theodorosgnk

Lead a multidisciplinary team combining machine learning and sensing for system automation. Optimization of sensor placement for large-scale monitoring using satellite imagery and auxiliary data sources. Reinforcement learning policies for robot sorting tasks. Applied machine learning using hyperspectral and multispectral data for classification and regression tasks. Supervision of project timelines, experimental validation, and deployment of ML solutions on real-world datasets.

Designed and optimized predictive credit scoring models using supervised learning and statistical modeling techniques, based on user behavioral and transactional datasets. Built automated analytical pipelines in Python and SQL for data processing and model deployment. Performed user segmentation and clustering to improve risk stratification and decision policies.

Developed multi-agent RL and regret-based algorithms for scheduling problems under stochastic dynamics. Formulated supervised and reinforcement learning approaches for computation offloading in split inference. Investigated meta-learning methods for policy optimization under non-stationary environments. Delivered internal seminars on online learning, regret minimization, and sequential decision-making methods.

Studied diversity and fairness in recommendation systems operating under cache-friendly content delivery. Developed reinforcement learning-based network control algorithms using Deep Q-Learning. Designed lightweight deep learning architectures for split inference, applied on time series forecasting.

Modeled the user consumption behavior on content platforms (e.g., YouTube, Spotify, Last.fm) using Markov chains and graph-based probabilistic models. Formulated the problem of re-ranking recommendations to incorporate low-cost promoted content in long user sessions. Solved the re-ranking problem via Convex Optimization and Markov Decision Processes. Analyzed efficiency-effectiveness trade-offs in MDP-based recommendation systems on large graph datasets. Studied recommendation-aware caching policies, exploiting the underlying recommender graphs.