Laboratorio Polin-Dipartimento di Matematica - Università di Torino
Workshop on mathematical and computer science methods for new assistive technologies for visual impairments


The Polin Laboratory organizes a Workshop on Mathematical and computer science methods for new assistive technologies for visual impairments on September 8th, 2020.

The issue of drafting and accessing texts containing formulae, graphs and tables by people with visual impairments will be examined by some of the most important experts in the sector. Significant consequences for motor disability (upper limbs) and Specific Learning Disorders (DSA) will be described as well. We will discuss strictly mathematical aspects as well as specific topics related to LaTeX, PDF/UA and to the realization of an OCR (Optical Character Recognition) software for mathematics. Issues related to the application of mathematical methods for computer vision and for the recognition and accessibility of graphs will be discussed as well. 

The workshop will be held online via the Google Meet platform. Details will be sent to the registered participants.

Speakers & Abstracts

  • Boris Doubrov - Dual LAB and PDF Association, Bruxelles

    Accessibility of Math in PDF: Standards and reality


    We discuss the challenges in presenting mathematical expressions in PDF documents in the accessible way. We go all the way from the existing and coming standards to the challenges of the authoring software and the screen readers.

  • Roberto Manduchi - Computer Vision Lab, UC Santa Cruz, USA

    Wayfinding with and without maps


    For people who cannot see or who cannot see well, orientation and wayfinding can be extremely challenging. I will present two cases of interest: a large transit hub, and an uninstrumented building. For each case, I will discuss the potentials and pitfalls of navigation technology. Specifically, I will address the following questions: What level of localization accuracy is needed, and what should be expected? How should directions be provided to be really effective? How can a map of the environment be obtained, and what can be done without a map?

  • Gregorio Pellegrino - Fondazione LIA, Milano

    State of the art and upcoming developments of accessible mathematical and chemical formulas on the web and in EPUBs


    The World Wide Web was born to share scientific research articles, but from the beginning it has shown its limits in presenting mathematical and chemical formulas. Standards have been developed over the years to describe mathematical formulas, but they are not yet widely implemented in browsers. The EPUB standard is based on the same technologies as the Open Web Platform, inheriting its strengths and weaknesses, including the management and support of chemical and mathematical formulas. The paper aims to present the state of the art and the next developments in this field, drawing from the international working groups working on it.

  • Volker Sorge - School of Computer Science, University of Birmingham (UK) & Progressive Accessibility Solutions, Ltd.

    Multi-modal STEM Accessibility Based on Rich Semantics


    Readers with different abilities and needs require diverse formats to work with and comprehend complex STEM content. Consequently a number of approaches and tools have been developed, to make content accessible that was exclusively authored for visual display and print. Not only are these approaches sometimes incompatible but often include the manual adding of meaning and information to content, one expression at a time. I shall argue that we will have to move away from reliance on a benevolent author, a knowledgable transcriber or the articifial restraints to the strict adherence to a particular input format. Instead semantics needs to be automatically generated, often on the fly, and embedded into sufficiently powerful representations to allow for the production of diverse accessible output regardless of the indented audience. On the one hand, this will allow authors, researchers and teachers to be creative in the content production unimpeded by often prohibitive constraints of accessibility requirements. On the other hand it will give readers with special needs independent access to information from which they were previously excluded in their choice of modality. I shall present a number of solutions for making STEM content accessible fully automatically. The key is sufficiently rich semantic interpretation that is produced algorithmically and embedded into the content's representation invisibly. In particular, I shall talk about the versatile rendering of mathematical formulas and the generation of accessible data visualisation straight from statistical models and demonstrate how they can be transformed into text, speech, sound and various tactile formats.

  • Masakazu Suzuki - Science Accessibility Net, Kyushu University, Japan

    Workflow to make STEM documents accessible


    Based on our recent experience to make, first, school text books of mathematics and sciences to Multimedia Daisy books with verified reading voices, and secondly, the lecture note series Koukyuroku of RIMS(*) to PDF with hidden texts including math expressions in LaTeX notations, I will talk about our workflow in each project. Since the fund is limited in both cases, the management of the collaboration of automatic processes and the manual works is very important. Unsolved problems and the steps to be improved in our workflow will also be mentioned. (*) RIMS=Research Institut for Mathematical Sciences Kyoto University.

  • Francesco Tudisco - GSSI Gran Sasso Science Institute, L' Aquila

    Learning with nonlinear eigenvectors


    Spectral methods for clustering and semi-supervised learning are both widely studied and applied in several problems in machine learning and computer vision. These methods are typically based on graph matrices (e.g. the Laplacian) and their eigenvectors. However, as data complexity grows, methods based on nonlinear mappings are growing in popularity as they can achieve much better performance. In this talk I will present some nonlinear versions of classical spectral methods we have recently developed for unsupervised and semi-supervised learning. I will discuss some theoretical properties as well as some computational results on standard benchmark classification datasets.

  • Francesco Vaccarino - Dipartimento di Scienze Matematiche, Politecnico di Torino

    Recent advances in Detecting Mathematical Expressions in Scientific Document Images by ANN


    We will present a  survey on detection method for mathematical expressions in scientific document images with a special focus on the promising use of convolutional network architectures


11.00 - 11.30 Suzuki

11.40 - 12.10 Pellegrino

12.20 - 13.20 Lunch break

13.20 - 13.50 Sorge

14.00 - 14.30 Tudisco

14.40 - 15.10 Doubrov

15.20 - 15.50 Vaccarino

16.00 - 16.30 Manduchi


To partecipate at the workshop the registration is mandatory; please complete the form by September 6, 2020.

If necessary please do not hesitate to contact us at