BECAS JAE 2023

Listado de Proyectos ofertados por los centros CSIC de la CONEXIÓN AIHUB

Identificador del proyecto: AIHUB-01
Tutor/es: Francesc Moreno-Noguer y Mariella Dimiccoli (mdimiccoli@iri.upc.edu)
Grupo de investigación: Grupo de Percepción y Manipulación, Institut de Robòtica i Informàtica Industrial, CSIC-UPC
Dirección del Centro: Barcelona
Título del proyecto: 3D human pose estimation from egocentric videos of social interactions
Resumen del proyecto: Understanding social interactions from a first-person perspective has compelling applications in Assistive Robotics and Augmented/Virtual reality. A crucial cue for social interaction understanding is the body pose of interacting people, which is paramount in nonverbal communication. However, estimating the body pose of the camera wearer from first-person (egocentric) videos is a challenging task since the camera wearer is largely out of view from a typical wearable camera. This project aims at addressing this challenge by leveraging inter-person interaction dynamics and 3D scene context. The method will be validated using the recently introduced Egobody dataset: https://sanweiliti.github.io/egobody/egobody.html. The student is expected to have excellent programming skills and familiarity with deep learning frameworks (preferably PyTorch). She/he will work in close collaboration with our expert team that has large experience in both 3D pose estimation and social interaction analysis from egocentric videos. Upon reaching the project goals, a scientific publication is expected to be submitted to a major conference or journal. 

---

Identificador del proyecto: AIHUB-02
Tutor/es: Pedro David García
Grupo de investigación: Instituto de Ciencias de Materiales de Madrid (ICMM)
Dirección del Centro: C/ Sor Juana Inés de la Cruz, 3 Campus CANTOBLANCO UAM, Madrid 
Título del proyecto: An Optomechanical Neural Network
Resumen del proyecto: In Von Neumann or Turing machines compute information implementing sequential algorithms that struggles with non-sequential and abstract problems such as image recognition, language translation, decision-making problems, and more. The human brain surpasses the computational power of standard computers when dealing with these tasks. The capacity to learn from examples defines human learning and inspires a radically novel way of information processing: artificial neural networks for neuromorphic computing. A potentially disruptive implementation of neuromorphic computing exploits photons as carriers of information and photonic meshes as an artificial neural network. In this project, we will explore the optomechanical coupling in silicon nanostructures that can be configured to perform neuromorphic computing. Our goal is to get familiar with an experimental setup to measure and readout the mechanical vibrations of the system with optical forces.

---

Identificador del proyecto: AIHUB-03
Tutor/es: Jesús Cerquides, Bjorn Komander (cerquide@iiia.csic.es)
Grupo de investigación: Instituto de Investigación en Inteligencia Artificial (IIIA)
Dirección del Centro: c/ Can Planes s/n, 08193 Barcelona 
Título del proyecto: Antecedentes: La deliberación democrática es un proceso de discusión inclusiva, informada y respetuosa entre ciudadanos, destinada a alcanzar una decisión colectiva sobre un tema en particular. Sin embargo, la deliberación a menudo se ve obstaculizada por la falta de información precisa, puntos de vista conflictivos y comunicación ineficaz. El uso de modelos de lenguaje grandes, como GPT-3, puede ayudar a abordar estos problemas al proporcionar acceso a un vasto corpus de información, facilitar argumentos más matizados y promover un discurso más respetuoso. Pregunta de Investigación: ¿Puede el uso de modelos de lenguaje grandes mejorar la calidad de la deliberación democrática, medida por la precisión de la información, la diversidad de puntos de vista y el nivel general de discurso respetuoso? Metodología: Proponemos un estudio experimental en el que se asignará al azar una muestra de ciudadanos a un grupo de control o a un grupo de intervención. El grupo de control participará en un formato de deliberación estándar, mientras que el grupo de intervención recibirá acceso a un modelo de lenguaje grande durante la deliberación. El modelo de lenguaje estará preentrenado en un corpus de información relevante y proporcionará asistencia en tiempo real a los participantes en forma de hechos sugeridos, argumentos contrarios y resúmenes. Recopilación y Análisis de Datos: La deliberación se registrará y transcribirá para su análisis. La precisión de la información presentada por los participantes se evaluará mediante verificadores de hechos independientes utilizando un esquema de codificación estandarizado. La diversidad de puntos de vista se evaluará mediante el análisis del número y la variedad de argumentos presentados. El nivel general de discurso respetuoso se evaluará utilizando una escala validada. Se llevarán a cabo análisis estadísticos para comparar los resultados entre los dos grupos. Resultados Esperados: Esperamos que el uso de modelos de lenguaje grandes mejore la calidad de la deliberación democrática al mejorar la precisión de la información, promover una gama más diversa de puntos de vista y aumentar el nivel de discurso respetuoso. Implicaciones: El uso de modelos de lenguaje grandes en la deliberación democrática podría tener implicaciones significativas para la política pública y la toma de decisiones democráticas. Al mejorar la calidad de la deliberación, los ciudadanos pueden estar mejor informados y más comprometidos en el proceso democrático, lo que lleva a resultados de política más efectivos y equitativos.

---

Identificador del proyecto: AIHUB-04
Tutor/es: Jesús Cerquides (cerquide@iiia.csic.es)
Grupo de investigación: Instituto de Investigación en Inteligencia Artificial (IIIA) – CSIC
Dirección del Centro: c/ Can Planes s/n, 08193 Barcelona
Título del proyecto: Developing Software for Analyzing ASVs Time Series Using Probabilistic Graphical Models
Resumen del proyecto: A Background: Amplicon sequencing has become a popular method for studying microbial communities. One important analysis task for amplicon sequence data is to analyze the dynamics of microbial communities over time. However, the analysis of ASVs time series data can be challenging due to the high variability and complexity of microbial communities. Currently available software for ASVs time series analysis requires significant computational expertise and can be time-consuming to use. Therefore, there is a need for software that can simplify the analysis of ASVs time series and improve the accuracy of microbial community analysis. Research Question: Can the development of software for analyzing ASVs time series based on probabilistic graphical models improve the accuracy and efficiency of microbial community analysis? Methodology: We propose the development of software for analyzing ASVs time series based on probabilistic graphical models, which will be designed to simplify the process of analyzing ASVs time series data for microbial community analysis. The software will be developed using a combination of open-source programming languages and packages and will support various types of PGMs commonly used in microbial community analysis. The software will allow users to perform various types of ASVs time series analysis, such as trend analysis, change-point detection, and forecasting. The software will be tested using a variety of ASVs time series datasets with different levels of complexity and variability to assess its accuracy and efficiency. Data Collection and Analysis: The accuracy and efficiency of the software will be evaluated using a variety of metrics, such as accuracy of trend analysis, sensitivity and specificity of change-point detection, and forecasting accuracy. The results will be compared to those obtained using currently available software to determine the efficacy of the developed software. Expected Results: We expect that the developed software will improve the accuracy and efficiency of microbial community analysis by simplifying the process of analysing ASVs time series data. The software will enable researchers to conduct more accurate and efficient analysis of their data, leading to more precise understanding of microbial communities over time. Additionally, the software will provide more flexibility in data preprocessing and modeling, allowing researchers to better tailor the analysis to their specific data and research questions. Implications: The development of software for analysing ASVs time series based on probabilistic graphical models has significant implications for the study of microbial communities. The availability of such software will enable researchers to conduct more accurate and efficient analysis of their data, leading to more precise understanding of microbial community dynamics over time. This, in turn, could lead to better management of microbial communities.
The project will be developed with Institut de Ciències del Mar and Real Jardín Botánico, both from CSIC

---

Identificador del proyecto: AIHUB-05
Tutor/es: Jose Manuel de La Rosa (jrosa@imse-cnm.csic.es)
Grupo de investigación: Instituto de Microelectrónica de Sevilla (IMSE) – CSIC
Dirección del Centro: Cl Américo Vespucio, 28. Parque Científico y Tecnológico Cartuja, 41092 Sevilla
Título del proyecto: Automatización del diseño de circuitos integrados mediante el uso de redes neuronales artificiales
Resumen del proyecto: Este trabajo tiene como objetivo el desarrollo de metodologías de optimización y síntesis automática de circuitos integrados analógicos, de señal mixta asistidos mediante algoritmos de Inteligencia Artificial. Como caso de estudio, el trabajo se aplicará al diseño de interfaces analógico/digitales integradas en tecnologías CMOS nanométricas.
http://www2.imse-cnm.csic.es/~jrosa/index.php/Funded-Projects

---

Identificador del proyecto: AIHUB-06
Tutor/es: Carlos Óscar Sorzano Sánchez 
Grupo de investigación: Biocomputing Unit, Centro Nacional Biotecnología (CNB) – CSIC
Dirección del Centro: c/Darwin, 3 Campus Univ. Autónoma, 28049 Cantoblanco – Madrid
Título del proyecto: Desarrollo de algoritmos de inteligencia artificial para el análisis de imagen de macromoléculas biológicas por microscopía electrónica
Resumen del proyecto: La microscopía electrónica de macromoléculas biológicas produce millones de imágenes, con una relación señal a ruido extremadamente baja, que deben ser procesadas de forma eficiente y robusta. En el grupo tenemos una amplia experiencia en el desarrollo de algoritmos de procesamiento de imágenes, que son utilizados por cientos de usuarios por todo el mundo. El estudiante que se una a esta beca participará de estos desarrollos.

---

Identificador del proyecto: AIHUB-07
Tutor/es: Sara Degli-Esposti 
Grupo de investigación: Grupo de Ética Aplicada, Instituto de Filosofía (IFS) – CSIC
Dirección del Centro: C/ de Albasanz, 26, 28037 Madrid
Título del proyecto: Ética de la IA y análisis de affordances de aplicaciones digitales
Resumen del proyecto: El presente plan de formación se centra en el análisis de aspectos éticos, sociales y psicológicos asociados con la adopción de las tecnologías digitales y del fenómeno de la datificación. La literatura relevante proviene de: communication and cultural studies, surveillance studies, ICT law and governance, social psychology, IA ethics, etc. Entre las metodologías más usadas es preciso mencionar: la etnografía digital y el análisis de affordances, análisis de comunidades en redes sociales, análisis del discurso con NLP, análisis de campañas de desinformación, etc. El plan de formación prevé el análisis de datos cuantitativos y cualitativos, la escritura de artículos académicos, la participación en proyectos de investigación si procede. Las actividades se pueden desarrollar tanto en inglés como en español. Dependiendo de la universidad y del convenio con el CSIC, la actividad puede considerarse práctica curricular y/o ser compatible con el desarrollo de un TFG/TFM.

---

Identificador del proyecto: AIHUB-08
Tutor/es: Eduardo R. Hernández
Grupo de investigación: Simulación y Modelización de Materiales, SIMOMA. Instituto de Ciencias de Materiales de Madrid (ICMM) – CSIC
Dirección del Centro: C/ Sor Juana Inés de la Cruz, 3 Campus CANTOBLANCO UAM, Madrid
Título del proyecto: Generative Graph Convolutional Neural Networks applied to the design of New Materials
Resumen del proyecto: Over the last decade or so, Machine Learning (ML) Techniques have found widespread applicability both in industry and science. Particularly, Deep Learning (DL), the branch of ML that is concerned with the design, training and deployment of Artificial Neural Networks (ANN), has demonstrated the ability to address problems that were previously intractable, such as near-human image classification, speech recognition, autonomous driving, etc. However, conventional ANNs have traditionally worked with structured data, such as a matrix of pixels in an image, while very often one is confronted with relational information that cannot be easily cast into a structured data form. This limitation has motivated the development of so-called Graph Neural Networks (GNN). Graphs provide a more general way of representing interrelated data structures which is free of the constraints inherent to structured data. Typically, a graph consists of nodes, encoding items of information or node properties, and edges, representing relations between the nodes in the graph; edges themselves can encode information, such as properties of the relation between the pair of linked nodes. GNNs can take graphs as their input both to be trained and to address a number of graph-related problems, such as node classification, property prediction (at the node or graph level), etc.
To the physicist, chemist or biologist, a graph is an ideal way of representing a crystal, a molecule, a disordered array of atoms (amorphous material), or even a complex bio-molecule. It is therefore not surprising that one can envisage many ways in which graph-based AI strategies may be useful in Condensed Matter and Molecular Physics, and Materials Science. In this project we aim to use Graph Convolutional Neural Networks to accelerate the design and theoretical analysis of crystalline, amorphous and molecular materials. Specifically, our aim is to design generative systems such as Variational Auto-Encoders (VAE) or Generative Adversarial Networks (GAN) capable of working with graphs, that can be trained on existing materials databases, and used in order to create new crystal or molecular structures with desirable structural or chemical/physical properties. This field of research is at the cutting edge of the application of Artificial Intelligence techniques to Materials Science and Condensed Matter Physics.

---

Identificador del proyecto: AIHUB-09
Tutor/es: Francisco del Monte (delmonte@icmm.csic.es)
Grupo de investigación: Instituto de Ciencias de Materiales de Madrid (ICMM) – CSIC
Dirección del Centro: C/ Sor Juana Inés de la Cruz, 3 Campus CANTOBLANCO UAM, Madrid
Título del proyecto: AI for design of low cost, safe, sustainable, and high performance electrolytes for electrochemical energy storage systems
Resumen del proyecto: The energy storage capability of any electrochemical energy storage (EES) device is closely related to the range of voltages in which the electrolyte is electrochemically stable (the so-called Electrochemical Stability Window, ESW). EES devices capable to provide good rate capabilities (RCs, associated to the power density and thus allowing rapid charging) and large cyclabilities are also quite valuable. Previous work carried out by members of the consortium using experimentally-based trial-and-error schemes has demonstrated that electrolytes containing the electrochemical active compound in pseudo-concentrated aqueous dilutions with one (or more than one) additional co-solvents may offer excellent EES performances (i.e., large ESW, high RC and excellent cyclability). However, finding the optimized electrolytes composition that perform best in EES is by no means trivial. The complex electrolyte composition (e.g., with single or double salts in solvent mixtures of water and some additional organic co-solvents) and the lack of a complete understanding of the relationship among composition, physicochemical and transport properties, solvation structures, and electrochemical performance makes non-particularly efficient those attempts at optimization described above (e.g., based on experimental trial-and-error schemes).
This project aims to investigate a holistic approach that combines molecular dynamics simulations and experiments to generate sufficient data for the design of AI models capable to predict the electrolyte composition that, under the constraints of cost, fire safety and sustainability, best perform in EES devices. The AI models developed in this project will contribute to the adoption of AI-based technological processes needed for the development of the modern industrial chemistry based on Green Chemistry.

---

Identificador del proyecto: AIHUB-10
Tutor/es: Mariella Dimiccoli (mdimiccoli@iri.upc.edu)
Grupo de investigación: Grupo de Percepción y Manipulación, Institut de Robòtica i Informàtica Industrial, CSIC-UPC
Dirección del Centro: C/ Llorens i Artigas 4-6, 08028, Barcelona
Título del proyecto: Learning to anticipate future actions
Resumen del proyecto: Anticipation is one of the main and most powerful neuro-cognitive mechanisms of our brain. We heavily rely on it for each of our daily activity: from preparing breakfast to driving a car we continuously figure out what will happen next to better interact with the environment, based on our knowledge of the world around us. However, anticipating the future still represents a big challenge for machines. One of the main reasons is the difficulty in acquiring, representing and leveraging the knowledge our environment for future predictions.
This project aims at addressing this challenge in the specific context of future action prediction by developing a deep learning model based on transformer networks that have proved to be effective in modeling long-range dependencies.The student is expected to have excellent programming skills, familiarity with deep learning frameworks (preferably PyTorch) and a good mathematical background. Working for seven months under this project will allow the student to learn about the most cutting-edge techniques in deep learning and computer vision, as well as to evaluate them in realistic scenarios. Upon reaching the project goals, a scientific publication is expected to be submitted to a major conference or journal.

---

Identificador del proyecto: AIHUB-11
Tutor/es: Sergio Gutiérrez Rodrigo and Luis Martín-Moreno
Grupo de investigación: Q-MAD, Quantum Materials and Devices. Instituto de Nanociencia y Materiales de Aragón (INMA) – CSIC-UNIZAR
Dirección del Centro: Campus San Francisco, Facultad de Ciencias. C/ Pedro Cerbuna, 12 – 50009 Zaragoza (España)
Título del proyecto: Partial differential equations solved with Neural Networks: applications in Nanooptics
Resumen del proyecto: The Nanophotonics research group at the Institute of Nanoscience and Materials of Aragon (INMA) has been actively investigating neural networks (NN) architectures to solve partial differential equations. Their work has yielded promising results [L. Medrano, L. Martin-Moreno, S. G. Rodrigo, “Solving differential equations with Deep Learning: a beginner’s guide”, (submitted to Eur. J. Phys.)], motivating further exploration of differential equations that play a crucial role in understanding physical phenomena. The proposal pretend to study optical systems from the perspective of the solution of their master equations. For example, optical solitons, a captivating topic in Nanooptics involving solutions to complex differential equations, and the time evolution of laser modes in optical cavities, which holds various applications in photonics.

---

Identificador del proyecto: AIHUB-12
Tutor/es: Maria Alberich Carramiñana, Franco Coltraro (maria.alberich@upc.edu)
Grupo de investigación: Grupo de Percepción y Manipulación, Institut de Robòtica i Informàtica Industrial, CSIC-UPC
Dirección del Centro: C/ Llorens i Artigas 4-6, 08028, Barcelona
Título del proyecto: Towards a new paradigm for regression and classification problems using Computational Topology
Resumen del proyecto: We will develop a predictive algorithm to tackle the classical classification and regression problem of Machine Learning from a completely new perspective using tools from Computational Topology. This novel method is expected to be as precise as state of the art Deep Learning methods while being completely explainable (i.e. not a black a box). The idea will be to study and reconstruct geometrically and topologically the space of features, using as main tools Delaunay triangulations and discrete Morse flows, in order to later be able to predict the label of an unknown point using its generalized barycentric coordinates. We are seeking a student with a deep interest in (differential and computational) Geometry and Artificial Intelligence. Programming skills in Python/MATLAB are required.

---

Identificador del proyecto: AIHUB-13
Tutor/es: Marcelo Bertalmío
Grupo de investigación: Instituto de Óptica “Daza de Valdés” (IO) – CSIC
Dirección del Centro: C/ Serrano, 121, 28006, Madrid
Título del proyecto: Development of neural networks with Intrinsically Nonlinear Receptive Fields
Resumen del proyecto: The responses of visual neurons, as well as visual perception phenomena in general, are highly nonlinear functions of the visual input, while vision models and artificial neural networks (ANN) are grounded on the notion of a linear receptive field (RF). The linear RF has a number of inherent problems: it changes with the input, it presupposes a set of basis functions for the visual system, and it conflicts with recent studies on dendritic computations. In a recent article we have proposed to model the RF in a nonlinear manner, introducing the intrinsically nonlinear receptive field (INRF): https://www.nature.com/articles/s41598-020-73113-0 The INRF model is more physiologically plausible, and we have proved that ANNs with INRF modules instead of linear filters have a remarkably improved performance and better emulate basic human perception. The main goal of this JAE-Intro project is to further develop neural networks based on the INRF formulation, collaborating in interdisciplinary research in vision science, mathematics and machine learning.

---

Identificador del proyecto: AIHUB-14
Tutor/es: Teresa Serrano Gotarredona (teresa.serrano@csic.es)
Grupo de investigación: Grupo Neuromorphs. Instituto de Microelectrónica de Sevilla (IMSE) – CSIC
Dirección del Centro: C/ Américo Vespucio, 28. Parque Científico y Tecnológico Cartuja, 41092 Sevilla · Spain
Título del proyecto: Algoritmo de atención visual bioinspirado
Resumen del proyecto: En el grupo neuromórfico del IMSE disponemos de diseños de sensores de visión bioinspirados que permiten controlar regiones de atención visual. Se propone realizar grabaciones con el sensor de visión del IMSE y utilizar dichas grabaciones para desarrollar algoritmos de atención que permitan localizar regiones de interés. Estos algoritmos se utilizarán para determinar en tiempo real las regiones de interés del sensor. Para esto, se estudiará la programación de los algoritmos en microcontroladores o FPGAs 

---

Identificador del proyecto: AIHUB-15
Tutor/es: Luis Alejandro Camuñas Mesa (camunas@imse-cnm.csic.es)
Grupo de investigación: Grupo Neuromorphs. Instituto de Microelectrónica de Sevilla (IMSE) – CSIC
Dirección del Centro: Cl Américo Vespucio, 28. Parque Científico y Tecnológico Cartuja, 41092 Sevilla · Spain
Título del proyecto: Bio-inspired Vision Processing Systems with Deep Learning
Resumen del proyecto: AI has a severe problem: it consumes an enormous amount of energy. Present day AI systems need to run on power-hungry GPU-driven data centers interconnected with the users through fast internet. This trend is unsustainable and today we know that with the present trend, by 2030 internet and data centers will consume 20% of the world’s electricity. On the other hand, the human brain consumes just 20W of power while being capable of cognitive tasks not yet mastered by man-made machines, and while continuously interacting with all body sensors and actuators. The brain uses a different technology than standard computers, which uses neurons that are over one million times slower, become defective and compute with poor precision. However, information encoding is done through population based nervous spikes that exploit spatial sparsity and time-driven computing principles. Spiking Neural Networks (SNNs) are the third generation of Neural Networks that try to imitate such computing principles. World-wide top computer industries are presently investing strongly in the potential of SNNs, with the hope to deploy low-energy AI on portable edge devices (phones, tablets, appliances, toys, security and surveillance). Examples are IBM with their TrueNorth chip and systems, Intel with their Loihi chip and related systems, or on the academic side the EU Flagship Project “The Human Brain Project” which has provided, among many other outputs, the SpiNNaker computer, an SNN machine capable of simulating in real time 1-billion neurons (1% of the human brain). At IMSE, the neuromorphic group has over 25 years of experience with SNN hardware, vision sensors, and computing algorithms. We have SpiNNaker and Loihi hardware devices, as well as vision sensors that directly provide spiking output information, similar to biological retinas, ready to be processed by SNN hardware. The neuromorphic group at IMSE has participated as co-founder of spin-off companies Prophesee (www.prophesee.ai) which produces spiking retina chips (called Dynamic Vision Sensors – DVS), and GrAI-Matter-Labs (www.graimatterlabs.ai) producing SNN processing hardware.
The project for the successful candidate will consist in developing neural processing algorithms for visual recognition, focusing on software implementation in a first stage and eventually on hardware platforms. For that the following tasks are identified: (a) getting familiar with the use of DVS cameras available at IMSE, understanding their operation principle and obtaining several datasets under different environments, (b) development of multi-layer neural networks for high-speed processing of visual information provided by DVS cameras, (c) implementation of complete neuromorphic systems with online learning capable of real-time object recognition. For hardware implementation, several platforms are available at IMSe, like SpiNNaker or Loihi computing platforms, and general purpose FPGAs or custom-made chips.
Therefore, the project can be adapted to the candidate’s preferences and prior training, emphasizing more the computational and algorithmic aspects or setting the strength onto more hardware specific aspects. Schedule and duration of the scholarship is negotiable with the successful candidate in order to adjust them to the candidate’s restrictions and preferences during the training at IMSE.

---

Identificador del proyecto: AIHUB-16
Tutor/es: Mireia Bargalló González (mireia.bargallo.gonzalez@csic.es)
Grupo de investigación: Grupo de Microfabricación e integración de sensores y fuentes de energía (MESSI). Instituto de Microelectrónica de Barcelona (IMB-CNM) – CSIC
Dirección del Centro: C/ dels Til·lers. Campus Universitat Autònoma de Barcelona (UAB). Bellaterra, Barcelona
Título del proyecto: Memristive Devices for Artificial Intelligence Hardware
Resumen del proyecto: Artificial Intelligence (AI) systems based on brain-inspired computing (neuromorphic) are an emerging field that aims to mimic the biological learning mechanisms of neurons and their synapses. In AI electronic systems, artificial neurons could be implemented using CMOS technology. However, there are currently no industrial solutions for the development of electronic synapses that would connect these neurons. Recently, it has been demonstrated that nanodevices based on resistive switching memristive structures (composed of Metal-Insulator-Metal, MIM structures) could be suitable devices for mimicking biological synapses due to their analog control over device resistance (synaptic strength) and potential device scaling. However, electronic synapses should also satisfy the biological synaptic learning rules. In this project, memristive devices based on silicon-based and printed technology will be fabricated. Once fabricated, biological learning rules will be tested using incremental programming algorithms. The obtained results will provide technological and operational guidelines for resistive synaptic devices in AI hardware.

---

Identificador del proyecto: AIHUB-17
Tutor/es: Cefe López (c.lopez@csic.es)
Grupo de investigación: Grupo de Cristales Fotónicos. Instituto de Ciencias de Materiales de Madrid (ICMM) – CSIC
Dirección del Centro: Campus de Cantoblanco, C. Sor Juana Inés de la Cruz, 3, 28049 Madrid
Título del proyecto: Plataforma de inteligencia artificial basada en “random lasers” de semiconductor para “reservoir computing”
Resumen del proyecto: La mayoría de las implementaciones de inteligencia artificial y aprendizaje automático (que se inspiran en o simulan el cerebro) se ejecutan en procesadores convencionales de silicio. Sin embargo, la inteligencia artificial requiere arquitecturas fundamentalmente diferentes para acercarse al funcionamiento del cerebro. Los fotones presentan ventajas frente a otros portadores de información como los electrones ya que, careciendo de masa e interacción entre ellos, pueden compartir canales de transmisión y ésta no es disipativa con las consiguientes ventajas en velocidad de computación y eficiencia energética. Los láseres estocásticos son dispositivos fotónicos emisores de luz fáciles de fabricar y que, debido a su capacidad de generar numerosísimos modos, pueden constituir una plataforma capaz de ejecutar computación en reservorio.
El proyecto que se propone se centra en el estudio de fuentes de luz laser basadas en medios difusivos y materiales semiconductores. Estas actividades siguen los resultados prometedores previamente obtenidos por nuestro grupo, véase Nat. Photonics 16, 219-225 (2022), y proponen la fabricación y estudio del funcionamiento de dispositivos obtenidos por ablación láser a partir de diodos láser comerciales.
Específicamente, se persigue establecer la correlación entre los parámetros de procesado y las características (espectrales, de eficiencia, y otras) de los dispositivos obtenidos y evaluar su potencial como reservorios para computación neuromórfica. Se parte de dispositivos comerciales y se procede a su caracterización, modificación y caracterización de los dispositivos modificados. Las tres fases de trabajo del estudiante orientadas a proporcionar conocimientos teóricos y prácticos serán:
– Caracterización de dispositivos comerciales: diodos laser mono y multimodo con emisión en el rojo y azul
– Modificación por ablación laser pulsada de los dispositivos comerciales: empleo de fuente de ablación de femto- y pico-segundos
– Caracterización de los dispositivos modificados: medidas de potencia óptica, espectro, distribución espacial de emisión, y coherencia espacial

---

Identificador del proyecto: AIHUB-18
Tutor/es: Manuel Alberto Matias Muriel (manuel@ifisc.uib-csic.es)
Grupo de investigación: Instituto de Física Interdisciplinar y Sistemas Complejos (IFISC) – CSIC-UIB
Dirección del Centro: Edifici Instituts Universitaris de Recerca. Campus Universitat de les Illes Balears. E-07122 Palma, Mallorca, Spain
Título del proyecto: Mapping the biodiversity of Mediterranean meadows from satellite images using machine learning techniques
Resumen del proyecto: Seagrass meadows, found in coastal marine ecosystems worldwide, play a vital role in enhancing coastal biodiversity, acting as carbon sinks, protecting coastlines from erosion, and improving water clarity through particle sedimentation. Unfortunately, the global seagrass extent has suffered significant losses, with approximately one third already vanished, primarily due to eutrophication, water quality deterioration, habitat destruction, overfishing, and climate change.
This project aims to contribute to the automatic characterization of seagrass biodiversity in the Balearic Islands by utilizing multispectral satellite images. The integration of these images holds the potential to enhance seagrass extent predictions. To achieve this, a supervised neural network will be trained using a biodiversity atlas generated from sonar techniques. Subsequently, the trained neural network will be employed to infer the presence of seagrasses in unmapped areas. Throughout the training process, various hyperparameters of the network will be optimized,

---

Identificador del proyecto: AIHUB-19
Tutor/es: José Ignacio Serrano Moreno y M. Dolores del Castillo Sobrino (jignacio.serrano@csic.es)
Grupo de investigación: Grupo Neural and Cognitive Engineering. Centro de Automática y Robótica (CAR) – CSIC
Dirección del Centro: Ctra. M300 Campo Real, Km 0,200, Arganda del Rey – 28500 Madrid
Título del proyecto: Modelo computacional cognitivo de comportamiento
Resumen del proyecto: Diseño y desarrollo de un modelo computacional cognitivo de comportamiento basado en la experiencia y el aprendizaje por analogía para poder ser implantado en robots. El modelo será híbrido (simbólico/conexionista) y será adaptable a la capacidad sensorial y de actuación de los robots donde se implante.

---

Identificador del proyecto: AIHUB-20
Tutor/es: David Ríos Insua (david.rios@icmat.es)
Grupo de investigación: Grupo DataLab, Instituto de Ciencias Matemáticas (ICMAT) – CSIC. https://datalab.icmat.es/
Dirección del Centro: C/ Nicolás Cabrera 13-15, Campus UAM, 28049 Madrid, Spain https://www.icmat.es/es/como-llegar/
Título del proyecto: Modelos predictivos para gestión de cáncer
Resumen del proyecto: Las actividades a desarrollar se vinculan a los proyectos ONCOSCREEN y MESIAS. En el primero se colaborará en desarrollar modelos de ayuda a la decisión para facilitar la adopción de medidas de screening y tratamientos frente al cáncer colorrectal. En el segundo se colaborará a construir y comparar modelos predictivos para distintos tipos de cánceres a partir de bases de datos disponibles en el Instituto.
Ficha del proyecto: https://oncoscreen.health/

---

Identificador del proyecto: AIHUB-21
Tutor/es: Silvia Gallego Queipo (sgallego@icmm.csic.es)
Grupo de investigación: Instituto de Ciencia de Materiales de Madrid (ICMM) – CSIC
Dirección del Centro: C/ Sor Juana Inés de la Cruz 3, 28049 Madrid, Spain
Título del proyecto: Machine Learning to aid spectral identification
Resumen del proyecto: Spectroscopies are unique techniques for materials characterization, but their interpretration depends on deconvolutional processes far from exact. The aim of this project is to use Machine Learning tools to explore the refinement of different commonly used spectroscopies, among them Mössbauer and X-ray photoemission. At present spectral identification is based on previous interpretations based on extensive databases and literature documentation, and strongly rely on human discrimination. We intend to develop AI algorithms enabling the identification of individual contributions to the convolutional signal, so as to improve the accuracy and gain insight in the physical processes contributing to the spectra. These algorithms can also serve as a guidance to speed up spectral interpretation under massive measurements. Our goal will be to provide a ML/AI model (a specific one for each type of spectroscopy) that can be trained on the basis of the appropriate database. To this end, simulations of spectra based on advanced tools such as density functional theory to build a complementary set of training data will also be considered.

---

Identificador del proyecto: AIHUB-22
Tutor/es: Vicent Costa y Pilar Dellunde (vicent@iiia.csic.es)
Grupo de investigación: Instituto de Investigación en Inteligencia Artificial (IIIA) – CSIC
Dirección del Centro: c/ Can Planes s/n, Campus UAB, 08193 Barcelona
Título del proyecto: Neurosymbolic AI: from Theory to Applications
Resumen del proyecto: Neurosymbolic artificial intelligence (AI) is a recent domain in AI that seeks to merge the knowledge-based symbolic approach with neural network-based methods. It is mainly motivated by application-level regards (e.g., explainability and interpretability) and algorithmic-level considerations (e.g., long-term planning and analogy) and intends to merge the strengths of both approaches and overcome their corresponding drawbacks.
The main goal of this project is to integrate principles and aspects from both approaches and to design hybrid systems in this emerging field of AI. The application domains would be related to tutors’ previous works (i.e., art painting style categorization) or others concerning people with different kinds of disability (e.g., evaluation of the quality of life of people with mental distress). The ideal candidates for this fellowship have programming skills and knowledge of logic and theoretical computer science and are concerned with the ethical aspects of AI systems design.

---

Identificador del proyecto: AIHUB-23
Tutor/es: Sergio Gutiérrez Rodrigo y Francisco J. Salgado-Remacha (sergut@unizar.es)
Grupo de investigación: Q-MAD, Quantum Materials and Devices. Instituto de Nanociencia y Materiales de Aragón (INMA) – CSIC-UNIZAR
Dirección del Centro: Campus San Francisco, Facultad de Ciencias. C/ Pedro Cerbuna, 12 – 50009 Zaragoza (España)
Título del proyecto: Ultrashort laser pulses retrieval using neural network operators
Resumen del proyecto: The development of new communication and computing technologies based on Quantum Optics requires the development of ultrafast pulsed laser sources. Mode-locked laser technology enables the generation of sub-picosecond pulses, surpassing the resolution limits of electronic systems. Consequently, the measurement of these pulses requires optical techniques, typically relying on the generation of a non-linear signal to extract information and using numerical algorithms for pulse reconstruction. To extract both the pulse amplitude and phase requires the use of specialized iterative algorithms. Due to technical limitations, these algorithms are still far from providing truly reliable experimental results.

We recently started a project in the Department of Applied Physics at the University of Zaragoza and at the Institute of Nanoscience and Materials of Aragon (INMA) with the goal of measuring the amplitude and phase of experimental optical ultrashort laser pulses in both spectral and temporal domains using neural networks. As part of this collaboration, a Final Degree Project has been developed with very promising results.

Therefore, our aim is to continue this work implementing new algorithms based on the recently developed neural network operators (NNO). NNOs enable to approximate nearly any nonlinear continuous mathematical operator with a neural network. We expect that NNOs will significantly enhance the speed and accuracy of ultrashort pulse recovery compared to existing methods employed to date.

---

Identificador del proyecto: AIHUB-24
Tutor/es: Júlia Borràs Sol (jborras@iri.upc.edu)
Grupo de investigación: Grupo de Percepción y Manipulación. Institut de Robòtica i Informàtica Industrial (IRII) – CSIC-UPC
Dirección del Centro: C/ Llorens i Artigas 4-6, 08028, Barcelona
Título del proyecto: Study on the influence of different grasp types for cloth manipulation
Resumen del proyecto: Using our new developed virtual reality system to manipulate clothes, the project will study the influence of using different grasp types and geometries on different manipulation primitives with and without dynamics, such as folding, unfolding, spreading a cloth, etc. We will analyse differences on trajectories, velocities and execution times. If we obtain results in the virtual reality simulation on time, we will reproduce some of the manipulation on real robots using 3D printed grippers.

---

Identificador del proyecto: AIHUB-25
Tutor/es: Adrià Colomé Figueras (acolome@iri.upc.edu)
Grupo de investigación: Grupo de Percepción y Manipulación. Institut de Robòtica i Informàtica Industrial (IRII) – CSIC-UPC
Dirección del Centro: C/ Llorens i Artigas 4-6, 08028, Barcelona
Título del proyecto: Modelos predictivos para la interacción física humano-robot
Resumen del proyecto: La interacción física entre humanos y robots en tareas asistenciales como ayudar a vestir a una persona, puede entenderse desde una perspectiva de control considerando las predicciones del comportamiento humano o el grado de aceptación de la persona en la tarea colaborativa, entre otros. En este proyecto, se pretende usar métodos que anticipan el movimiento humano para mejorar la adaptabilidad del comportamiento del robot a la realización de la tarea.

---

Identificador del proyecto: AIHUB-26
Tutor/es: Nuria E. Campillo y Eduardo Hernandez (nuria.campillo@csic.es)
Grupo de investigación: Instituto de Ciencias Matemáticas (ICMAT) -CSIC
Dirección del Centro: Campus UAB, Carrer dels Tillers, s/n, 08193 Bellaterra, Barcelona
Título del proyecto: Multitask Graph Neural Networks for Ames Mutagenicity Prediction
Resumen del proyecto: In drug development, it is important to know as soon as possible not only the biological activity but also the type or drug properties as well as the toxicity. One key toxicity end-point is the mutagenic capacity of a compound. The Ames mutagenicity test constitutes the most frequently used assay to estimate the mutagenic potential of drug
candidates. The Ames test is an in vitro model that consists in the detection of mutations in different Salmonella typhimurium strains in the presence of the compound of interest. The Organisation for Economic Co-operation and Development (OECD) Guidelines for the Testing of Chemicals points out that at least five strains of bacteria
should be used to conduct an Ames test.
In this context it would be useful to develop AI based QSAR models that could adequately predict the outcome of the Ames test for a given compound. Our aim in this project is to develop a graph-based multitasking model that is capable of predicting the outcome of the Ames test for each of the recommended Salmonella trains (not just the global outcome of the test). Such a model could be extremely useful in the context of the chemical and pharmaceutical industry
The training project that we propose is considered in the context of a multidisciplinary project whose global objective is the development of new DYRK1A inhibitors as potential treatment to Alzheimer diseases using AI tools. The specific objectives are:
1. De novo design of inhibitors of DYR1A using AI tools
2. Development of an IA-based QSAR to predict mutagenesis
3. Synthesis of the proposed molecules
4. Biological evaluation

The student of the JAE-intro will focus on the second objective although she/he will be involved in the development of the other objectives.

SPECIFIC GOAL. Development of a IA-based QSAR to predict mutagenesis employing a graph-based multitasking model

---

Identificador del proyecto: AIHUB-27
Tutor/es: Guillem Alenyà y Alberto Olivares-Alarcos (galenya@iri.upc.edu)
Grupo de investigación: Grupo de Percepción y Manipulación. Institut de Robòtica i Informàtica Industrial, CSIC-UPC
Dirección del Centro: C/ Llorens i Artigas 4-6, 08028, Barcelona
Título del proyecto: Robot tasks in collaborative and adaptive scenarios with humans
Resumen del proyecto: The interaction between humans and robots in collaborative and assistive scenarios increases the uncertainty during the tasks’ execution, requiring online adaptations of robots’ plans. How and why a robot updated its plan is usually not clear for humans. Hence, reliable robots must be able to model and reason about their collaborations and potential adaptations, and also to explain them.

The aim of this project is to design and implement several collaborative and adaptive tasks in which robots interact with humans (e.g. in assistive scenarios). The implemented robot behaviors will allow robots to react to the changes in the environment (e.g. exogenous effects of human actions). Such reactivity will require the use of sensors (e.g., force sensors, cameras) and off-the-shelf libraries and methods for a proper robot perception. During the execution of the tasks, an existent ontological model will allow the robot to represent the task knowledge to reason whether the different interactions are or not examples of collaborations, and also, whether adaptations are needed or not. Furthermore, the stored knowledge will be used to provide explanations to humans of the robot’s decisions using an existing methodology developed in a previous work.

During the project, the student will have access to our laboratory equipment (computers, sensors, robots, etc.) and also will engage with our experienced personnel. At the end of this project, the student will have gained experience programming robots, using perception sensors and techniques. Furthermore, they will also have the opportunity to learn about knowledge representation formalisms (e.g. ontologies) and explainability approaches, relevant topics in the robotics state-of-the-art.

---

Identificador del proyecto: AIHUB-28
Tutor/es: Arantza Oyanguren and Luca Fiorini (Arantza.Oyanguren@ific.uv.es)
Grupo de investigación: Física Experimental de Altas Energías en Colisionadores y Grupo de Física del Bosón de Higgs con el Experimento ATLAS del LHC, Instituto de Física Corpuscular de Valencia (IFIC), CSIC-UV
Dirección del Centro: Parc Científic de la Universitat de València C/ Catedrático José Beltrán, 2 E-46980 Paterna · Valencia
Título del proyecto: Bayesian adversarial machine learning methods in regression settings
Resumen del proyecto: In multiple domains such as malware detection, automated driving systems, or fraud detection, machine learning algorithms are susceptible of being attacked by malicious agents who are able to perturb the value of the covariates of instances to attain certain goals. Such problems pertain to the field of adversarial machine learning and have been dealt with mostly through game-theoretic ideas with strong underlying common knowledge assumptions. These are not realistic in numerous application domains in relation to security. We have provided an alternative Bayesian framework that accounts for the lack of knowledge about the attacker’s behavior using adversarial risk analysis. Its focus has been on classification problems. In this project we shall focus on regression problems (i.e. with continuous response variables). Tasks include: conceptual extensions to cover regression problems, design of computational strategies to find robustified regression algorithms (including large scale settings) and applications (with emphasis on competitive business decision in finance).

---

Identificador del proyecto: AIHUB-29
Tutor/es: Angela Ribeiro Seijas (angela.ribeiro@csic.es)
Grupo de investigación: Grupo de Percepción Artificial, Centro de Automática y Robótica, CSIC
Dirección del Centro: CTRA. DE CAMPO REAL KM 0,200 LA POVEDA Arganda del Rey (Madrid)
Título del proyecto: Estrategia evolutiva para la generación y entrenamiento automáticos de Redes Neuronales Convolucionales (CNN) eficientes para la clasificación de imágenes
Resumen del proyecto: Las redes neuronales convolucionales (CNN) se aplican con mucho éxito en visión por computador, en concreto en tareas de detección y clasificación de objetos. Las primeras arquitecturas CNN se han obtenido a partir de elaborados procesos de diseño en el que la pericia del diseñador ha sido la clave. Así los desarrollos disponibles son fruto de años de esfuerzo e ingenio.
El grupo GPA está actualmente trabajando en la aplicación de estrategias evolutivas para la obtención y entrenamiento de arquitecturas CNN optimizadas para entornos y tareas específicos. El estudiante se integraría en esta línea de investigación que ya ha proporcionado resultados muy interesantes. Con el trabajo propuesto, tendrá la oportunidad de acercarse a áreas de IA tan interesantes como la visión por computador, el aprendizaje automático, el aprendizaje profundo, la optimización a través de algoritmos evolutivos; todo ello aplicado a la resolución de un problema complejo importante en la agricultura como es la detección temprana de plagas. Tendrá asimismo la oportunidad de acceder al centro de supercomputación CESGA y participar en la elaboración de un artículo científico. Posibilidad de desarrollo del TFM.

---

Identificador del proyecto: AIHUB-30
Tutor/es: David Arroyo (david.arroyo@csic.es)
Grupo de investigación: Instituto de Tecnologías Físicas y de la Información, CSIC
Dirección del Centro: C/ Serrano, 144. 28006 – Madrid 
Título del proyecto: Estudio de amenazas avanzadas persistentes (APTs -Advanced Persistent Threats-) y su relación con campañas de desinformación
Resumen del proyecto: Uno de los grandes retos en el ámbito de la ciberseguridad y la seguridad nacional viene dado por la capacidad de describir, anticipar y contener campañas de APTs. Las APTs son grupos organizados normalmente con el respaldo de algún estado (aunque no siempre tienen que ser así) que consiguen combinar una avanzada capacidad de recolección de inteligencia sobre organismos y estados, y con esa inteligencia despliegan estrategias sofisticadas de ciberataque. En el contexto actual son cada vez más las ocasiones en las que esa estrategia involucra el uso de técnicas de desinformación. Las operaciones de información y la estrategia de zona gris, por tanto, deben ser tenidas en cuenta a la hora de caracterizar APTs. El plan formativo de esta JAE-Intro tendrá por objeto identificar fuentes de interés para estudiar APTs y su relación con campañas de desinformación. El proyecto de formación se hará en base a los resultados del proyecto europeo TRESCA y en cooperación con el equipo de trabajo del proyecto XAI-Disinfodemics. La actividad realizada en estos proyectos puede ser consultada aquí: https://dargcsic.github.io/publications/