Lagrange forms which are not found in range functions based on Taylor expansion. Experimentally, this yields at least a twofold speedup in EVAL.
Room B1.17

23 November 2022 - 23 November 2022

As a servo control engineer, who entered robotics control field afterwards, I found that there is a discrepancy between servo control and robot control. This talk is about this discrepancy between them, and also about how the servo control methodologies can be applied to the robot control. Model-based control using Disturbance Observer (DOB) has been a very powerful tool in the servo control, and this talk introduces how this DOB techniques can be applied to robot control. Force control including Series Elastic Actuators as well as force sensor feedback is one of the main application of this approach.
East Campus USI-SUPSI - Room A1.02 15h00

23 November 2022 - 23 November 2022

“All models are wrong, but some are useful”, I clearly remember the moment when I first heard this well-known aphorism by George Box stepping into the Faculty of Statistical Sciences at the Sapienza University of Rome in 1993. It became a sort of mantra, and my challenge became to prove it wrong: some models “must be” correct and give – as a map – a reliable idea of the territory although they cannot represent every single detail. During this talk I will show my successes and inevitably my failures in many statistical collaborations and projects that I have accumulated since then: in particular, I will present an overview of some of my activities around statistical methods and applications in randomized clinical trials, observational studies and my view that no perfect analysis and/or modelling approach can rescue a badly designed study. I will also discuss about the current infodemic of systematic (and unfortunately not so systematic) reviews and meta-analyses and how the Bayesian approach to this type of apparently simple and low-dimensional datasets can avoid misleading medical conclusions. Lastly, emphasis will be given to the multidisciplinarity of Health Technology Assessment and how the evidence-generation loop to inform medical intervention’ evaluations is a natural fit to the Bayesian paradigm that today’s posterior is tomorrow’s prior.
Room C2.09, 2nd floor Sector C, East Campus USI-SUPSI

17 November 2022

The EU MCSA VIRTUOUS project aims to create a virtual tongue through an integrated computational framework able to screen food for natural ligands targeting taste receptors. VIRTUOUS integrates drug discovery techniques, big-data and machine learning algorithms to predict the organoleptic profile of Mediterranean ingredients based on their chemical composition. Recent developments in the VIRTUOUS project will be presented in the workshop, ranging from new algorithms for advanced modelling simulation, machine learning tools to identify the most relevant descriptors affecting compounds’ taste, and standardized procedures for sensory analysis of virgin olive oils. New results on application of artificial intelligence for quality assessment of extra-virgin olive oil from fluorescence spectrum will be shown. Finally, a first DEMO of the Virtuous platform will be presented
East Campus USI-SUPSI - Room C1.02

13 October 2022 - 13 October 2022

Atmospheric Science commonly deals with complex spatiotemporal fields. Methods based on deep learning (DL), such as convolutional neural networks, have turned out to be powerful tools for analyzing such data. However, predictive DL models are typically trained to optimize loss functions such as the root-mean-square error, which leads to blurry predictions and does not give a quantitative estimate of the uncertainty of the prediction, whose importance is particularly emphasized in the weather and climate fields. Alternatively, probabilistic losses such as cross entropy can produce pointwise uncertainties, but still fail to represent the spatial correlations in the uncertainty. Generative models are able to produce diverse, realistic samples. This makes them – and especially their conditional variants – well suited for representing uncertainty through sample diversity. In the recent years, generative adversarial networks (GANs), have found applications in weather and climate data processing. They can be used for common problems in this field, such as generating physical fields from the corresponding in-situ and remote sensing observations, increasing the resolution of observed data, or predicting the time evolution of data fields. In this presentation, I will give an overview on the applications of generative models in the atmospheric science, with an emphasis on my own work in processing cloud and precipitation observations with them. I will also discuss more generally which problems in climate science could (or already do) benefit from generative models. Furthermore, I will discuss the current challenges and open questions for training generative models for weather and climate applications, and in validating and interpreting their results.
Room B1.14 - East Campus - Lugano

7 October 2022 - 7 October 2022

PH curves are parametric polynomial curves for which the norm of the first derivative is still a polynomial. Introduced by Farouki and Sakkalis in 1990, they found several applications in manufacturing, numerical control machines and robotics. Here we consider their exponential counterparts, based on exponential polynomials, in order to expand the design options for this class of curves. The construction of EPH curves revolves on quaternions with exponential polynomial coefficients, while their evaluation presents some numerical issues which are addressed by the new algorithm proposed.
East Campus USI-SUPSI, Lugano

16 September 2022 - 16 September 2022

We revisit the laser model with cavity loss modulation, from which evidence of chaos and generalized multistability was discovered in 1982 [1]. Multistability refers to the coexistence of two or more attractors in nonlinear dynamical systems. Despite its relative simplicity, the adopted model shows us how the multistability depends on the dissipation of the system. The model is then tested under the action of a secondary sinusoidal perturbation, which can remove bistability when a suitable relative phase is chosen [2]. Such a control strategy is universally known as “phase control” and it was first implemented in the same physical system but at different resonance frequencies [3]. The potential of this control technique has been validated on other paradigmatic chaotic systems such as the Duffing oscillator. In this particular case, it has been verified that the control is more sensitive when applied to the cubic term of the nonlinearity [4]. We recently demonstrated that phase control, which is classified as a nonfeedback method, can be converted to a closed loop control (feedback method) when a suitable adaptive filter is used on the chaotic signal to be processed [5].
East Campus USI-SUPSI - 11:30-12:30

2 September 2022 - 2 September 2022

For centuries mathematics has been an activity carried out by humans for humans. In recent years, a new perspective has arisen, in which mathematics is an activity that humans and machines perform for humans and machines. In the seminar, we exploit this duality within Computer Aided Geometric Design (CAGD) and deep learning frameworks. We consider the problem of constructing spline models starting from data observations and their necessary parameterization. This latter step, namely computing the parametric values associated with each observation, highly affects the shape and accuracy of the final spline model. In particular, we propose a data-driven parameterization based on convolutional neural networks which take in input the relative distances of a variable number of data points and return a suitable parameterization of randomly measured points. We show, with numerical examples, that the proposed scheme leads to improve the spline model accuracy, it is flexible with respect to the input data dimension and can generalize with respect to different kinds of data.
East Campus USI-SUPSI Room D1.13 14:30-16:00

29 August 2022 - 29 August 2022

IDSIA is pleased to organise a mini-workshop on the very relevant topic of explainable AI and scientific understanding which will see the participation of Dr Florian J Bone and Dr Emanuele Ratti.
Room D1.06, Sector D, first floor, Campus Est, USI-SUPSI, Lugano-Viganello