Robert Ghilduta

Hands on Deep Reinforcement Learning Lab

While Large Language Models have captured widespread attention, the reinforcement learning (RL) techniques that shape their behavior through post-training remain opaque to many practitioners. This hands-on workshop demystifies deep reinforcement learning by building intuition from the ground up. Participants will start with fundamental concepts of deep learning and reinforcement learning in simple, visual robotic environments where actions and rewards are immediately observable.

Through interactive exercises, you’ll write reward functions, observe agent behaviors, and debug learning failures in real-time—experiences that directly translate to understanding how RL shapes modern AI systems. By manipulating rewards in grid worlds and simulated robots, you’ll develop crucial insights into reward hacking, exploration-exploitation tradeoffs, and the challenges of specification that persist in post-training. No prior RL experience is required, but basic Python knowledge and familiarity with neural networks is helpful.

Robert Ghilduta is a telecommunications and Software Defined Radio expert, and President of Nuand LLC. Robert’s expertise in Digital Signal Processing and his company’s wireless products have been widely used in industry and defense, including by many departments within the U.S. Department of Defense. During his career, Robert has worked on and led the development of MEMS, Mixed Signal and Digital Logic VLSI designs, and broadband modem IP. Robert is also an active angel investor based in the San Francisco Bay Area, focusing on early-stage information security, telecommunication, and deep technology startups. Previously, Robert was a Member of Technical Staff at Meraki (acquired by Cisco), where he led the software engineering of several generations of enterprise Access Point products. Robert earned a B.S. and M.S. in Computer Engineer with a focus in Electrical Engineering from the Rochester Institute of Technology.