2025-11-05 –, Blue Language: English
Unmanned vassals, from ships to mini-submarines, shaping the new age of marine warfare. However, this transformation, occurring in traditional and high-risk environment, should be both better and transformative from our current state to the future. To this end, we developed admiral-driven machine learning framework for marine operations and resource allocation. This frameworks allows to produce admiral approved solution while still using state-of-the-art machine learning methods to obtain mathematical optimum for different needs. In this talk, we discuss the process we followed to developed this framework with real-world examples, sampled data from secret operation, and the code that could make it all happen.
In this talk, we present an admiral-driven machine learning framework for decision-making in marine operations, focusing on how to align algorithmic outputs with human strategic oversight in high-risk environments. The framework leverages metric learning via Inverse Reinforcement Learning (IRL) to capture admiral-level preferences from demonstrations and historical data, ensuring that solutions remain interpretable and trustworthy. At the same time, it incorporates multi-objective optimization under operational constraints, balancing competing goals such as resource allocation, mission success probability, and platform survivability. We will discuss the algorithms, modeling process, and implementation challenges, highlighting how mathematical optimality can be reconciled with operational realities. While some background in machine learning, reinforcement learning, and constrained optimization will help, the talk is designed to be accessible to a broad technical audience with an interest in autonomous systems. By the end of the session, participants will understand how IRL-based metric learning can embed human expertise into AI systems, how constrained multi-objective optimization can generate actionable and balanced solutions, and how human-in-the-loop frameworks can advance the deployment of machine learning in critical domains such as marine warfare.
Background knowledge: Familiarity with machine learning is required. Some familiarity with RL and multi-agent models is helpful but not a must.
Key takeaways:
1. How metric learning through Inverse Reinforcement Learning can capture expert (or user) preferences and embed them into machine learning models.
2. Practical methods for performing multi-objective optimization in constrained, high-risk environments.
3. Ho to develop a solution that combines ML with human strategic oversight when the human does not know the answer.
Previous knowledge expected
Prof' Teddy Lazebnik is a applied mathematics and computer science researcher with extensive experience leading RnD teams in both the life sciences and financial domains. Over the past decade and a half, Teddy has honed his software development skills, including nine years of experience managing development teams of up to fourteen professionals. Teddy has a proven track record in system architecture, developing production-ready algorithms, and collaborating with clients. His expertise includes bio-physical simulations, big data analysis, and data analysis for information systems. His research is focused on applying advanced mathematics and computer science to the life sciences and socio-economic domains, covering areas such as AI-driven personalized treatment protocols, drug discovery, eXplainable AI, socio-economic systems modeling and simulation, and optimal policy detection from financial data.