Max Robotics — Global Service Robot Directory + US Certification

About

Sina Ghiassian is a researcher whose work sits at the intersection of representation learning and reinforcement learning (RL), with a particular focus on making RL systems more robust and scalable. His most cited work, "Improving Performance in Reinforcement Learning by Breaking Generalization in Neural Networks" (2020, 8 citations), tackles a fundamental challenge in deep RL: the tendency of neural networks to over-generalize, which can hinder learning in complex environments. Ghiassian proposes novel methods to deliberately break harmful generalization, allowing agents to learn more effective representations without relying on extensive domain-specific prior knowledge—a key step toward more practical, scalable RL systems. In related work, "Prediction in Intelligence: An Empirical Comparison of Off-policy Algorithms on Robots" (2019, 4 citations), he explores how off-policy learning and general value functions (GVFs) can enable robots to continually make predictions about the world, a capability he argues may be central to intelligence. By empirically comparing algorithms on physical robots, Ghiassian bridges theory and real-world application, demonstrating how representation learning and prediction can work together to build more adaptive, intelligent agents. His contributions are shaping how researchers think about generalization, representation, and lifelong learning in RL.