This work improves Hindsight Learning for goal-directed tasks in reinforcement learning.
problem Sparse reward problems in reinforcement learning, especially goal-directed tasks.
method Improves Hindsight Experience Replay and Hindsight Policy Gradients for robotic tasks.
result Improved learning in Hindsight Experience Replay and Hindsight Policy Gradients.
PCHID improves sample efficiency in reinforcement learning tasks.
problem Sparse rewards make learning policies difficult in reinforcement learning.
method Hindsight Inverse Dynamics with Hindsight Experience Replay and Policy Continuation.
result PCHID significantly improves sample efficiency and final performance on multi-goal tasks.
HTRPO tackles sparse rewards in RL with improved stability and performance.
problem Sparse rewards in reinforcement learning.
method HTRPO extends TRPO with hindsight and QKL for better policy update stability.
result HTRPO consistently outperforms TRPO and HPG in various sparse reward tasks.
CDP improves RL performance and sample-efficiency by prioritizing rare goal states.
problem Learning from imbalanced data in RL.
method Curiosity-Driven Prioritization (CDP) framework.
result CDP improves both performance and sample-efficiency of RL agents.
HGAIL learns policies without expert demonstrations.
problem Lack of expert demonstrations in imitation learning.
method Combines hindsight and GAIL to learn policies.
result Comparable performance to current methods, with curriculum learning.
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.
New approach generates optimal disturbances for controller verification.
problem Optimizing disturbances for controller verification with blackbox access.
method Online learning approach that adaptively generates disturbances based on controller inputs.
result New algorithm (MOTR) outperforms existing methods in simulated examples.
HO2 learns options from data efficiently, improving robot manipulation tasks.
problem Learning options from raw pixel inputs in 3D robot manipulation tasks.
method HO2 infers likely option choices and trains all policy components off-policy.
result HO2 outperforms existing methods on 3D robot manipulation tasks.
Reinforcement Learning (RL) algorithms can suffer from poor sample efficiency when rewards are delayed and sparse. We introduce a solution that enables agents to learn temporally extended actions at multiple levels of abstraction in a sample efficient and automated fashion. Our approach combines universal value functio…
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.
A new EM framework for goal-conditioned RL improves performance on sparse reward tasks.
problem Handling sparse rewards in goal-conditioned reinforcement learning.
method A graphical model framework with an EM algorithm that includes a learning-in-hindsight E-step and a supervised M-step.
result hEM significantly outperforms model-free baselines on goal-conditioned benchmarks with sparse rewards.
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.
HL algorithms improve resource allocation in cloud environments.
problem Sequential decision-making under uncertainty with exogenous variables.
method HL algorithms leverage exogenous variable samples to infer counterfactual consequences.
result HL algorithms outperform classic methods and reinforcement learning in resource allocation.
Improved Q-learning with hindsight factor reduces overestimation errors.
problem Overestimation errors in Q-learning leading to suboptimal policies.
method Introducing a hindsight factor as an additional loss term in Q-learning.
result The hindsight factor significantly reduces overestimation errors and improves stability.
Paper introduces a new G ⋆ G^\star G ⋆ regret measure for online convex optimization with smooth losses.
problem Online convex optimization with smooth losses.
method Introduces a new G ⋆ G^\star G ⋆ regret measure that depends on the cumulative squared gradient norm. result The G ⋆ G^\star G ⋆ regret can be arbitrarily sharper than existing measures when losses have vanishing curvature. Extends Hindsight Experience Replay to learn from multiple reward functions.
problem Learning policies for multiple reward functions without trial runs.
method Develops a method to use a reward function to calculate rewards for states not encountered, and learns policies for multiple reward functions.
result A single policy can generalize across all linear combinations of multi-objective rewards.
A novel RL objective and prioritization framework improve performance and sample-efficiency in multi-goal tasks.
problem Learning diverse goals in multi-goal reinforcement learning.
method Maximum entropy regularization for objective and prioritization framework.
result Promising improvements in performance and sample-efficiency on multi-goal robotic tasks.
Aims to optimize goal sampling in deep reinforcement learning.
problem The sampling of goals affects deep reinforcement learning performance.
method Curriculum goal masking method to focus on medium difficulty goals.
result Focusing on medium difficulty goals leads to better learning performance.
HIGhER uses language to generate new instructions for better learning from mistakes.
problem Improving instruction following in reinforcement learning environments.
method Hindsight Generation for Experience Replay (HIGhER) that learns from mistakes and relabels episodes.
result HIGhER enhances instruction following in reinforcement learning environments.
A contextual bandit method evaluates and improves inventory control policies.
problem Evaluating and improving periodic review inventory control policies with nonstationary demand.
method Contextual bandit-based algorithm to evaluate and tweak policies.
result The method achieves favorable guarantees in both theory and practice.
In this paper, we consider an online optimization process, where the objective functions are not convex (nor concave) but instead belong to a broad class of continuous submodular functions. We first propose a variant of the Frank-Wolfe algorithm that has access to the full gradient of the objective functions. We show t…
New algorithms learn POMDPs efficiently with hindsight observability.
problem Hardness of learning in POMDPs due to partial observability.
method Hindsight Observable Markov Decision Process (HOMDP) and new algorithms for tabular and function approximation settings.
result Sample-efficient learning in POMDPs with optimal dependence on latent state and observation cardinalities.
New method combines curiosity and hindsight for stacking blocks.
problem Sparse rewards in reinforcement learning.
method Curiosity-driven exploration combined with hindsight and curriculum learning.
result First to stack more than two blocks using only sparse reward.
Develops a new approach to learn value predictors from data.
problem Learning effective value predictors from data in reinforcement learning.
method Combines model learning and model-free methods to learn what to model.
result Significantly improves value prediction in simple and complex domains.
We optimize rebalancing options by limiting asset allocations to a few choices, reducing the price and guaranteeing near-optimal performance.
problem Optimizing rebalancing strategies under discrete hindsight optimization.
method Restricting the set of rebalancing rules to a small number of asset allocations.
result Guaranteed near-optimal performance with a rock-bottom option price.
Framework uses hindsight regret to audit marketing budget allocations.
problem Lack of principled way to assess strategic budget allocations.
method Hindsight regret framework based on constraint-faithful benchmark.
result Identifies practical trade-off between allocation flexibility and detectability.
New algorithms assign credit to past decisions based on hindsight.
problem Efficient credit assignment in reinforcement learning.
method Explicit credit assignment using new data in hindsight.
result Value functions can be rewritten to include hindsight credit assignment.
Adaptive TD learning reduces bias in policy evaluation by switching between TD and MC methods.
problem Achieving accurate policy evaluation with Temporal Difference (TD) learning in the presence of state-specific uncertainty.
method Adaptive switching between TD and Monte Carlo (MC) methods, using learned confidence intervals to detect and mitigate bias.
result The proposed adaptive algorithm outperforms existing methods in policy evaluation tasks.
Efficient RL in partially observable risk-sensitive environments with hindsight observations.
problem Risk-sensitive reinforcement learning in partially observable environments.
method Integrates hindsight observations into POMDP framework, develops novel RL algorithm.
result Achieves polynomial regret with provable efficiency, outperforming existing methods.
Efficient algorithm controls unknown systems with adversarial perturbations.
problem Controlling unknown linear systems with adversarial perturbations and convex losses.
method Measures regret against an optimal linear policy, provides efficient algorithm with sublinear regret bound.
result First efficient algorithm with sublinear regret bound of T^{2/3}.
Algorithm maximizes wealth from best pairs rebalancing rule in hindsight.
problem Maximizing wealth from best pairs rebalancing rule in hindsight.
method Extends Ordentlich and Cover's max-min universal portfolio to achieve a percentage of the hindsight-optimized wealth.
result Achieves a compound-annual growth rate arbitrarily close to the best pairs rebalancing rule in hindsight.
HGG generates goals to improve sample efficiency in robotic tasks.
problem Efficiency in reinforcement learning with sparse reward signals.
method Generates valuable hindsight goals for reinforcement learning.
result Significantly improved sample efficiency over HER.
Interactive learning with hindsight instruction feedback achieves better performance than traditional methods.
problem Learning with expert supervision of optimal responses is often impractical or requires expert knowledge.
method Study of interactive learning with hindsight instruction feedback, introducing LORIL algorithm.
result LORIL algorithm achieves T \sqrt{T} T regret, improving over baselines in domains with low-rank instruction-response distribution. Study on learning strategies in adaptive Markov games with policy regret as metric.
problem Learning in dynamic Markov games with adaptive opponents is challenging.
method Introduced policy regret as a new learning metric and developed algorithms for consistent adaptive adversaries.
result Achieved T \sqrt{T} T policy regret against certain adaptive adversaries. Dynamical-VAE learns causal dynamics from POMDPs using future information.
problem Learning accurate state representations from partial observations in POMDPs.
method Dynamical Variational Auto-Encoder (DVAE) with hindsight framework.
result DVAE uncovers causal graph more effectively than history-based methods.
New method uses hindsight to make exploration robust in stochastic environments.
problem Exploration in sparse-reward or reward-free environments, especially in stochastic settings.
method Learn representations of the future that capture unpredictable aspects, using them to predict and reward only the predictable parts of the world.
result Improves exploration in Atari games and Montezuma's Revenge, robust to stochasticity.
Policy gradient aims to maximize expected return using gradient ascent.
problem Finding a policy that maximizes expected return in a given class of policies.
method Gradient ascent applied to a differentiable model of the policy, estimating the gradient of expected return.
result Policy gradient methods require on-policy data for gradient estimation, limiting sample efficiency.
This paper prices and replicates the best continuously-rebalanced portfolio in hindsight.
problem Deriving the price of a financial derivative based on the best continuously-rebalanced portfolio in hindsight.
method Analyzing the best continuously-rebalanced portfolio in hindsight for a single-stock Black-Scholes market and a general market with correlated stocks.
result The replicating strategy compounds wealth at the same asymptotic rate as the best levered rebalancing rule in hindsight, beating the market asymptotically.
Off-policy stochastic actor-critic methods rely on approximating the stochastic policy gradient in order to derive an optimal policy. One may also derive the optimal policy by approximating the action-value gradient. The use of action-value gradients is desirable as policy improvement occurs along the direction of stee…
ARM policy gradient reduces variance for binary actions.
problem High variance in policy gradients for binary actions.
method Augment-Reinforce-Merge (ARM) policy gradient estimator.
result ARM estimator achieves significant variance reduction and faster convergence.
A new theorem and algorithm solve off-policy policy gradient problems.
problem Solving the theoretical gap in off-policy policy gradient methods.
method Introduced an off-policy policy gradient theorem using emphatic weightings and developed the ACE algorithm.
result Demonstrated ACE finds the optimal solution in off-policy learning, unlike previous methods.
DE via conjugate policies improves exploration and policy performance.
problem Effective exploration in policy gradient methods.
method DE via conjugate policies.
result DE improves policy performance and exploration effectiveness.
Faster policy learning via continuous-time gradients.
problem Efficiently estimating policy gradients for continuous-time systems.
method Approximating continuous-time gradients directly, using adaptive discretization.
result More efficient policy gradient estimator leads to faster learning.
STORM-PG uses momentum for faster policy gradient updates.
problem Improving policy gradient methods for reinforcement learning.
method Introduces STORM-PG, a SARAH-based algorithm with exponential moving average.
result Achieves O ( 1 / ε 3 ) O(1/ε^3) O ( 1/ ε 3 ) sample complexity, matching best-known rate. Paper explains DRL strategies for portfolio management using linear models.
problem Difficulty in understanding DRL-based trading strategies.
method Empirical approach using linear models and integrated gradients.
result DRL agents show stronger multi-step prediction power than machine learning methods.
Paper develops online learning algorithms for quaternion ARMA models.
problem Adaptive learning for autoregressive moving average (ARMA) models in quaternion domain.
method Transformed learning problem into full information optimization task, solved using gradient descent and Newton's method.
result Online algorithms achieve asymptotic performance approaching best ARMA model.
Optimizes Thompson sampling policies using policy gradient methods.
problem Improving Thompson sampling in bandit problems.
method Applies policy gradient algorithms to optimize Thompson sampling policies.
result Direct policy search on Thompson sampling improves performance.
Paper tackles efficient policy gradient estimation from off-policy data.
problem Estimating policy gradients from off-policy data is challenging and inefficient.
method Derives asymptotic lower bounds, proposes a meta-algorithm with 3-way robustness, and establishes convergence guarantees.
result Meta-algorithm achieves the lower bound on mean-squared error without parametric assumptions.