In this master thesis it will be investigated the critical trade-off between model accuracy and robustness in the context of artificial intelligence (AI), focusing on large language models (LLMs). It will provide a comprehensive review of the state of the art regarding LLMs architecture, pre-training phase and the techniques used for enhancing models performance. Then it will focus on the problem regarding feature engineering, which is essential for building robust models.
The development phase introduces a pipeline that uses LLMs capabilities with the objective of suggesting new features for improving the performance of a given model. For filtering out the generated features it has been developed a system that uses several correlation scores for avoiding adding noise to the model.
The impact of the new generated features will be evaluated, providing insights into the trade-of between accuracy and robustness. This will permit to understand the importance of balancing this trade-off for developing more reliable AI systems.
In this master thesis it will be investigated the critical trade-off between model accuracy and robustness in the context of artificial intelligence (AI), focusing on large language models (LLMs). It will provide a comprehensive review of the state of the art regarding LLMs architecture, pre-training phase and the techniques used for enhancing models performance. Then it will focus on the problem regarding feature engineering, which is essential for building robust models.
The development phase introduces a pipeline that uses LLMs capabilities with the objective of suggesting new features for improving the performance of a given model. For filtering out the generated features it has been developed a system that uses several correlation scores for avoiding adding noise to the model.
The impact of the new generated features will be evaluated, providing insights into the trade-of between accuracy and robustness. This will permit to understand the importance of balancing this trade-off for developing more reliable AI systems. Read More
Related posts:
A Graph Neural Network for EEG-Based Emotion Recognition With Contrastive Learning and Generative Adversarial Neural Network Data Augmentation
Validación de explicabilidad post-hoc de modelos de tipo caja-negra usando grandes modelos de lenguaje
Mathematical models to estimate leaf area of grapevine varieties Chardonnay, Marselan and Pinot Noir
A Novel Cross-Attention-Based Pedestrian Visual–Inertial Odometry With Analyses Demonstrating Challenges in Dense Optical Flow
Small is big: Meta bets on AI models for mobile devices
Evaluating emotional and subjective responses in synthetic art-related dialogues: A multi-stage framework with large language models