Many continuous control tasks have bounded action spaces. When policy gradient methods are applied to such tasks, out-of-bound actions need to be clipped before execution, while policies are usually optimized as if the actions are not clipped. We propose a policy gradient estimator that exploits the knowledge of action…
On-device research index
arXiv research
A locally-built, LLM-digested index of recent arXiv papers in quant finance, geometry/topology, and statistical ML — keyword search served straight from SQLite on this machine.
168,657 papers · 148 categories
Trend · papers per month
48 results for “action estimation”
Proposes MDR estimator for unbiased OPE with large action spaces.
problem Severe bias and variance tradeoffs in OPE with large action spaces.
method Marginalized Doubly Robust (MDR) estimator, reducing variance and bias.
result MDR estimator is unbiased under weaker assumptions than MIPS.
New OPE estimator improves offline policy evaluation for large action spaces.
problem Existing OPE estimators fail with large action spaces, leading to extreme bias and variance.
method Proposes a new estimator using marginalized importance weights and action embeddings.
result Empirical performance improvement enables reliable OPE even with many actions.
A new method learns action representations for reinforcement learning.
problem Efficient action-value estimation in reinforcement learning.
method Action hypergraph networks framework for learning action representations.
result Hypergraph Q-networks show effectiveness on various domains.
New method reduces bias and variance in OPE for large action spaces.
problem High bias and variance in OPE for large, combinatorial action spaces.
method Factored action spaces and decomposed importance sampling.
result Decomposed IS estimators have less variance than non-decomposed versions.
PQR estimates reward functions from actions and states without assuming state-only rewards.
problem Estimating reward functions from actions and states without state-only assumptions.
method Deep learning approach that sequentially estimates policy, Q-function, and reward.
result PQR uniquely recovers true reward with known transitions and bounds error with unknown transitions.
New estimator improves off-policy evaluation for large action spaces.
problem Conventional importance-weighting approaches suffer from excessive variance in off-policy evaluation for large discrete action spaces.
method Proposes OffCEM estimator based on conjunct effect model (CEM), applying importance weighting only to action clusters and using model-based reward estimation for residual effects.
result Proposed estimator is unbiased under local correctness condition, providing substantial improvements in OPE especially with many actions.
Reinforcement learning (RL) in discrete action space is ubiquitous in real-world applications, but its complexity grows exponentially with the action-space dimension, making it challenging to apply existing on-policy gradient based deep RL algorithms efficiently. To effectively operate in multidimensional discrete acti…
Estimation of importance sampling weights for off-policy evaluation of contextual bandits often results in imbalance - a mismatch between the desired and the actual distribution of state-action pairs after weighting. In this work we present balanced off-policy evaluation (B-OPE), a generic method for estimating weights…
New findings show optimization is crucial for OPL in large action spaces.
problem Challenges in optimizing policies for large action spaces in offline contextual bandits.
method Weighed log-likelihood objectives and estimator-aware policy parametrization.
result Simple weighted log-likelihood objectives enjoy better optimization properties and recover competitive policies.
In this paper, we consider a best action identification problem in the stochastic linear bandit setup with a fixed confident constraint. In the considered best action identification problem, instead of minimizing the accumulative regret as done in existing works, the learner aims to obtain an accurate estimate of the u…
Paper tackles RL with continuous actions and unmeasured confounders.
problem Offline policy learning with continuous actions and unmeasured confounders.
method Developed a novel identification result and a minimax estimator for nonparametric policy value estimation.
result Introduced a policy-gradient-based algorithm to identify the optimal policy.
In complex tasks, such as those with large combinatorial action spaces, random exploration may be too inefficient to achieve meaningful learning progress. In this work, we use a curriculum of progressively growing action spaces to accelerate learning. We assume the environment is out of our control, but that the agent …
CAEL-MIPS learns embeddings to improve MIPS for better OPE in contextual bandits.
problem High variance in IPS weighting for OPE in large action spaces.
method Context-Action Embedding Learning (CAEL) for MIPS to minimize MSE.
result CAEL-MIPS outperforms baselines in MSE for OPE in contextual bandits.
Hidden Markov Model Estimation-Based Q-learning for Partially Observable Markov Decision Processcs.LG
The objective is to study an on-line Hidden Markov model (HMM) estimation-based Q-learning algorithm for partially observable Markov decision process (POMDP) on finite state and action sets. When the full state observation is available, Q-learning finds the optimal action-value function given the current action (Q func…
In recent years, deep reinforcement learning has been shown to be adept at solving sequential decision processes with high-dimensional state spaces such as in the Atari games. Many reinforcement learning problems, however, involve high-dimensional discrete action spaces as well as high-dimensional state spaces. This pa…
The paper explores proper actions and their relation to representation theory, with new quantitative methods.
problem Understanding proper actions and their connection to representation theory.
method Geometric criteria, sharpness measure, and dynamical volume estimates.
result New quantitative methods have established temperedness criteria for unitary representations.
The paper tackles counterfactual learning for stochastic policies with continuous actions.
problem Learning stochastic policies with continuous actions from logged data.
method Introduces a joint kernel embedding of contexts and actions to model continuous actions, and uses proximal point algorithms and smooth estimators for optimization.
result Demonstrates the benefits of using proximal point algorithms and smooth estimators for counterfactual learning.
ESAC improves reinforcement learning by lookahead and intuition.
problem Designing efficient reinforcement learning architectures for intelligent agents.
method Integrates selector, tuner, model-learner, estimator, and lookahead into ESAC.
result ESAC outperforms other architectures in optimizing policies.
New method QMLE performs well in complex action spaces without policy gradients.
problem Why policy gradients outperform action-value methods in complex action spaces.
method QMLE framework for action-value methods based on three principles.
result QMLE performs comparably to policy gradient methods in complex action spaces.
Deep Reinforcement Learning improves with Weighted Q-Learning to reduce bias and uncertainty.
problem Overestimation and high variance in Q-Learning cause learning algorithms to diverge in complex environments.
method Deep Weighted Q-Learning (Deep WQL) uses Dropout and Monte Carlo sampling to approximate WQL's weights and reduce bias.
result Deep WQL reduces bias and improves performance on benchmarks compared to existing methods.
To address the challenge of backpropagating the gradient through categorical variables, we propose the augment-REINFORCE-swap-merge (ARSM) gradient estimator that is unbiased and has low variance. ARSM first uses variable augmentation, REINFORCE, and Rao-Blackwellization to re-express the gradient as an expectation und…
Mean Actor Criticstat.ML
We propose a new algorithm, Mean Actor-Critic (MAC), for discrete-action continuous-state reinforcement learning. MAC is a policy gradient algorithm that uses the agent's explicit representation of all action values to estimate the gradient of the policy, rather than using only the actions that were actually executed. …
Improved TD learning reduces batch sampling error.
problem TD learning can be inaccurate in batch settings due to action weighting.
method Introduced PSEC-TD(0) to correct action weighting using importance sampling.
result PSEC-TD(0) produces more accurate value function estimates.
Many complex domains, such as robotics control and real-time strategy (RTS) games, require an agent to learn a continuous control. In the former, an agent learns a policy over and in the latter, over a discrete set of actions each of which is parametrized by a continuous parameter. Such problems are natu…
While often stated as an instance of the likelihood ratio trick [Rubinstein, 1989], the original policy gradient theorem [Sutton, 1999] involves an integral over the action space. When this integral can be computed, the resulting "all-action" estimator [Sutton, 2001] provides a conditioning effect [Bratley, 1987] reduc…
We consider the problem of learning the optimal action-value function in the discounted-reward Markov decision processes (MDPs). We prove a new PAC bound on the sample-complexity of model-based value iteration algorithm in the presence of the generative model, which indicates that for an MDP with N state-action pairs a…
This paper improves MADDPG's performance in discrete grid-world scenarios.
problem MADDPG struggles with discrete action spaces in grid-world tasks.
method Integrates alternative Gumbel-Softmax estimators into MADDPG.
result One alternative estimator significantly improves performance in grid-world tasks.
Linear disentangled representations improve unsupervised action estimation.
problem Learning linear disentangled representations for unsupervised action estimation.
method Developed a method to induce irreducible representations in VAE models without labeled action sequences.
result Linear disentangled representations are a desirable property for unsupervised action estimation.
A new method, Count-MORL, improves offline reinforcement learning by using state-action frequency.
problem Improving offline reinforcement learning performance.
method Integrates count-based conservatism into model-based offline reinforcement learning.
result The learned policy is near-optimal and outperforms existing methods.
A new method recovers rewards from behavior policies using classification and regression.
problem Recovering meaningful rewards from observed behavior in reinforcement learning.
method GenPQR, a modular procedure that estimates behavior policy, evaluates soft Q-function, and recovers normalized reward using classification and regression.
result GenPQR matches or improves reward recovery compared to DeepPQR, while being simpler and more modular.
New method reduces bias in learning from large action spaces using selective importance sampling.
problem Learning from large-scale recommendation systems with bandit feedback and supervised labels.
method Selective Importance Sampling (sIS) and Policy Optimization for eXtreme Models (POXM) algorithm.
result POXM method significantly outperforms existing methods in learning from bandit feedback on XMC tasks.
We consider a purely algebraic result. Then given a circle or cyclic group of prime order action on a manifold, we will use it to estimate the lower bound of the number of fixed points. We also give an obstruction to the existence of action on manifolds with isolated fixed points when is a prime.
Estimates distances between fixed points of certain Teichmüller actions.
problem Estimating distances between fixed points of specific Teichmüller actions.
method Uses a pants decomposition and quasi-isometry to estimate distances.
result Provides a method to estimate distances between fixed points.
A new RL algorithm POWR learns world models to estimate action-values.
problem Inaccessibility of explicit action-value functions in RL.
method Learning a world model using conditional mean embeddings and deriving action-value function via matrix operations.
result POWR algorithm converges to global optimum with proven rates.
New algorithm estimates treatment effects for more efficient contextual bandits.
problem Contextual bandits struggle with action-independent reward redundancies.
method Reduces contextual bandits to heterogeneous treatment effect estimation.
result Heterogeneous treatment effect estimation leads to more efficient model estimation.
We consider off-policy evaluation and optimization with continuous action spaces. We focus on observational data where the data collection policy is unknown and needs to be estimated. We take a semi-parametric approach where the value function takes a known parametric form in the treatment, but we are agnostic on how i…
Decoupled Q-Chunkingcs.LG
Decouples critic chunk length from policy to improve policy reactivity and performance.
problem Bootstrapping bias and difficulty in extracting optimal policies from chunked critics.
method Optimizes policy against a distilled critic for partial action chunks, allowing shorter chunks for policy.
result Reliably outperforms prior methods on long-horizon offline goal-conditioned tasks.
Regularizes predictions to encourage beneficial user actions.
problem Learning models affect user actions, but provide no guarantees.
method Introduces look-ahead regularization to anticipate and encourage beneficial actions.
result Effective in promoting beneficial user actions in real and synthetic data.
New method minimizes decision errors in large treatment spaces.
problem Improving decision-making in large treatment spaces with biased observational data.
method Loss minimizes classification error of actions in large action space.
result Proves improved decision-making performance in large combinatorial action spaces.
We consider off-policy policy evaluation when the trajectory data are generated by multiple behavior policies. Recent work has shown the key role played by the state or state-action stationary distribution corrections in the infinite horizon context for off-policy policy evaluation. We propose estimated mixture policy …
POTEC tackles off-policy learning in large action spaces, improving effectiveness.
problem Existing OPL methods fail in large discrete action spaces due to bias or variance issues.
method Two-stage algorithm: cluster selection via policy-based approach, action selection via regression-based approach.
result POTEC provides substantial improvements in off-policy learning effectiveness, especially in large and structured action spaces.
In traditional reinforcement learning, an agent maximizes the reward collected during its interaction with the environment by approximating the optimal policy through the estimation of value functions. Typically, given a state s and action a, the corresponding value is the expected discounted sum of rewards. The optima…
Q-Distribution Guided Q-Learning corrects overestimation of uncertain OOD actions in offline RL.
problem Overestimation of Q-values for out-of-distribution actions in offline reinforcement learning.
method QDQ applies a pessimistic adjustment to Q-values in uncertain OOD regions based on a consistency model.
result QDQ improves performance on the D4RL benchmark and achieves significant improvements across many tasks.
New methods use vector search and nearest-neighbor matching for policy learning in causal inference.
problem Learning optimal policies in causal inference with limited data.
method RAG-based policy learning with vector search and nearest-neighbor matching.
result The methods bound the within-candidate choice regret and evaluate the one-step method directly as a policy.
ZOSPI improves RL policies with global value function exploitation.
problem Limited sample efficiency of PG algorithms in RL.
method Zeroth-order supervised policy improvement, leveraging global value function estimation.
result ZOSPI achieves competitive results with remarkable sample efficiency.
Uplifting Banditsstat.ML
UCB algorithms estimate uplifts in multi-variable reward systems.
problem Marketing campaigns and recommender systems with multiple outcome variables.
method UCB-style algorithms for estimating uplifts over a baseline.
result Sublinear regret bounds for multiple problem variants.
We propose expected policy gradients (EPG), which unify stochastic policy gradients (SPG) and deterministic policy gradients (DPG) for reinforcement learning. Inspired by expected sarsa, EPG integrates (or sums) across actions when estimating the gradient, instead of relying only on the action in the sampled trajectory…