Internship Topics

Apply to one of the internship topics we have for you this year!

This year, CTM offers you 23 internship topics! The topics cover all the 4 areas of the centre: Wireless Networks [WiN], Optical and Electronic Technologies [OET], Multimedia Communication Technologies [MCT], and Visual Computing & Machine Intelligence [VCMI].

Look up the several internship topics in detail by following this link. The descriptions contain the objectives, work-plan, an supervisors of each topic.

Pick up to 3 topics which interest you the most, and fill in the application form indicating them in order of preference. We will try to match you with the topics that interest you the most!

Here you can find a summary of each topic, and direct link to the proposal:

1. Automated 3D Models from (few) Images[MCT]

“Did you ever want to fly around a place you can only see in photos? Thanks to recent machine learning research, we can, sometimes in a matter of a few seconds, create good 3D models from a series of images that show a target from multiple angles by predicting the scene’s depth. In this project, we will explore such techniques, train our neural radiance field (NeRF) models, and compare performance across different inputs.”

Supervisors:
Nuno Pereira (nuno.a.pereira@inesctec.pt)
Pedro Carvalho (pedro.m.carvalho@inesctec.pt)


2. Exploring Places in VR[MCT]

“Exploring a 3D scene in VR can be a great way to discover a real-world location. However, creating a photo-realistic 3D scene can be challenging. Taking advantage of the accessibility of the web and better workflows and tools for content creation, researchers have recently created engaging VR experiences. In this project, we will create a similar VR experience of a space of our choice and document the workflow and choices adopted.”

Supervisors:
Nuno Pereira (nuno.a.pereira@inesctec.pt)
Luís Corte Real (luis.corte-real@inesctec.pt)


3. Volumetric Video Streaming[MCT]

“Volumetric video allows users to see the video content from any position and angle and has been used for live tele-presence in Augmented or Virtual Reality, capture live performances, or motion capture. These videos require large volumes of data, and thus creating a system for streaming of volumetric video can be non-trivial. In this project, we want to explore volumetric video streaming by creating an end-to-end system for capture and streaming.”

Supervisors:
Nuno Pereira (nuno.a.pereira@inesctec.pt)
Paula Viana (paula.viana@inesctec.pt)


4. Synthetic Object Generation for Natural Images[MCT]

“Datasets have assumed a growing importance, to develop deep learning models varied tasks, for example processing visual data. Preparing datasets for training these models is time-consuming and can be costly. Thus, the use of synthetic data is a compromise solution to overcome these obstacles. We intend to test a small set of methodologies for generating synthetic data, focusing on people and objects inside vehicles, which can be used to train models for this type of situation.”

Supervisors:
Pedro Carvalho (pedro.m.carvalho@inesctec.pt)
Nuno Pereira (nuno.a.pereira@inesctec.pt)


5. Image Segmentation for Precision Livestock Farming[MCT]

“Like other areas, livestock has increasingly adopted technological solutions to increase the efficiency of processes, but the inherent characteristics of the scenario pose some challenges. For example, the subjects are live animals, and producers show some resistance to new technology. So, the use of imaging has been of increasing interest as a non-invasive approach, but still presents difficulties, especially in uncontrolled scenarios. We intend to explore a small set of image processing methodologies to segment animals in uncontrolled environments.”

Supervisors:
Pedro Carvalho (pedro.m.carvalho@inesctec.pt)


6. Experimental Characterization of a Smart Antenna at 3 GHz[OET]

“Smart antennas can electronically reconfigure radiation patterns to form the maximum radiation toward a desired direction. Reflectarrays are a good solution for a low‐cost smart antenna because they employ spatial feeding instead of traditional feed networks. This type of antenna consists of a planar array of elements. Each element is designed to incorporate a certainly reflected phase when the incident wave illuminates it, in order to shape the beam in a desired direction. In this work, we will characterize a 2-bit 20×20 reflectarray at 3 GHz.”

Supervisors:
Sofia Inácio (sofia.i.inacio@inesctec.pt)
Luís Pessoa (luis.m.pessoa@inesctec.pt)


7. Face Recognition Door Lock System with RISC-V[OET]

“This project will explore how to build a Face Recognition Door Lock System, using a ESP32-CAM chip, which combines a micro-controller and Wi-Fi capabilities, making it ideal for IoT applications. Face recognition adds an extra layer of security to any digital door lock system. The ESP32-CAM is based on RISC-V, an open-source instruction set architecture (ISA), which is increasingly relevant due to its simplicity, flexibility, and scalability. It is designed to be highly customizable to adapt to specific applications. This project will allow us to explore the intersection of hardware, and software, and how the integration of both can lead to innovative solutions.

Supervisors:
João C. Ferreira (joao.c.ferreira@inesctec.pt)
Nuno Paulino (nuno.m.paulino@inesctec.pt)


8. Implementation of Analog Memory[OET]

Artificial neural networks have been increasingly used to solve multiple problems, ranging from a simple function approximation to a complex real-time object detection or autonomous driving. As the complexity of the software evolves to meet the demand, the hardware must also improve, allowing for faster training and inference times. Given this need, new computing paradigms have been explored, such as the use of local analog/digital memories or memristors to increase system efficiency. This proposal aims at the design and implementation of an analog memory on a printed circuit board, using discrete components.

Supervisors:
Guilherme Carvalho (guilherme.l.leitao@inesctec.pt)
Vítor Grade Tavares (vgt@fe.up.pt)


9. Multi-bit Digital Control of Planar Antenna Arrays[OET]

Antenna arrays are devices with more than one antenna, each called and element. One type of array is the planar array, where the elements are organized in a flat surface. By controlling each element, it is possible to define the radiation direction, which is promising for future RF applications. The control of the element is done with a set of diodes, i.e., a set of N-bits. More bits means better the control, but also more time to perform the complex calculations required. For large arrays (20×20) this affects how fast the direction can be changed. This work will implement a multi-bit control of a reflectarray, improving on the existing 1-bit control.

Supervisors:
Nuno Paulino (nuno.m.paulino@inesctec.pt)
Luís M. Pessoa (luis.m.pessoa@inesctec.pt)


10. UHF-RFID Antenna Performance Analysis[OET]

In near-field applications, RFID tag detection has to be limited within an assigned confined volume close to the antenna surface, so a proper reader antenna design is needed. Based on an existing commercial UHF-RFID read/write antenna, usable for various smart shelf retail applications, a new design was adapted, simulated and manufactured. The antenna is an open-ended or shorted microstrip feed line coupled to periodic planar metal strips. This work will characterize this near-field UHF-RFID (865-960 MHz) antenna experimentally. Namely, the analysis of tags detection varying their distance and position with respect to the antenna surface.”

Supervisors:
Sofia Inácio (sofia.i.inacio@inesctec.pt)
Luís Pessoa (luis.m.pessoa@inesctec.pt)


11. Microscopic Image Analysis for Urine Biomarker Characterization[VCMI]

“Breast cancer has up to 70% risk of recurrence, the highest of all cancer types, resulting also in the highest expense for healthcare systems. The current care method consists of regular invasive cystoscopy and cytology, which require hospital visits. A study revealed that 63% of cytologies resulted in undetermined diagnoses. A non-invasive procedure with comparable or better sensitivity and specificity would help on the following up of patients. This work will design a computer vision algorithm for microscopic image analysis for urine biomarker characterization, and validate the algorithm with ground truth data.”

Supervisors:
Hélder Oliveira (helder.f.oliveira@inesctec.pt)
Tania Pereira (tania.pereira@inesctec.pt)
Raphaël Canadas (rfcanadas@med.up.pt)


12. Generative Adversarial Networks for 3D Medical Data[VCMI]

“Breast cancer (BC) is the cancer with the highest incidence worldwide, and a leading cause of cancer fatalities. Current treatment involves surgery combined with radiotherapy. However, 2/3 of early detected BC tumors are clinically unpalpable. These require invasive, less accurate localization procedures for a conservative therapeutic approach. To succeed, a scientific problem must first be solved: how does the breast, a highly deformable organ, change shape when a patient is in different postures while scanned using different medical imaging modalities? We aim to develop a novel, hybrid in silico/physics informed machine learning approach to generate artificial data based on generative models, and validate the algorithm with ground truth data.”

Supervisors:
Hélder Oliveira (helder.f.oliveira@inesctec.pt)
Tania Pereira (tania.pereira@inesctec.pt)
Tiago Gonçalves (tiago.f.goncalves@inesctec.pt)


13. Transformers for Times Series Forecast[VCMI]

“Time series forecasting with transformers refers to using transformer-based neural networks to predict future time series values. Transformers can handle sequential data effectively, by capturing long-term dependencies in the data, which is important for accurate time series forecasting. They have been particularly successful in Natural Language Processing tasks, like language translation and text generation. This work will Transformer-XL and other architecture variations, employing the “relative positional encoding” technique to handle long-term dependencies in the data effectively, and compare with other recurrent architectures.”

Supervisors:
Hugo S. Oliveira (hugo.oliveira@fc.up.pt)


14. Mixed Supervised Learning for Colorectal Cancer Segmentation[VCMI]

“Manual segmentation of tissues from the giga-pixel Haematoxylin and Eosin (H&E)-stained Whole Slide Images (WSIs) requires expert knowledge and is very time-consuming. This limits the available datasets for of this domain for deep neural networks approaches. Several works have addressed this constraint by resorting to self-supervised and semi-supervised learning. Self-supervised learning followed by supervised fine-tuning is based on transfer learning, a widely used strategy in computer vision. This work proposes a semi-supervised learning paradigm to learn a Whole Slide Image segmentation task efficiently.”

Supervisors:
João D. Nunes (joao.d.fernandes@inesctec.pt)
Tania Pereira (tania.pereira@inesctec.pt)


15. Deep Learning Models for Glioma Characterization using Genomic Data[VCMI]

“Glioblastoma (GBM) has one of the worst 5-year survival rates of all cancers. While genomic studies of the disease have been performed, alterations in the non-coding regulatory regions of GBM have largely remained unexplored. In this project, the objective will be to use the AI models and public datasets to help make better tumor predictions and characterization. The main objective will be the development of deep learning model to predict patient survival, and validate the algorithm with ground truth data.

Supervisors:
Tania Pereira (tania.pereira@inesctec.pt)
Hélder Oliveira (helder.f.oliveira@inesctec.pt)
Jennifer Boer (RMIT Australia)


16. Deep Detection of Glioma Biomarkers using MRI and WSI[VCMI]

“Adult-type diffuse gliomas are the most frequent malignant tumors of the central nervous system. The current diagnostics to detect these gliomas includes Magnetic Resonance Imaging (MRI) and tissue acquisition. Then a biopsy is taken, and routinely assessed by molecular analysis to differentiate diffuse gliomas. Recent results of deep learning models combining MRI and digital pathology exams (WSI) promise to provide complimentary information on the tumor phenotype and, consequently, the tumor genotype. This internship will address using machine learning and deep learning techniques to detect the two most important glioma biomarkers.”

Supervisors:
Tomé Albuquerque (tome.m.albuquerque@inesctect.pt)


17. Explainable Artificial Intelligence[VCMI]

“The information learnt by face recognition systems that rely on deep learning models is not transparent to humans. These highly complex systems learn correlations from non-causal events and infer potential causal relations. So, despite having extraordinary performance, they may be weak against adversarial attacks or unseen samples. For example, these systems can be biased against gender or biases. This work will use PyTorch and other machine-learning frameworks to use new domain and semantic information, for example, eye color and face symmetry, which are usually useful but not always leveraged by current models.”

Supervisors:
Ana F. Sequeira (ana.f.sequeira@inesctec.pt)
Pedro Neto (pedro.d.carneiro@inesctec.pt)


18. Framework for Obstacle-aware Positioning of Mobile Robotic Platform in 6G Networks[WiN]

5G networks provide enhanced mobile broadband, high reliability, and low latency for a massive number devices. Future 6G networks aims at going one step further, in order to enable ubiquitous connectivity while dynamically reconfiguring the positioning of wireless communications cells on-demand. The main objective of this summer internship is the development of a framework for the obstacle-aware positioning of a communications cell carried by a mobile robotic platform, using information from video cameras on the platform, and an Application Programming Interface (API) for controlling it’s position.

Supervisors:
André Coelho (andre.f.coelho@inesctec.pt)
Rui Campos (rui.l.campos@inesctec.pt)


19. Obstacle-aware Positioning Algorithm for a Simple Scenario[WiN]

“During natural or man-made disasters, the reliability of communication infrastructure is crucial. Events like wildfires, earthquakes, floods, and even cyber and terrorist attacks can leave networks unavailable. That’s where Next Generation Networks (NGN) come in. These innovative systems aim to overcome challenges in difficult scenarios, including through the use of unmanned aerial vehicles (UAVs) to provide on-demand wireless connectivity. However, the success relies highly on Line-of-Sight (LoS), which can be easily blocked by obstacles. We’ll be tackling this challenge head-on by providing LoS to user using UAVs, and by evaluating the link using network Simulator 3 (ns-3).

Supervisors:
Kamran Shafafi (kamran.shafafi@inesctec.pt)


20. Online Learning for Wi-Fi Throughput Optimization using Reinforcement Learning[WiN]

“New versions of Wi-Fi have introduced features that allow for higher throughput. However, it is increasingly complex to develop algorithms to optimize the new network parameters. We will address this problem using Reinforcement Learning (RL) to intelligently adapt the throughput. RL algorithms are typically developed and evaluated in simplistic simulations that consider offline learning sufficient. In this summer internship, we aim to evolve the algorithm towards an online learning approach, to approximate its application in a real context.”

Supervisors:
Rúben Queirós (ruben.m.queiros@inesctec.pt)
Hélder Fontes (helder.m.fontes@inesctec.pt)


21. IoT Link Optimization Using Matlab-Simulink Models[WiN]

“MATLAB and Simulink models will be used to to explore the available models and simulation examples for design blocks related to the Internet-of-Things and wireless communication systems. A representative model of a typical communications link will be selected and analyzed, with its main blocks and functionalities to be identified. Further research will be conducted to derive possible configuration improvements tol the baseline model. Finally, the obtained wireless link optimization will be evaluated by comparing the simulation results of the original model to the modified model.”

Supervisors:
Nuno Almeida (nalmeida@inesctec.pt)


22. Performance Evaluation of a Multimodal Underwater Wireless Communications[WiN]

“The concept of the Blue Economy is related to both industrial and commercial activities at sea and has received increasing interest in recent years, including offshore wind farms, environmental monitoring and deep-sea mining. Operating in the ocean requires expensive resources and logistics for supporting manned missions, especially underwater, where broadband communications are limited to short-range applications. Two technologies can be used for these scenarios: radio and optical. This topic offers a unique opportunity to work on cutting-edge multimodal underwater communications, by conducting experiments in a freshwater tank and working with state-of-the-art radio and acoustic modems.

Supervisors:
João P. Loureiro (joao.p.loureiro@inesctec.pt)
Filipe B. Teixeira (fbt@inesctec.pt)
Rui Campos (rcampos@inesctec.pt)


23. Starlink Performance Characterization[WiN]

“This internship is part of the OVERWATCH project, which addresses the development of a backup communications solution for emergency scenarios. This communications solution is based on a Tethered Drone, consisting of aerial and terrestrial components. The aerial component acts as a communications mast, with configurable height, and has a Wi-Fi Access Point and a 5G Base Station. The terrestrial component includes the Internet connection hardware via Low Earth Orbit (LEO) Satellite. This satellite link, based on Starlink’s commercial hardware, will serve as the Internet access backhaul for this communications solution. Since this technology is recent, it is important to characterize the performance of this backhaul, to understand the impact on the quality of service offered. The main objective of this work is the the development of a framework for the periodic evaluation of the network performance.”

Supervisors:
Helder Martins Fontes (helder.m.fontes@inesctec.pt)
Rui Lopes Campos (rui.l.campos@inesctec.pt)