mlsanityData to 13 Oct 2026

Formalizing Task-Space Complexity for Zero-Shot Generalization

arXiv 2606.20967 ↗First posted 18 Jun 2026cs.LG cs.SY eess.SYcited by 0

Not on the map. Why?

Abstract (arXiv)

Policies must operate across diverse conditions, yet a single policy is often conservative while fully adaptive schemes can be complex. We study zero-shot generalization in contextual dynamical systems and introduce a performance-centric, directional task dissimilarity--the signed divergence--that upper bounds the generalization gap from a source context to a target context. The signed divergence induces varepsilon-tolerance sets that certify when a source policy class generalizes, and it yields a concrete notion of task-space complexity: the minimum number of source contexts needed so that every target context incurs at most varepsilon generalization gap. Under a mild local smoothness assumption on performance, the induced tolerance sets admit certified inner/outer balls and instance-dependent volume bounds on task-space complexity. In the finite-oracle setting, source selection reduces to set cover; a greedy strategy inherits the standard H(n) approximation guarantee. Using a Mass-Spring-Damper system with linear-quadratic regulator (LQR) controllers and a nonlinear CartPole system with deep reinforcement learning controllers, we show that greedy selection achieves the same varepsilon-coverage with fewer policies than uniform or random baselines. Our approach delivers a performance-based task similarity measure and practical certificates for building generalizable control with simple policies.

Key facts. No citing papers in the map yet (data through 13 Oct 2026). Builds on 23 of its 37 references that are in the map. First posted on arXiv 18 Jun 2026.

Builds on

23 of 37 references.

References · 23

  1. 2017Proximal Policy Optimization Algorithmscited by 14,592
  2. 2017Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networkscited by 4,870
  3. 2017Domain Randomization for Transferring Deep Neural Networks from Simulation to the Real Worldcited by 1,411
  4. 2017A Survey on Multi-Task Learningcited by 840
  5. 2018Taskonomy: Disentangling Task Transfer Learningcited by 633
  6. 2019Which Tasks Should Be Learned Together in Multi-task Learning?cited by 368
  7. 2015Contextual Markov Decision Processescited by 215
  8. 2015Actor-Mimic: Deep Multitask and Transfer Reinforcement Learningcited by 177
  9. 2021A Survey of Zero-shot Generalisation in Deep Reinforcement Learningcited by 164
  10. 2019Scalable methods for computing state similarity in deterministic Markov Decision Processescited by 133
  11. 2021Contrastive Behavioral Similarity Embeddings for Generalization in Reinforcement Learningcited by 123
  12. 2017Markov Decision Processes with Continuous Side Informationcited by 54
11 more
  1. 2023Multi-Task Reinforcement Learning with Mixture of Orthogonal Expertscited by 45
  2. 2022Contextualize Me -- The Case for Context in Reinforcement Learningcited by 41
  3. 2021When Is Generalizable Reinforcement Learning Tractable?cited by 19
  4. 2025Know When to Explore: Difficulty-Aware Certainty as a Guide for LLM Reinforcement Learningcited by 12
  5. 2024Statistical Context Detection for Deep Lifelong Reinforcement Learningcited by 7
  6. 2024Model-Based Transfer Learning for Contextual Reinforcement Learningcited by 7
  7. 2023Temporal Transfer Learning for Traffic Optimization with Coarse-grained Advisory Autonomycited by 6
  8. 2022Synergistic Scheduling of Learning and Allocation of Tasks in Human-Robot Teamscited by 5
  9. 2025Provable Zero-Shot Generalization in Offline Reinforcement Learningcited by 3
  10. 2025Pessimism Principle Can Be Effective: Towards a Framework for Zero-Shot Transfer Reinforcement Learningcited by 2
  11. 2021Suboptimal coverings for continuous spaces of control taskscited by 1

Built on by

Not cited yet.