PGS uses neural networks to improve policies online without search trees.
problem Limited scalability of Monte Carlo Tree Search (MCTS) for high branching factor games.
method Adapts a neural network simulation policy via policy gradient updates, avoiding search trees.
result PGS achieves comparable performance to MCTS and defeats strong Hex agents.
Trust-region methods have yielded state-of-the-art results in policy search. A common approach is to use KL-divergence to bound the region of trust resulting in a natural gradient policy update. We show that the natural gradient and trust region optimization are equivalent if we use the natural parameterization of a st…
Paper proposes MCTSPO for better reinforcement learning policy optimization.
problem Local optima and saddle points in gradient-based methods and poor initialization in gradient-free methods.
method Monte-Carlo tree search combined with gradient-free optimization.
result Improved performance on reinforcement learning tasks with deceptive or sparse reward functions.
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.
Introduces GAMPS for better model-based policy learning.
problem Misspecified model classes lead to poor policy estimates.
method Exploits current policy to learn approximate transition model, focusing on relevant parts of the environment.
result Empirically validated GAMPS on benchmark domains, demonstrating improved properties.
Bayesian optimization improves policy search in reinforcement learning.
problem Finding optimal policies with high variance estimates from random samples.
method Develops an algorithm combining Bayesian optimization and policy gradients.
result Improves sample complexity and reduces variance in empirical evaluations.
New approach learns walk and trot gaits from simulated quadruped using strategic exploration.
problem Learning symmetric gaits (walk and trot) from high-dimensional action spaces.
method Introduced symmetry properties into initial covariance of Gaussian search distribution for strategic exploration. Used episode-based likelihood ratio policy gradient and relative entropy policy search.
result Significant performance enhancement in learning walk and trot gaits compared to random gaits.
Paper presents a new policy gradient theorem using weak derivatives for reinforcement learning.
problem Continuous state-action reinforcement learning problems.
method Introduced an alternative policy gradient theorem using weak derivatives.
result The new approach yields algorithms that converge almost surely to stationary points of the value function.
The paper proposes a new policy training objective to reduce exploration in self-play.
problem Training policies to mimic MCTS search behavior can lead to excessive exploration.
method Derive a policy gradient expression using MCTS value estimates to reduce exploration.
result Empirically evaluated policies show reduced exploration compared to MCTS-based training.
Previously, the exploding gradient problem has been explained to be central in deep learning and model-based reinforcement learning, because it causes numerical issues and instability in optimization. Our experiments in model-based reinforcement learning imply that the problem is not just a numerical issue, but it may …
New actor-critic method reduces sample complexity for reinforcement learning.
problem Improving sample complexity for actor-critic algorithms in reinforcement learning.
method Integrates Monte Carlo rollouts into policy search steps for better control over bias.
result Established sample complexity for actor-critic algorithms with policy gradient.
Direct policy gradients optimize policies in discrete action spaces using sampling.
problem Optimizing policies in discrete action spaces with direct methods.
method Combining direct optimization and A⋆ sampling for policy gradient approximation. result DirPG algorithms can incorporate domain knowledge and have higher probability of sampling informative gradients.
Transformer learns to search through reinforcement learning, mimicking DFS.
problem Understanding how transformers learn search capabilities in RL.
method Two-head transformer, depth-wise curriculum, discounted returns.
result Transformer policy generalizes depth and prioritizes high-probability branches.
Enhances RL performance with a population-guided parallel learning scheme.
problem Improving off-policy reinforcement learning performance.
method Population-guided parallel learning scheme with shared experience replay buffer and soft policy update.
result Monotone improvement of the expected cumulative return proved theoretically and demonstrated in practice.
The policy gradient approach is a flexible and powerful reinforcement learning method particularly for problems with continuous actions such as robot control. A common challenge in this scenario is how to reduce the variance of policy gradient estimates for reliable policy updates. In this paper, we combine the followi…
Credit assignment in Meta-reinforcement learning (Meta-RL) is still poorly understood. Existing methods either neglect credit assignment to pre-adaptation behavior or implement it naively. This leads to poor sample-efficiency during meta-training as well as ineffective task identification strategies. This paper provide…
Paper uses CMAB to improve NAS efficiency and accuracy.
problem Improving efficiency and accuracy of NAS for DNNs.
method Formulated NAS as CMAB, used Nested Monte-Carlo Search.
result Discovered cell structure achieves comparable accuracy to state-of-the-art, 20x faster.
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. …
Learning policies on data synthesized by models can in principle quench the thirst of reinforcement learning algorithms for large amounts of real experience, which is often costly to acquire. However, simulating plausible experience de novo is a hard problem for many complex environments, often resulting in biases for …
A new method reduces variance in training early-stage rankers for large-scale search systems.
problem Training early-stage rankers for large-scale search systems is challenging due to exploding variance in policy gradient methods.
method Proposes credit-assigned policy gradient (CA-PG) to mitigate variance in training early-stage rankers.
result CA-PG significantly reduces variance in training early-stage rankers compared to vanilla policy gradient.
Improves neural network search in combinatorial spaces of mathematical symbols.
problem Early commitment and initialization bias limit exploration in neural network search.
method Entropy regularization and distribution initialization methods.
result Improves performance, increases sample efficiency, lowers solution complexity.
Paper proposes a policy-search algorithm to learn entropy-maximizing exploration policies in reward-free environments.
problem Reward-free learning in high-dimensional, continuous-control domains.
method Maximum Entropy POLicy optimization (MEPOL) algorithm that maximizes a non-parametric state entropy estimate.
result MEPOL learns a maximum-entropy exploration policy that facilitates learning various reward-based tasks.
New algorithm converges to optimal filter for predicting linear dynamical systems.
problem Direct policy search for optimal dynamic filters in partially observable systems.
method Regularizer enforcing informativity over filter states.
result Gradient descent converges to globally optimal solution at rate O(1/T).
Paper speeds up policy optimization for large recommendation systems.
problem Offline optimization of large-scale recommendation systems is computationally expensive.
method Derives an approximation of policy learning algorithms that scales logarithmically with the catalogue size.
result Our algorithm is an order of magnitude faster than naive approaches while producing equally good policies.
A new method improves stability in policy learning for continuous control tasks.
problem Stability issues in policy gradient methods when policies are close to deterministic.
method Target Distribution Learning (TDL) alternates between proposing a target distribution and training the policy network to approach it.
result TDL leads to more stable policy improvements over iterations compared to existing methods.
Deep neuroevolution and deep reinforcement learning (deep RL) algorithms are two popular approaches to policy search. The former is widely applicable and rather stable, but suffers from low sample efficiency. By contrast, the latter is more sample efficient, but the most sample efficient variants are also rather unstab…
Memory-efficient algorithm reduces variance in off-policy RL.
problem High variance in off-policy policy optimization.
method Memory-efficient, stochastically variance-reduced algorithm using off-policy samples.
result Empirically validated effectiveness of the proposed algorithm.
This paper introduces a new method to evaluate multiple policies simultaneously.
problem Estimating the value of many policies for a single set of states.
method Developed a scalable, differentiable fingerprinting mechanism to represent complex policies.
result The method can produce policies that outperform those that generated the training data, in zero-shot manner.
Novel RL-based NPG improves multi-objective NAS efficiency and performance.
problem Discovering optimal neural architectures with multiple conflicting objectives.
method Non-stationary policy gradient with adaptive reward functions and shared model.
result Framework efficiently approximates full Pareto front and achieves superior performance.
RPG improves sample efficiency in RL by learning optimal action ranks.
problem Sample inefficiency in reinforcement learning.
method Ranking Policy Gradient (RPG) method that learns optimal rank of discrete actions.
result RPG reduces sample complexity and improves sample efficiency for large-scale problems.
Accelerates policy optimization in RL with optimistic and adaptive updates.
problem Improving policy optimization methods in reinforcement learning.
method Integrates foresight into policy improvement step via optimistic and adaptive updates.
result Designs an optimistic policy gradient algorithm, adaptive via meta-gradient learning.
Evolutionary Strategies optimize hyper-parameters for off-policy learning.
problem Hyper-parameter sensitivity in off-policy learning.
method Application of Evolutionary Strategies for online hyper-parameter tuning.
result Our method outperforms state-of-the-art baselines.
FiDi-RL combines deep RL with FiDi policy search for efficient continuous control.
problem Challenges in continuous control tasks using reinforcement learning.
method Combines Deep Deterministic Policy Gradients (DDPG) with Augment Random Search (ARS) using FiDi policy search.
result Improves data efficiency and stability of ARS.
Study improves policy search in continuous control by using heavy-tailed distributions.
problem Challenges in continuous space policy search due to non-convexity and myopic-farsighted incentives.
method Introduced heavy-tailed policy parameterizations and analyzed convergence rates and stability.
result Convergence rate to stationarity depends on policy's tail index and exploration tolerance.
Genetic algorithms have been widely used in many practical optimization problems. Inspired by natural selection, operators, including mutation, crossover and selection, provide effective heuristics for search and black-box optimization. However, they have not been shown useful for deep reinforcement learning, possibly …
Prognosticator improves performance in non-stationary MDPs.
problem Real-world applications often violate the stationary assumption in reinforcement learning.
method A policy gradient algorithm that maximizes a forecast of future performance.
result Prognosticator outperforms online adaptation techniques in non-stationary MDPs.
New algorithm reduces variance in Monte Carlo simulations using deep neural networks and policy gradients.
problem Reducing variance in Monte Carlo simulations for estimating function values.
method Optimal correlation search using deep neural networks and policy gradients.
result Optimal correlation function reduces variance by approximating and calibrating policy.
ETGL-DDPG improves DDPG for sparse reward control with new exploration and replay techniques.
problem Sparse reward continuous control in reinforcement learning.
method Introduces εt-greedy search and GDRB framework for efficient exploration and reward use. result ETGL-DDPG outperforms DDPG and other methods on sparse-reward continuous benchmarks.
The performance of policy gradient methods is sensitive to hyperparameter settings that must be tuned for any new application. Widely used grid search methods for tuning hyperparameters are sample inefficient and computationally expensive. More advanced methods like Population Based Training that learn optimal schedule…
New method for RL with general utilities using variational policy gradient.
problem Optimizing policies with general concave utility functions in RL.
method Derives Variational Policy Gradient Theorem, develops variational Monte Carlo gradient estimation algorithm.
result Global convergence to optimal policy for general objectives, exponential convergence under strong convexity.
Deep RL learns 2-opt heuristics to improve TSP solutions.
problem Improving TSP solutions beyond initial heuristics.
method Deep reinforcement learning to learn 2-opt operations.
result Learned policies improve solutions faster than previous methods.
The objective in a traditional reinforcement learning (RL) problem is to find a policy that optimizes the expected value of a performance metric such as the infinite-horizon cumulative discounted or long-run average cost/reward. In practice, optimizing the expected value alone may not be satisfactory, in that it may be…
Paper learns domain randomization distributions for robust robot policies.
problem Finding good domain randomization parameters for simulation without real data.
method Gradient-based search methods to learn domain randomization distribution.
result Improvements in jump-start and asymptotic performance when transferring policies.
The paper studies how search and distillation improve reasoning in large language models.
problem Improving reasoning capabilities of large language models.
method Viewing chain-of-thought generation as a metastable Markov process, proving benefits of search and distillation.
result Search protocol rewards sparse edges, reducing the expected number of steps to reach different clusters.
In this work, we explore how probabilistic programs can be used to represent policies in sequential decision problems. In this formulation, a probabilistic program is a black-box stochastic simulator for both the problem domain and the agent. We relate classic policy gradient techniques to recently introduced black-box…
Armijo line-search speeds up gradient descent for various functions.
problem Improving convergence rate of gradient descent.
method Applying Armijo line-search to adjust step-size in gradient descent.
result GD with Armijo line-search converges faster than GD with a fixed step-size.
Bayesian optimization tackles expensive discrete and mixed parameter spaces.
problem Optimizing expensive functions with discrete and mixed parameters.
method Probabilistic reparameterization to maximize expectation of AF over continuous parameters.
result Our approach provably converges to a maximizer of the AF and enjoys the same regret bounds as standard BO.
We propose a method to optimise the parameters of a policy which will be used to safely perform a given task in a data-efficient manner. We train a Gaussian process model to capture the system dynamics, based on the PILCO framework. Our model has useful analytic properties, which allow closed form computation of error …