Rewriting history improves RL algorithms for solving multiple tasks.
problem Improving sample efficiency in multi-task reinforcement learning.
method Introducing hindsight relabeling as inverse RL to generalize goal-relabeling techniques.
result Relabeling data using inverse RL accelerates learning in multi-task settings.
Multi-task Inverse Reinforcement Learning (IRL) is the problem of inferring multiple reward functions from expert demonstrations. Prior work, built on Bayesian IRL, is unable to scale to complex environments due to computational constraints. This paper contributes a formulation of multi-task IRL in the more computation…
Generalized Hindsight improves RL by reusing data from one task for another.
problem High sample complexity in reinforcement learning due to wasted uninformative data.
method Approximate inverse reinforcement learning to relabel behaviors with better-suited tasks.
result Efficient reuse of samples, reducing sample complexity on multi-task RL tasks.
New method learns multiple reward functions for complex tasks.
problem Learning reward functions for tasks with multiple ways of solving.
method Combines maximum entropy approach with Dirichlet process clustering.
result Method accurately learns reward functions for complex tasks.
Reliable and effective multi-task learning is a prerequisite for the development of robotic agents that can quickly learn to accomplish related, everyday tasks. However, in the reinforcement learning domain, multi-task learning has not exhibited the same level of success as in other domains, such as computer vision. In…
Exploration in multi-task reinforcement learning is critical in training agents to deduce the underlying MDP. Many of the existing exploration frameworks such as E3, Rmax, Thompson sampling assume a single stationary MDP and are not suitable for system identification in the multi-task setting. We present a nove…
This work extends HiP-MDPs to robust state abstractions for multi-task and meta-reinforcement learning.
problem Limited observability of state in HiP-MDPs for real-world scenarios with rich observation spaces.
method Inspired by Block MDPs, the work extends HiP-MDPs to enable robust state abstractions for multi-task and meta-reinforcement learning.
result Transfer and generalization bounds based on task and state similarity, and sample complexity bounds that depend on the aggregate number of samples across tasks.
Transferring knowledge across a sequence of reinforcement-learning tasks is challenging, and has a number of important applications. Though there is encouraging empirical evidence that transfer can improve performance in subsequent reinforcement-learning tasks, there has been very little theoretical analysis. In this p…
A new method for learning policies in multiple environments.
problem Learning policies that work in different but related environments.
method Decentralized entropy-regularized policy gradient method.
result The method can learn effective policies in multiple environments.
Proof of convergence for multi-objective optimization using inverse reinforcement learning.
problem Proving convergence in multi-objective optimization problems.
method Wasserstein inverse reinforcement learning with projective subgradient method and gradient descent.
result Convergence of inverse reinforcement learning for multi-objective optimization.
Sharing knowledge between tasks is vital for efficient learning in a multi-task setting. However, most research so far has focused on the easier case where knowledge transfer is not harmful, i.e., where knowledge from one task cannot negatively impact the performance on another task. In contrast, we present an approach…
Meta-reinforcement learning algorithms can enable robots to acquire new skills much more quickly, by leveraging prior experience to learn how to learn. However, much of the current research on meta-reinforcement learning focuses on task distributions that are very narrow. For example, a commonly used meta-reinforcement…
This paper investigates multi-task reinforcement learning in non-Markovian decision making, showing benefits in sample efficiency.
problem Investigating multi-task reinforcement learning in non-Markovian decision making processes.
method Developed a joint model class for tasks and used the η-bracketing number to quantify complexity and similarity. result Multi-task reinforcement learning can improve sample efficiency in non-Markovian decision making processes.
TempLe learns transition templates for efficient multi-task RL.
problem Efficiently transferring knowledge across different RL tasks with varying state/action spaces.
method Generates transition dynamics templates to abstract similarities between tasks.
result Achieves significantly lower sample complexity than single-task or multi-task methods.
Soft modularization improves sample efficiency and performance in reinforcement learning.
problem Challenges in training multiple tasks jointly in reinforcement learning.
method Explicit modularization technique on policy representation, soft modularization method.
result Improves sample efficiency and performance over strong baselines in robotics manipulation tasks.
Adversarial online multi-task RL with task separation.
problem Minimize regret in an adversarial online multi-task setting with unknown MDPs.
method Prove minimax and instance-specific lower bounds, develop a clustering algorithm with optimal sample complexity and regret.
result Tight sample complexity and regret bounds for adversarial online multi-task RL.
In this paper we investigate two hypothesis regarding the use of deep reinforcement learning in multiple tasks. The first hypothesis is driven by the question of whether a deep reinforcement learning algorithm, trained on two similar tasks, is able to outperform two single-task, individually trained algorithms, by more…
While deep learning and deep reinforcement learning (RL) systems have demonstrated impressive results in domains such as image classification, game playing, and robotic control, data efficiency remains a major challenge. Multi-task learning has emerged as a promising approach for sharing structure across multiple tasks…
Proof shows imitation of expert's reward and solutions in multi-objective optimization.
problem Multi-objective optimization with reward and solution imitation.
method Wasserstein inverse reinforcement learning.
result Wasserstein inverse reinforcement learning enables imitation of expert's reward and solutions in multi-objective optimization.
Paper proposes IRL methods for limited interaction scenarios.
problem Learning with limited teacher interaction.
method Curriculum Inverse Reinforcement Learning (CIRL) and Self-Paced Inverse Reinforcement Learning (SPIRL).
result Training strategies can accelerate learning compared to random or batch methods.
This work simplifies IRL by using potential-based reward shaping.
problem Computational inefficiency in inverse reinforcement learning.
method Potential-based reward shaping to reduce RL sub-problems.
result Reduces computational burden of inverse reinforcement learning.
Unified view of federated learning and distributed RL using local stochastic approximation.
problem Finding the root of an operator composed of local operators in a network of agents with dependent data.
method Local stochastic approximation over a network of agents with Markov process-dependent data.
result Convergence rates of local stochastic approximation for both constant and time-varying step sizes, within a logarithmic factor of independent data.
Develops a new method to improve performance in multi-objective learning problems.
problem Gradient bias in multi-objective learning leading to degraded performance.
method Stochastic Multi-objective gradient Correction (MoCo) method that guarantees convergence without increasing batch size.
result Demonstrates effectiveness of MoCo method in simulations on multi-task learning.
Most reinforcement learning algorithms are inefficient for learning multiple tasks in complex robotic systems, where different tasks share a set of actions. In such environments a compound policy may be learnt with shared neural network parameters, which performs multiple tasks concurrently. However such compound polic…
We state the problem of inverse reinforcement learning in terms of preference elicitation, resulting in a principled (Bayesian) statistical formulation. This generalises previous work on Bayesian inverse reinforcement learning and allows us to obtain a posterior distribution on the agent's preferences, policy and optio…
Develops statistical framework for resolving reward function ambiguity in inverse reinforcement learning.
problem Non-uniqueness of reward functions in inverse reinforcement learning.
method Entropy regularization combined with least-squares reconstruction of the reward from the soft Bellman residual.
result Least-squares reward function is unique and consistent with the expert policy.
REFIL learns from imagined sub-group interactions to improve multi-agent reinforcement learning.
problem Learning efficiency in multi-agent reinforcement learning settings.
method Randomized Entity-wise Factorization (REFIL) approach.
result REFIL outperforms all strong baselines in challenging multi-task StarCraft settings.
Combines human and AI to optimize fund managers' investment decisions.
problem Improving fund managers' investment practices.
method Combines Inverse Reinforcement Learning and Reinforcement Learning.
result Improves fund managers' investment performance.
AIRL learns robust, generalizable reward functions from demonstrations.
problem Learning robust reward functions from demonstrations for changing environments.
method Adversarial Inverse Reinforcement Learning (AIRL) with hierarchical disentangled rewards.
result Generalizable policies and comparable results to state-of-the-art methods.
Modeling driver trajectories using inverse reinforcement learning and random utility.
problem Modeling rational driver behavior in road networks from sparse sensor data.
method Apply random utility theory to model unknown reward function, introduce extended state, and use Markov decision process.
result Maximum entropy inverse reinforcement learning is a special case of the proposed approach.
Robots learn new tasks autonomously with minimal human intervention.
problem Lack of scalable data collection for robot learning.
method Multi-task imitation learning with autonomous data collection and one-shot generalization.
result Robots can continuously improve through autonomous data collection without reinforcement learning.
Proposes a new distance metric for multi-marginal optimal transport.
problem Computational scalability in multi-marginal optimal transport.
method Random one-dimensional projections to construct sliced multi-marginal Wasserstein distance.
result Sliced multi-marginal Wasserstein distance is a metric with dimension-free sample complexity.
Improves imitation learning in RL by learning reward function efficiently.
problem Lack of effective reward function approximation in AIRL for imitation tasks.
method Proposes Off-Policy AIRL that combines adversarial learning with efficient reward function approximation.
result Shows superior imitation performance and efficiency compared to state-of-the-art AIL algorithms.
TOPPO improves PPO for MTRL by balancing critic gradients, outperforming SAC.
problem Critic-side gradient ill-conditioning in PPO for MTRL.
method Critic Balancing modules to improve gradient conditioning and balance task updates.
result TOPPO achieves stronger mean and tail-task performance than SAC-family and ARS-family baselines.
This paper sets a lower bound for sample complexity in inverse reinforcement learning.
problem Finding a reward function that generates a desired optimal policy in MDPs.
method Information-theoretic lower bound using geometric construction and Fano's inequality.
result An O(nlogn) sample complexity lower bound for IRL problems. Inverse reinforcement learning (IRL) is the problem of finding a reward function that generates a given optimal policy for a given Markov Decision Process. This paper looks at an algorithmic-independent geometric analysis of the IRL problem with finite states and actions. A L1-regularized Support Vector Machine formula…
We tackle the Multi-task Batch Reinforcement Learning problem. Given multiple datasets collected from different tasks, we train a multi-task policy to perform well in unseen tasks sampled from the same distribution. The task identities of the unseen tasks are not provided. To perform well, the policy must infer the tas…
GDT improves reinforcement learning by matching future state information efficiently.
problem Efficient learning of multi-task policies from trajectory data.
method Generalized Decision Transformer (GDT) for offline hindsight information matching.
result GDT enables effective offline multi-task state-marginal matching and imitation learning.
The design of a reward function often poses a major practical challenge to real-world applications of reinforcement learning. Approaches such as inverse reinforcement learning attempt to overcome this challenge, but require expert demonstrations, which can be difficult or expensive to obtain in practice. We propose var…
Reinforcement learning in complex environments is a challenging problem. In particular, the success of reinforcement learning algorithms depends on a well-designed reward function. Inverse reinforcement learning (IRL) solves the problem of recovering reward functions from expert demonstrations. In this paper, we solve …
New approach uses inverse reinforcement learning to improve language model training.
problem Training large language models using imitation learning methods.
method Developed a new method of inverse reinforcement learning to optimize sequences directly.
result IRL-based fine-tuning leads to better performance and diversity in language generation.
We address the problem of inverse reinforcement learning in Markov decision processes where the agent is risk-sensitive. In particular, we model risk-sensitivity in a reinforcement learning framework by making use of models of human decision-making having their origins in behavioral psychology, behavioral economics, an…
Augmenting reinforcement learning with imitation learning is often hailed as a method by which to improve upon learning from scratch. However, most existing methods for integrating these two techniques are subject to several strong assumptions---chief among them that information about demonstrator actions is available.…
MixTS uses a mixture prior to analyze Thompson Sampling in multi-task learning.
problem Analyzing Thompson Sampling in environments with uncertain and multi-class problems.
method Developed MixTS by incorporating a mixture prior into Thompson Sampling and using a novel proof technique for mixture distributions.
result Proved Bayes regret bounds for MixTS in linear bandits and finite-horizon reinforcement learning.
The reinforcement learning community has made great strides in designing algorithms capable of exceeding human performance on specific tasks. These algorithms are mostly trained one task at the time, each new task requiring to train a brand new agent instance. This means the learning algorithm is general, but each solu…
New algorithm reduces performance loss in IRL with mismatched transition dynamics.
problem Performance degradation in inverse reinforcement learning due to mismatched transition dynamics.
method Proposed a robust Maximum Causal Entropy (MCE) IRL algorithm leveraging robust reinforcement learning insights.
result Empirically demonstrated stable performance improvement under transition dynamics mismatches.
Proposes new stochastic algorithms for multi-objective optimization.
problem Multi-objective optimization in machine learning problems.
method Direction-oriented multi-objective formulation and Stochastic Direction-oriented Multi-objective Gradient descent (SDMGrad).
result Stochastic algorithms converge to Pareto stationary points with improved complexities.
New approach transfers rewards learned in one environment to reinforcement learning in a new environment.
problem Transfer of rewards learned using inverse reinforcement learning from one environment to a new, different environment.
method Formulate the problem as a joint system of Bellman equations, develop minimax estimators for the target soft-q-function, solve the source and target system of equations jointly. result The coupled approach removes the first-order influence of source Bellman residual error compared to the sequential approach.