Tutorials
Empirical Design in Reinforcement Learning for Physical AI
Adam White ⋅ Martha White ⋅ Andrew Patterson
It is no secret that reinforcement learning algorithms are hard to apply in the real world, but why? The answer is not obvious, so we start with a concrete example: on a realistic industrial simulator, multiple widely used algorithms perform worse than a simple baseline. Attendees then dig into a hands-on session to find out why. Next, we step back and discuss what makes the real world hard. These challenges rarely show up in standard benchmarks, and our current algorithms were never designed for them. The rest of the tutorial is about bringing those concerns into the lab. We cover empirical methodology, new benchmarks, and the many ways researchers accidentally cheat. We also take a hard look at common bad practice in RL: too few random seeds, untuned baselines, and overconfident claims built on incorrect confidence intervals. A second hands-on session shows how easily these mistakes can mislead us about what will work in deployment. We do not offer a single checklist for running RL experiments. Instead, attendees leave with three key messages and practical tools they can use in their own work.
Show more
Multi-Objective Optimization for Modern Machine Learning: Foundations, Recent Advances, and Emerging Applications
Jia (Kevin) Liu ⋅ Chaosheng Dong ⋅ Lisha Chen ⋅ Tianyi Chen ⋅ Kaiyi Ji
Machine learning (ML) tasks are increasingly required to satisfy multiple, often competing, objectives simultaneously. Multi-objective optimization (MOO) provides a principled mathematical framework for modeling and solving such problems by systematically characterizing trade-offs among conflicting goals and identifying Pareto-optimal solutions. This tutorial introduces the foundations and recent advances of MOO for machine learning and artificial intelligence. We begin with fundamental concepts, such as Pareto optimality, Pareto fronts, scalarization techniques, preference-based optimization, and gradient-based approaches. We then present modern algorithmic developments that have significantly expanded the applicability of MOO in complex machine learning settings, including distributed MOO, multi-objective bilevel optimization for hierarchical decision-making, multi-objective reinforcement learning, and their use cases. The tutorial is designed for a broad NeurIPS audience, including graduate students, researchers, and practitioners working in ML, artificial intelligence (AI), and optimization-related fields. We will focus on state-of-the-art algorithmic and theoretical developments, while also emphasizing intuition and connections to ML applications.
Show more
From Scores to Evidence: Robust Evaluation of Foundation Models
Ana Gjorgjevikj ⋅ Barbara Koroušić Seljak ⋅ Tome Eftimov
View full details
The Principles of Diffusion Models: Toward Fast and Scalable Diffusion LLMs
Chieh-Hsin Lai ⋅ Subham Sahoo ⋅ Stefano Ermon
View full details
Uncertainty Quantification: From Detecting LLM Hallucinations to Strengthening Reasoning and AI Agents
Artem Shelmanov ⋅ Ekaterina Fadeeva ⋅ Maxim Panov
View full details
From Heuristics to Guarantees: Bandit and RL Blueprint for Reliable Agentic-AI
Aadirupa Saha ⋅ Arun Verma ⋅ Djallel Bouneffouf
View full details
Causal Representation Learning
Julius von Kügelgen ⋅ Francesco Locatello ⋅ Kun Zhang
View full details
Principles of Deep Representation Learning via Low-Dimensional Models
Yi Ma ⋅ Qing Qu ⋅ Peng Wang
View full details
Distilling Reasoning: From Chain-of-Thought to Agentic Capabilities in Smaller Models
Vishrav Chaudhary ⋅ Akhil Arora ⋅ Subbarao Kambhampati
View full details
Neuro-Symbolic Learning in the Era of Large Language Models
Haoxuan Li ⋅ Lan-Zhe Guo ⋅ Wang-Zhou Dai
View full details
Formal Guarantees for Frontier AI Systems
Gagandeep Singh ⋅ Isha Chaudhary ⋅ Nalin Wadhwa
View full details
Successful Page Load