The Artificial Intelligence Research Institute (IIIA) offers up to 5 scholarships for the introduction to a research career, in the context of CSIC's JAE Intro Programme.

Training plans and mentors:

IIIA-01: Developing Software for Analyzing ASVs Time Series Using Probabilistic Graphical Models
Mentor: Dr. Jesús Cerquides (cerquide@iiia.csic.es)

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 analyzing 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 analyzing 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 Jardin Botánico, both from CSIC.

IIIA-02: Enhancing Democratic Deliberation with Large Language Models: An Experimental Study
Mentor: Dr. Jesús Cerquides (cerquide@iiia.csic.es)

Background: Democratic deliberation is a process of inclusive, informed and respectful discussion among citizens, aimed at reaching a collective decision on a particular issue. However, deliberation can often be hindered by a lack of accurate information, conflicting viewpoints, and ineffective communication. The use of large language models, such as GPT-3, may help to address these issues by providing access to a vast corpus of information, facilitating more nuanced arguments, and promoting more respectful discourse.

Research Question: Can the use of large language models improve the quality of democratic deliberation, as measured by the accuracy of information, diversity of viewpoints, and overall level of respectful discourse?

Methodology: We propose an experimental study in which a sample of citizens will be randomly assigned to either a control group or an intervention group. The control group will participate in a standard deliberation format, while the intervention group will be provided with access to a large language model during the deliberation. The language model will be pre-trained on a corpus of relevant information and will provide real-time assistance to the participants in the form of suggested facts, counterarguments, and summaries.

Data Collection and Analysis: The deliberation will be recorded and transcribed for analysis. The accuracy of information presented by participants will be assessed by independent fact-checkers using a standardized coding scheme. The diversity of viewpoints will be assessed by analyzing the number and range of arguments presented. The overall level of respectful discourse will be assessed using a validated scale. Statistical analyses will be conducted to compare the outcomes between the two groups.

Expected Results: We expect that the use of large language models will improve the quality of democratic deliberation by enhancing the accuracy of information, promoting a more diverse range of viewpoints, and increasing the level of respectful discourse.

Implications: The use of large language models in democratic deliberation could have significant implications for public policy and democratic decision-making. By improving the quality of deliberation, citizens may be better informed and more engaged in the democratic process, leading to more effective and equitable policy outcomes.

IIIA-03: Recommending personalized and explainable learning pathways for young learners in Africa
Mentor: Dr. Filippo Bistaffa (filippo.bistaffa@iiia.csic.es)

Background: Launched in June 2022 for a two-year pilot period, UNICEF’s Yoma (youth agency marketplace) Operational Research (Yoma OR) project aims to support African countries in developing learning to earning opportunities by involving local youth – growing their digital skills while finding concrete answers to the climate resilience challenges their communities face. In practical terms, Yoma is an online platform that collects courses, activities, and learning opportunities offered by different providers such as Google, Meta, SAP, Atingi.

Research Question: Can we recommend personalized and explainable “learning pathways” (i.e., sequences of courses, activities, or learning opportunities on the Yoma platform) that allow young learners to acquire the professional skills necessary to succeed in their desired jobs?

Methodology: Despite its apparent simplicity, our question involves the employment of different solution strategies and technological tools. We propose the following research methodology to pursue our goal:
1.    Data collection and analysis: Courses and learning opportunities are offered by different providers, so their structure and the skills they provide are presented in different ways. Our first task is to collect this heterogenous input and extract the information necessary for the following steps in our methodology. To do this, we will use classical data extraction and analysis techniques, but we will also explore whether the impressive capabilities of modern Large Language Models (LLMs) such as ChatGPT or Bard can be of any help.
2.    Once we have acquired the information about each course, our second task is to model and solve the problem of computing the learning pathway that maximizes the affinity with the skills required by the job preferred by the user. To tackle this task we can leverage an existing Integer Linear Programming (ILP) formulation that we already developed for a similar problem (i.e., recommending university courses) and that can be adapted to our scenario.
3.    Our last (but not least) task is to make sure that our recommendations can be explained and understood by learners. Arguably, if learners do not understand our recommendations, they won’t follow them, so all our previous work will be useless. Hence, we plan to explore different ways to provide “explanations” about computed solutions. Such explanations will also allow users to better understand whether the input they provided really represents their true preferences and, if not, to adjust the behavior of the algorithm accordingly.

Expected Results: Ideally, we expect a small prototype that encompasses at least the first 2 points of our methodology. The inclusion of explainability (point 3) can be seen as a bonus that can be explored if possible.

IIIA-04: New Variants of the MiCRO Negotiation Strategy
Mentor: Dr. Dave de Jonge (davedejonge@iiia.csic.es)

BACKGROUND: AUTOMATED NEGOTIATION
The topic of automated negotiation deals with the question how autonomous software agents can negotiate with each other.

In this field it is assumed there is a set of agents that need to solve a problem together, even though they have conflicting interests. So, the solution that is best for one agent, may not be the best solution for another agent. This means that the agents need to compromise and find a solution that is acceptable to everyone. In order to come to such an agreement, the agents may propose solutions to each another, and each agent may accept or reject the proposals it receives from the other agents. This requires each agent to make a trade-off between its own interests and the interests of the other agents. After all, an agent that is purely selfish and is only willing to accept the solution that is best for itself, will never be able to get the cooperation of the other agents, and therefore only be worse off.

A typical example is the case of a buyer and a seller that are bargaining over the price of a car. While the seller aims to sell the car for the highest possible price, he still needs to make sure the price is low enough for the buyer to accept the deal, and vice versa.

BACKGROUND: THE MiCRO STRATEGY
Recently, an extremely simple new negotiation algorithm, called MiCRO, was introduced by dr. Dave de Jonge which was shown to outperform almost all existing state-of-the-art negotiation algorithms, even though MiCRO is much simpler than those other algorithms. Unfortunately, however, MiCRO is only applicable to negotiations between no more than two agents, and only to problems for which the number of possible solutions is relatively small (less than a million).

To deal with these limitations, dr. de Jonge has proposed some ideas on how MiCRO could be generalized to negotiations among more than two agents, and to negotiations with a larger number of possible solutions (several millions).

GOALS OF THIS PROJECT
The goal of this project is for the student to implement these ideas (in Java or Python), perform experiments, and determine how well these new variants of MiCRO perform against state-of-the-art negotiation algorithms, and under which parameter settings. And perhaps, based on the results of those experiments, the student could even figure out ways to improve MiCRO even further.

Optionally, the task can be made more challenging, by trying to implement an even more advanced algorithm that is applicable to astronomically large test cases (e.g. with 10 to the power 100 possible solutions). This would require the use of more complex search techniques, such as genetic algorithms or tree search.

IIIA-06: Towards Explainable AI by Way of Embodied Cognition
Mentor: Dr. Marco Schorlemmer (marco@iiia.csic.es)

Machine-learning systems based on deep neural networks are currently pattern-matching black boxes that make it difficult for both developers and users to understand when a particular set-up of a neural network is going to be successfully trained and deployed in a trustworthy and robust manner.

The aim of this project is to make deep-learning architectures more transparent to developers and users alike by increasing their degree of explainability by design, with those built-in concepts that are currently lacking and which may help to reveal their underlying assumptions and behaviour.

We will draw from the insights of contemporary cognitive science on embodied cognition, which claims that human conceptualisation and understanding are largley grounded on our bodily experience and the interactions we establish with the environment at a sensorimotor level. We will explore how, taking this perspective of cognition as a reference, we can contribute to one of the fundamental ethical objectives of AI for the coming years, namely the objective of explainability.

At IIIA-CSIC we have developed mathematical and computational models of embodied cognition, appying them to mathematical conceptualisation, diagrammatic reasoning and musical creativity. For this particular project we will team up with researchers from UAB’s Philosophy Department with expertise on embodied and enactive approaches to cognition.

This is a highly interdisciplinary project, bringing together techniques from cognitive linguistics, computer science, mathematics, and philosophy.

IIIA-07: Improve the level of interaction of NPCs in 3D scenarios for simulation
Mentor: Dr. Jordi Sabater-Mir (jsabater@iiia.csic.es)

In the context of simulations in which immersive 3D scenarios are used, the interaction with NPCs (characters controlled by the simulation) is a key element. Until recently, this interaction was either predefined (with the consequent lack of flexibility and adaptability to changes in the simulation) or left much to be desired from the point of view of its realism. With the appearance of LLMs (Language Large Models) the opportunity to use the capacity of these models to improve the interaction with the user using natural language opens up. The difficulty consists in connecting the LLM with the logic of the simulation in such a way that the output of the LLM makes sense and is coherent with the evolution of it. The training plan will be linked to the RHYMAS project: Real-time Hybrid Multiscale Agent-based Simulations for emergency training (PID2020-113594RB-100), a project in collaboration with the "Escola de Bombers, Protecció Civil i Agents Rurals de Catalunya". The training plan will involve working with LLMs both for the input and output of the interaction (thus focused on natural language) and its connection with the simulation logic, emergency simulations based on multi-agent systems, use of engines of games (Unreal Engine) and the use of immersive technologies (virtual reality). The student will work with the research team of the project and will participate in the meetings as well as in the various training activities that are carried out at the IIIA-CSIC, such as the weekly seminars in which they will have the opportunity to learn, from the hands of researchers from first level, other lines of research within the area of Artificial Intelligence.