INESC-ID RCAAP Portal
Not a member yet
15883 research outputs found
Sort by
How to build a bomb and other tales: How red teaming in large language models improve safety
No full textINESC-ID @ eRisk 2025: Exploring Fine-Tuned, Similarity-Based, and Prompt-Based Approaches to Depression Symptom Identification
No full textThis research publication presents an exploration of fine-tuned, similarity-based, and prompt-based approaches for identifying depression symptoms from language use. The authors' team developed and validated five independent test runs, two of which employed ensemble methods, achieving the highest scores in the official Information Retrieval evaluation, outperforming submissions from 16 other teams. By leveraging fine-tuning of foundation models and synthetic data generation, the authors demonstrated a promising approach for detecting depression symptoms in online platforms
Accelerating NTT with RISC-V Vector Extension for Fully Homomorphic Encryption
No full textThis research paper presents an optimization technique for accelerating Fully Homomorphic Encryption (FHE) using RISC-V Vector Extension. The authors leverage vectorized implementations to accelerate the Number Theoretic Transform (NTT) and Inverse-NTT operations in the Open-Source FHE library, OpenFHE, achieving a maximum speedup of 27.05x for NTT and INTT components. By targeting the CKKS scheme, they demonstrate significant acceleration benefits, showcasing the potential of RISC-V Vector Extension for portable and scalable FHE acceleration on unmodified systems
Towards cyberbullying detection: Building, benchmarking and longitudinal analysis of aggressiveness and conflicts/attacks datasets from Twitter
No full textThis study presents two Portuguese language datasets for cyberbullying detection: Aggressiveness and Conflicts/Attacks. The main contribution of this work is the development of these novel datasets, which provide a comprehensive representation of online aggression and conflicts, enabling the creation of more accurate machine learning models for detecting cyberbullying. By incorporating contextual factors and longitudinal analysis, this study enhances the understanding of online interactions and provides a new approach for identifying cyberbullying, ultimately contributing to the development of effective tools for preventing and intervening in cyberbullying situations
Improving Coding Efficiency of Massive Parallel Intra Prediction Using Alternative References
No full text
