Filippos Bellos

Conducted research on time series forecasting, leveraging multimodal approaches and aligning LLM with time series embedding spaces. Developed AI in Healthcare solutions, including medical VQA with LLM fine-tuning and clinical note generation from conversations. Led development of an advanced AI agent system for a DARPA project, utilizing Generative AI models for user interaction and scene analysis

Responsible for CNN implementations in Python and C for performance tests in FPGA accelerators and various computer vision projects. Led development in an EU-funded project for cloud tracking, collaborating with Inaccess. Managed C code and its parallelization using CUDA.

Assisted with research on organic optoelectronic devices, primarily focusing on OLEDs and OPCs. Investigated their interface engineering to enhance device performance and characteristics. Contributed to experimental setup and data analysis.

Worked on semi-supervised learning for image classification, focusing on the use of pseudolabels generated by model predictions or graph-based methods. Investigated iterative improvement of pseudolabel quality, incorporating ideas from learning with noisy labels. Developed methods that improved state-of-the-art performance, particularly in few-label settings.

Led deep learning development on projects including near real-time ship detection in Synthetic Aperture Radar images. Implemented SSD in TensorFlow and modified YOLOv3/YOLOv4 using C and CUDA, training models on existing and custom datasets. Developed a model for classifying Gaia observations of unresolved galaxies using galaxy spectra.

Pursued a Master of Science in Electrical and Computer Engineering with a specialization in Machine Learning and Computer Vision. Relevant coursework included Machine Learning, Advanced Topics in Computer Vision, Natural Language Processing, and Biomedical AI.

Activities and societies: Diploma Thesis: “Iterative label cleaning for Semi-Supervised Learning”. Projects on real-time object detection of ships, video classification, and parallelized CNN for image classification.

Completed a Master of Science in Control and Computing, focusing on machine learning and computer vision. Developed a diploma thesis on 'Iterative label cleaning for Semi-Supervised Learning' and engaged in projects on real-time object detection, video classification, and parallel computing.

Activities and societies: Thesis: “Study and FPGA implementation of high-throughput digital filters Finite Impulse Response”.

Obtained a Bachelor of Science in Physics with a major in Electronics, Computers, Telecommunications, and Automation. Completed a thesis on 'Study and FPGA implementation of high-throughput digital filters Finite Impulse Response'.