MOnitoring Outbreak events for Disease surveillance in a data science context (MOOD) explores the extraction, analysis, and visualization of health-related data (including also environmental data) to enhance the utility of event-based epidemiological surveillance. INESC-ID is hiring two researchers to work on work-package 3 within the project, specifically targeting tasks related to: Natural language processing for the extraction of information related to disease outbreaks from online media news. Specific tasks to be researched within this context include (a) key-phrase extraction, and (b) text geo-coding/geo-parsing. Deep learning methods for time series modelling and forecasting, envisioning the application of recurrent neural networks and/or state-space models to time series of disease counts.

The student will join the project DIVINA in the scope of its first and third tasks, which aim to develop a new compiler from JavaScript to a simple intermediate language for JavaScript analysis and specification as well as reference implementations for a large subset of the JavaScript and Node.js built-in libraries. The student will mainly work on the validation and testing of the developed compiler and reference implementations. This work will involve both testing the compiler and reference implementations against Test262, the official JavaScript conformance testsuite, as well as measuring its performance and coverage against those of other JavaScript reference implementations. The student will additionally produce documentation for the compiler and reference implementations and create the required infrastructure for making these tools available online.

The candidate is expected to design, implement, and evaluate a tool to extend a Domain-Driven Design (DDD) Domain-specific Language (DSL), that formalizes and maps the most relevant tactical and strategic DDD patterns into a human readable coding language, with support for the mapping of decomposition artifacts from a monolith migration tool. Additionally, the candidate is expected to map the candidate decompositions generated by the monolith migration tool into the extended language, such that its modelling by a software architect can be guided using the information collected from the monolith.

The candidate is expected to design, implement, and evaluate an algorithm to ensure that broadcast messages are received by all participants in a distributed computation in the same order, without performing explicit coordination. Instead, total order should be achieved optimistically, by adding delays to messages before delivery to the application. The algorithm is expected to be subsequently used by the data consistency layer of the DACOMICO project.

The candidate is expected to design, implement, and evaluate a middleware layer that is able to expose to a distributed microservices that execute as part of a same functionality a set of data versions that are consistent with Transactional Causal Consistency.

The candidate is expected to design, implement, and evaluate a tool to automatically detect anomalies that may occur in a given decomposition of a monolithic application in a set of microservices.

The main goal of this work will consist in the design and development of the solutions to address the project’s challenges of text summarization, entity extraction, text indexing and retrieval.

The WSmart Route+ project studies a new paradigm on smart route management. Data analytics combined with route optimization will be explored to improve service routes (for instance waste collection) cooperation supporting the adoption of a route planning based on dynamic routes as opposed to the traditional blind collection that is based on static routes. The Research problem involves optimization of sensor placement and planning of service routes for several real-world situations (e.g.: Recycling Bin network, Fire detection stations, etc). The student will work with Stochastic and Classical Optimization methods, and Machine Learning, comparing classical algorithms with recent approaches, e.g. deep reinforcement learning or submodularity, and hopefully developing new ones.

The project has the following objectives: Requirement analysis and task analysis Development of a VR reading room for radiologists Perform user tests with radiologists Data analysis and validation of the developed app to support the adoption of VR technologies for radiologists VR lab manager

State-of-the-art AI models are unsustainably hunger for data and computational resources: the recent GPT-3 model developed by OpenAI has 175 billion parameters, was trained on 500 billion words, and took 3.14E23 flops to train (on a domestic GPU, it would cost $4.6M and it would require 355 years to train such a model). This has a daunting carbon footprint and poses serious environmental concerns. Recent literature has shown that the training of a state of the art ML model for NLP can emit 5x more CO2 than a car during its entire lifespan. The assignee of this research grant will investigate and develop new scientific methods and technologies for making AI systems more energy and data efficient -- an important goal for Data4Green and the European Green Deal. More in detail, the focus of this grant will be on investigating the adoption of emerging hardware mechanisms (In-memory processing devices, Non-Volatile RAM, Transactional Memory) to increase efficiency of storage and intra-node synchronization.

Propose and develop strategies for the integration of Electric Vehicles in microgrids and electricity markets. Propose optimization algorithms based in bi-level and stochastic optimization.

Design of a power management unit (PMU) integrated circuit (ASIC) to power an array of sensor nodes in an MRI application, that must be resilient to intense magnetic fields. The PMU must be sized to power and voltages levels required for various sensor nodes and should have the following adaptive functions: - Be able to adapt to varied input voltages to adapt to different system configurations. - Be able to adapt output voltages to optimize efficiency.