The PhD Students from IECS Doctoral School and Doctorate Programme in Industrial Innovation will showcase their research projects and will answer the questions of the participants. The jury will choose the 3 best presentations to reward.
The jury is made up of the following companies
Moderator: prof. Bruno Crispo
Industrial Innovation PhD – IID
Alessandro Dai Prè
Implementing R&D in a SME: a focus on innovative heat exchangers
Year 1 – Cycle 36
Heat Exchangers (HEXs) directly determine efficiency and sustainability of industrial processes and plants. Pillow Plate HEXs are a more performant and cost effective alternative to classic HEXs, but lack a fast and validated design method, which is being developed by us with an Agile-like methodology.
Viktória Vozárová
Diagnosability in Multi-Agent Systems
Year 1 – Cycle 36
In multi-agent systems, the agents observe different parts of the system and they share knowledge in a distributed fashion. Our goal is to express industrial diagnosability properties for such systems using epistemic logic and to provide formal methods for verification of these properties.
Hyunho Mo
Evolutionary Neural Architecture Search for Remaining Useful Life Prediction
Year 2 – Cycle 35
We introduce the topic of RUL prediction from sensory information. Then, a manually designed DNN architecture which is called multi-head CNN-LSTM is proposed for the RUL prediction. Finally, we introduce a way of applying EA-based NAS on top of the high-level structure of a multi-head CNN-LSTM.
Q&A section
Information Engineering adn Computer Science PhD – IECS
Federica Lago
More Real than Real: Studying Human Perception of Synthetic Faces
Year 3 – Cycle 34
The realism of recent Deepfakes can lead to severe consequences in case of malicious use. Thus, we designed an online experiment to assess human ability to correctly spot them. A demo version of our experiment is available here http://mmlabsites.disi.unitn.it/face-perception-demo. Challenge your ability to distinguish between real and synthetic faces!”
Nandu Chandran Nair
Is this Enough? Evaluation of Malayalam Wordnet
Year 3 – Cycle 34
The quality of a product is the degree to which a product meets the Customer’s expectations, which must also be valid in the case of lexical semantic resources. Conducting a periodic evaluation of resources is essential to ensure that they meet anative speaker’s expectations and are free from errors. This paper defines the possible errors that a lexical semantic resource can contain, how they may impact downstream applications and explains the steps applied to evaluate and quantify the quality of Malayalam WordNet. Malayalam isone of the classical languages of India. We Propose an approach allowing to subset the part of the WordNet tied to the lowest qual-ity scores. We aim to work on this subset in a crowdsourcing context to improve the quality of the resource and avoid downstream errors.
Andrea Montibeller
GPU in Multimedia Forensics: Why not only for Deep Learning?
Year 1 – Cycle 36
Due to deep learning, GPUs became widely used even in multimedia forensics focusing on methods aimed at supporting the law in courtroom cases but colliding with legislative limitations that make them not a valid source of proof.
In this poster, we propose a way in which GPUs can be used to solve common multimedia forensics problems without involving deep learning.
Burcu Sayin Günel
Active learning for calibration
Year 3 – Cycle 34
Calibration is extremely important in machine learning tasks where failures are costly, and a fallback option to delegate decisions to humans exists. This study investigates if active learning can improve the model calibration by detecting unknown unknowns and decrease high confidence errors.
Q&A section
Federico Mento
New Developments in Lung Ultrasound
Year 2 – Cycle 35
Lung ultrasound (LUS) is used by clinicians to evaluate the state of the lung surface, and is based on the visual evaluation of B-line vertical artifacts. We proposed a multi-frequency quantitative approach that showed a high potentiality in supporting the diagnosis of fibrotic patients.
Michele Grisafi
The weaknesses of IoT – A pressing security challenge
Year 1 – Cycle 36
IoT involves plenty of devices in many sectors, often as part of a network or in security-critical systems, thus becoming valuable targets for cyber-attacks. Unfortunately, most of these devices are low-powered and resource-constrained, thus hard to defend with traditional security techniques.
Mohamed Nabih Ali Mohamed Nawar
Speech enhancement for automatic Speech Recognition
Year 2 – Cycle 35
Automatic speech recognition must be robust to different types of noise. Hence, to mitigate the noise effect is to integrate a speech enhancement (SE) module. In literature, integrating SE with ASR back-end is still an open issue, because the noise deteriorates the ASR performance.
Alessandro Torresani
A V-SLAM guided Mobile Mapping System for Photogrammetric Image Acquisitions
Year 3 – Cycle 34
In the context of 3D reconstruction of complex indoor/outdoor environments, mobile mapping systems represent now the most effective and widespread solutions.
However, their capability to understand the context and detect potential acquisition errors is still limited.
How can we make them smarter?
Leonardo Lucio Custode
Interpretable methods for reinforcement learning
Year 2 – Cycle 35
AI lately achieved impressive breakthroughs in several fields. However, these systems cannot be easily understood by humans. Interpretable AI is a field that aims to fix this problem.
Our results show that we were able to obtain competitive solutions that are very easy to interpret.
Q&A section