8 papers
Latent Fact-Checking: Detecting Misinformation through Activation Engineering
Pedro T. Barcelos, Otávio Parraga, Marcelo M. Delucis +3
The proliferation of misinformation online has driven demand for scalable detection systems. While most existing approaches rely on surface-level linguistic features or external kn…
Continual Learning for Sequential Personalization of Small Language Models: A Stability Monitoring Analysis
Thomas S. Paula, Lucas S. Kupssinskü, Rodrigo C. Barros
Small Language Models (SLMs) are increasingly being considered for deployment on edge devices such as laptops, enabling private, low-latency, and locally personalized applications.…
Low-Rank Adapters Initialization via Gradient Surgery for Continual Learning
Joana Pasquali, Ramiro N. Barros, Arthur S. Bianchessi +7
LoRA is widely adopted for continual fine-tuning of Large Language Models due to its parameter efficiency, modularity across tasks, and compatibility with replay strategies. Howeve…
Inference-Time Machine Unlearning via Gated Activation Redirection
Vinícius Conte Turani, Otávio Parraga, João Vitor Boer Abitante +7
Large Language Models (LLMs) memorize vast amounts of training data, raising concerns regarding privacy, copyright infringement, and safety. Machine unlearning seeks to remove the…
Quantization-Robust LLM Unlearning via Low-Rank Adaptation
João Vitor Boer Abitante, Joana Meneguzzo Pasquali, Luan Fonseca Garcia +4
Large Language Model (LLM) unlearning aims to remove targeted knowledge from a trained model, but practical deployments often require post-training quantization (PTQ) for efficient…
A Framework for Benchmarking Fairness-Utility Trade-offs in Text-to-Image Models via Pareto Frontiers
Marco N. Bochernitsan, Rodrigo C. Barros, Lucas S. Kupssinskü
Achieving fairness in text-to-image generation demands mitigating social biases without compromising visual fidelity, a challenge critical to responsible AI. Current fairness evalu…
Inference Time Debiasing Concepts in Diffusion Models
Lucas S. Kupssinskü, Marco N. Bochernitsan, Jordan Kopper +2
We propose DeCoDi, a debiasing procedure for text-to-image diffusion-based models that changes the inference procedure, does not significantly change image quality, has negligible…
Bayesian Attention Mechanism: A Probabilistic Framework for Positional Encoding and Context Length Extrapolation
Arthur S. Bianchessi, Yasmin C. Aguirre, Rodrigo C. Barros +1
Transformer-based language models rely on positional encoding (PE) to handle token order and support context length extrapolation. However, existing PE methods lack theoretical cla…