Titulo Estágio
Towards Generalization in Tabular Models with LLM-Learned Concepts
Áreas de especialidade
Sistemas Inteligentes
Local do Estágio
DEI
Enquadramento
The field of Natural Language Processing (NLP) has witnessed significant advancements in recent years, enabling computers to understand and generate human language at unprecedented levels. This progress has led to the development of powerful predictive models capable of solving various language-related tasks, including sentiment analysis, machine translation, text summarization, and question answering. While these models have achieved impressive performance, they often suffer from a lack of interpretability, making it challenging to understand the underlying decision-making processes and trust their outputs.
The need for interpretable models in NLP arises from the growing demand for transparency, accountability, and fairness in the deployment of AI systems. It is crucial to understand why a particular decision is made, especially in sensitive domains such as healthcare, finance, and law, where the consequences of erroneous predictions can be severe. Additionally, interpretability help
Objetivo
This thesis proposal aims to explore and develop techniques for creating interpretable models in the domain of NLP. The research will investigate novel methods that not only achieve high performance but also provide human-interpretable explanations for their predictions. The goal is to strike a balance between model complexity and interpretability, ensuring that the resulting models can be readily understood and validated by human users.
Plano de Trabalhos - Semestre 1
1-Literature Review
-Conduct an extensive review of existing literature on interpretability in NLP and related fields.
-Identify current approaches, methodologies, and challenges in developing interpretable models.
2-Problem Definition and Research Questions
-Clearly define the research problem and identify specific research questions to be addressed.
-Refine the scope of the study and determine the key objectives of the research.
3-Data Collection and Preprocessing
-Identify relevant benchmark datasets and acquire necessary permissions or access.
-Preprocess the data, including cleaning, tokenization, and normalization, to prepare it for model training and evaluation.
4-Model Selection and Implementation
-Explore different NLP models and select appropriate architectures for the research objectives.
-Implement and fine-tune the chosen models using the preprocessed data.
5-Interpretable Model Development
-Investigate and develop novel techniques to enhance interpretability in the selected NLP models.
-Implement interpretability methods such as attention mechanisms, explainable embeddings, or rule-based approaches.
6-Evaluation and Preliminary Results
-Evaluate the performance and interpretability of the developed models using appropriate evaluation metrics.
-Analyze and interpret the preliminary results to identify strengths, weaknesses, and areas of improvement.
Plano de Trabalhos - Semestre 2
7-Model Refinement and Experimentation
-Incorporate feedback and insights gained from the preliminary results to refine the developed models.
-Conduct additional experiments and iterations to enhance model performance and interpretability.
8-User Studies and Expert Evaluations
-Design and conduct user studies to assess the usefulness and effectiveness of the interpretable models.
-Collaborate with domain experts to obtain evaluations and feedback on the models' interpretability and practical applications.
9-Comparative Analysis and Discussion
-Compare the developed interpretable models with existing approaches in terms of performance and interpretability.
-Analyze the strengths and limitations of the proposed techniques and highlight their contributions to the field.
10-Discussion and Conclusion
-Summarize the key findings and insights from the research.
-Discuss the implications of the research in the context of responsible AI deployment and decision-making processes.
-Formulate conclusions based on the achieved objectives and address any open research questions.
11-Dissertation Writing and Finalization
-Prepare a comprehensive dissertation document, including an introduction, literature review, methodology, results, discussion, and conclusion.
-Review, revise, and proofread the dissertation for clarity, coherence, and accuracy.
Condições
.
Observações
.
Orientador
Pedro Henriques Abreu
pha@dei.uc.pt 📩