A5027978607- Digital Media Forensic Detection
- Digital and Cyber Forensics
- Face recognition and analysis
- Biometric Identification and Security
- Retinal Imaging and Analysis
- Anomaly Detection Techniques and Applications
- Face and Expression Recognition
- Natural Language Processing Techniques
- AI in cancer detection
- Advanced Steganography and Watermarking Techniques
- Radiomics and Machine Learning in Medical Imaging
- Fire effects on ecosystems
- Generative Adversarial Networks and Image Synthesis
- COVID-19 diagnosis using AI
- Video Analysis and Summarization
- Digital Imaging for Blood Diseases
- Network Security and Intrusion Detection
- Topic Modeling
- Systems Engineering Methodologies and Applications
- Agriculture, Land Use, Rural Development
- Data Visualization and Analytics
- Advanced Image and Video Retrieval Techniques
- Gene expression and cancer classification
- Statistical and Computational Modeling
- Parkinson's Disease and Spinal Disorders
Universidade Estadual de Campinas (UNICAMP)
2012-2023
Universidade de Brasília
2020
Institute of Electrical and Electronics Engineers
2020
Regional Municipality of Niagara
2020
IEEE Computer Society
2020
Universidade Federal de Santa Catarina
2006
There are two common ways in which developers incorporating proprietary and domain-specific data when building applications of Large Language Models (LLMs): Retrieval-Augmented Generation (RAG) Fine-Tuning. RAG augments the prompt with external data, while fine-Tuning incorporates additional knowledge into model itself. However, pros cons both approaches not well understood. In this paper, we propose a pipeline for fine-tuning RAG, present tradeoffs multiple popular LLMs, including...
In recent years, software-based face presentation attack detection (PAD) methods have seen a great progress. However, most existing schemes are not able to generalize well in more realistic conditions. The objective of this competition is evaluate and compare the generalization performances mobile PAD techniques under some real-world variations, including unseen input sensors, instruments (PAI) illumination conditions, on larger scale OULU-NPU dataset using its standard evaluation protocols...
With the widespread use of biometric authentication comes exploitation presentation attacks, possibly undermining effectiveness these technologies in real-world setups. One example takes place when an impostor, aiming at unlocking someone else's smartphone, deceives built-in face recognition system by presenting a printed image user. In this work, we study problem automatically detecting attacks against methods, considering use-case fast device and hardware constraints mobile devices. To...
Fine-tuning large language models (LLMs) to align with user preferences is challenging due the high cost of quality human annotations in Reinforcement Learning from Human Feedback (RLHF) and generalizability limitations AI Feedback. To address these challenges, we propose RLTHF, a human-AI hybrid framework that combines LLM-based initial alignment selective achieve full-human annotation minimal effort. RLTHF identifies hard-to-annotate samples mislabeled by LLMs using reward model's...
Screening of Diabetic Retinopathy (DR) with timely treatment prevents blindness. Several researchers have focused their work on the development computer-aided lesion-specific detectors. Combining detectors is a complex task as frequently different properties and constraints are not designed under unified framework. We extend our previous for detecting DR lesions based points interest visual words to include additional most common investigate fusion techniques combine classifiers...
One of the major modern threats to society is propagation misinformation — fake news, science denialism, hate speech fueled by social media's widespread adoption. On leading platforms, millions automated and profiles exist only for this purpose. step mitigate problem verifying authenticity profiles, which proves be an infeasible task done manually. Recent data-driven methods accurately tackle performing automatic authorship attribution, although important aspect often overlooked: model...
Most pictures shared online are accompanied by temporal metadata (i.e., the day and time they were taken), which makes it possible to associate an image content with real-world events. Maliciously manipulating this can convey a distorted version of reality. In work, we present emerging problem detecting timestamp manipulation. We propose end-to-end approach verify whether purported capture outdoor is consistent its geographic location. consider manipulations done in hour and/or month...
We live in a connected society which no major event - from music concerts to terrorist attempts happens without being recorded by smartphone and shared instantly the world. This flow of generated data is often unstructured carries reliable information with respect time capture such media pieces. Consequently, posterior reconstruction, understanding fact-checking that are hindered if not properly organized. In this work, we train data-driven method chronologically sort images originated real...
We discuss the problem of restructuring visual data from different heterogeneous sources to analyze an event interest. present X-coherence: a pipeline seeking organize and represent pieces data, tying them coherently with real world one another. also outline research challenges while X-coherence.
The complexity increase and variety of equipments controlled by embedded computers generate the need a multidisciplinary approach for process development those equipments, involving areas software, mechanics, electric electronic engineering. System engineering uses different techniques to concurrently integrate these views. In this sense, it is being specified OMG modeling language, denominated SysML (System Modeling Language). It intends include in single specification an integrated view...
Mobile devices have their popularity and affordability greatly increased in recent years. As a consequence of ubiquity, these now carry all sorts personal data that should be accessed only by owner. Even though knowledge-based procedures are still the main methods to secure owner's identity, recently biometric traits been employed for more effortless authentication. In this work, authors propose facial verification method optimised mobile environment. It consists two-tiered procedure...
Monitoring organic matter is pivotal for maintaining soil health and can help inform sustainable management practices. While sensor-based information offers higher-fidelity reliable insights into changes, sampling measuring sensor data cost-prohibitive. We propose a multi-modal, scalable framework that estimate from remote sensing data, more readily available source while leveraging sparse improving generalization. Using the we preserve underlying causal relations among attributes matter....
Mitigating climate change requires transforming agriculture to minimize environ mental impact and build resilience. Regenerative agricultural practices enhance soil organic carbon (SOC) levels, thus improving health sequestering carbon. A challenge increasing regenerative is cheaply measuring SOC over time understanding how affected by other environmental factors farm management practices. To address this challenge, we introduce an AI-driven Soil Organic Carbon Copilot that automates the...
This work presents a machine learning approach to aid in the classification of Parkinson’s disease (PD). Answers 30-question non-motor symptoms questionnaire are used as input for two classifiers that focus on differentiating subjects (PD) from healthy and PD vs. patients with differential diagnoses. The method was evaluated using cross-validation technique, results surpass those literature.
Synthetic realities are digital creations or augmentations that contextually generated through the use of Artificial Intelligence (AI) methods, leveraging extensive amounts data to construct new narratives realities, regardless intent deceive. In this paper, we delve into concept synthetic and their implications for Digital Forensics society at large within rapidly advancing field AI. We highlight crucial need development forensic techniques capable identifying harmful distinguishing them...
RESUMO A Ciência Forense Digital surgiu da necessidade de tratar problemas forenses na era digital. Seu mais recente desafio está relacionado ao surgimento das mídias sociais, intensificado pelos avanços Inteligência Artificial. produção massiva dados nas sociais tornou a análise forense complexa, especialmente pelo aperfeiçoamento modelos computacionais capazes gerar conteúdo artificial com alto realismo. Assim, tem-se aplicação técnicas Artificial para esse imenso volume informação. Neste...
For better usage of idle resources in a symmetric multiprocessing environment, cloud computing providers often exploit the boundaries parallelism by imposing high CPU subscription rates over their virtualization systems. Moreover, unsuitable resource allocation can significantly impair performance during intensive workloads and increase infrastructure expenditures unnecessarily. It becomes an increasing challenge when dealing with microservices architecture container encapsulated...