Ismael Garrido Muñoz

PhD - Universidad de Jaén (2021 - Ongoing)

Bias analysis, detection and mitigation in deep learning-based language models

I am currently pursuing a PhD in deep learning Biases in Deep Learning, focusing on the development of ethical AI systems and enhancing AI transparency. My research aims to address and mitigate biases in AI models to ensure fairness and equity. Additionally, I am dedicated to improving AI interpretability, striving to move beyond the "black box" nature of deep learning algorithms. By promoting ethical standards and fostering a clear understanding of AI, my work contributes to creating more reliable and trustworthy AI technologies. To achieve these goals, I employ a combination of machine learning, natural language processing, and data visualization techniques. I develop tools and methodologies to detect and mitigate biases in deep learning models, ensuring they operate fairly and transparently.

Máster Universitario en Ingeniería Informática - Universidad de Jaén (2016 - 2018)

TFM: Confirmation bias analysis from massive social media data.

Grado en Ingeniería Informática - Universidad de Jaén (2012 - 2016)

TFG: Follow-up of press topics

Senior Developer en Liquid Barcodes Liquid Barcodes (2022 - Current)

Full stack development. API as a Service/SASS for an advertisement/loyalty platform for the retail sector, using microservices (C#, RabbitMQ, Mongo, Redis, AWS) along with some other produts and client side apps (Angular/Vue/React) that either serve as a frontend or are used to manage the advertisment campaigns. Worked on CICD pipelines (Jenkins/Github Actions) within a Kubernetes Cluster, data (mostly Tableu/python), internal tooling.

Full Stack developer at Diputación de JaénUD Ibérica (Abril 2020 - 2021)

Responsible for multiple web services and apps: Sede Electrónica, Acceso (Oauth like handling extra login methods & integrations available to Spanish Citizens), VideoActas, Contro Horario, Plan Director Web (back-office) with .NET related tecnologies, Vue, Angular, Oracle/Informix/SqlServer. Developer of a custom frontend framework integrating the legacy MVC with the new MVVM to accelerate the development of new apps.

Cloud Developer at OfimaticaOfimatica TSS (Julio 2019 - Abril 2020)

Developed a comprehensive cloud-based SaaS solution tailored for the hotel industry covering the full business vertical of hotel operations, utilizing a multi-tenant architecture and CQRS (Command Query Responsibility Segregation) pattern. Worked on performance improving scalability of the system. (C#, Angular, Azure, Azure Queues, MongoDB)

Researcher at the sinbad2 group sinbad2.ujaen.es (Enero 2018 - Julio 2019)

Development of Flintstones 4 modular software (Fuzzy LINguisTic DeciSion TOols eNhacemEnt Suite) for running multi criteria decision making algorithms. Used by research groups. Implemented multiple research papers into it. (Eclipse RCP/Java 8, Web SVG for custom charts)

Precise bias location in Language Models

Recent advancements in large language models like GPT-4, Gemini, Llama 3, and Claude 3, have brought these systems closer to replicating human linguistic capabilities. These models, enhanced by vast amounts of data, improved computational resources, and increasingly complex architectures, are now capable of performing tasks that appear to require an understanding of language. However, this "understanding" is superficial, as the models function based on statistical correlations between tokens rather than true comprehension. Despite their usefulness, these models can also encode and perpetuate biases and prejudices, raising concerns about their impartiality. To address this issue, the paper proposes a technique for detecting where biases are embedded within the model's hidden states, focusing on smaller models with the intention of applying these findings to larger models.

Exploring gender bias in Spanish deep learning models (Read)

Visualization tool developed during the investigation of the bias present in deep learning language models in Spanish. The tool allows us to explore in detail the outcome of the response of the models we present with a set of template sentences, allowing us to compare the behavior of the models when the templates are presented with a context that alludes to a man or a woman. The exploration of the data in the tool is performed at various levels of detail, from visualizing the model output itself with its weights to visualizing the aggregation of the results by categories.

Analysis, Detection and Mitigation of Biases in Deep Learning Language Models (Read)

Recent advances in artificial intelligence have made it possible to make our everyday lives better. However, these models capture the biases present in society and incorporate them into their knowledge. Model will have vastly differents results depending on attributes such as the subject’s gender, race or religion. Bias in AI is encompassed in study areas such as Fairness or Explainability.

A Survey on Bias in Deep NLP (Read)

Deep neural networks are hegemonic approaches to many machine learning areas, including natural language processing (NLP). These networks, somehow, learn a probability distribution of words and relations across the training collection used, inheriting the potential flaws, inconsistencies and biases contained in such a collection. As pre-trained models have been found to be very useful approaches to transfer learning, dealing with bias has become a relevant issue in this new scenario. We introduce bias in a formal way and explore how it has been treated in several networks, in terms of detection and correction.