Curiosity-Critic improves world model training by focusing on cumulative prediction error.
problem Training world models with intrinsic rewards that consider cumulative prediction error.
method Curiosity-Critic uses a surrogate reward based on the difference between current and asymptotic prediction errors, estimated online by a co-trained critic.
result Curiosity-Critic outperforms other methods in training speed and final world model accuracy.
New method uses reward prediction error for efficient exploration.
problem Efficient exploration in reinforcement learning, especially in complex environments.
method Reward prediction error (RPE) as an intrinsic motivation for exploration, combined with a deep reinforcement learning method (QXplore).
result QXplore outperforms state-novelty methods in diverse tasks, especially when state novelty is not correlated with improved reward.
This study proposes hidden state curiosity to enhance RL models' resilience against noise.
problem Curiosity traps in RL models distract agents from discovering novel experiences.
method Proposed hidden state curiosity based on the Free Energy Principle to reward agents for KL divergence between predictive priors and posteriors.
result Agents with hidden state curiosity are more resilient against curiosity traps compared to those with prediction error curiosity.
The Artificial Prediction Market is a recent machine learning technique for multi-class classification, inspired from the financial markets. It involves a number of trained market participants that bet on the possible outcomes and are rewarded if they predict correctly. This paper generalizes the scope of the Artificia…
Reinforcement learning algorithms rely on carefully engineering environment rewards that are extrinsic to the agent. However, annotating each environment with hand-designed, dense rewards is not scalable, motivating the need for developing reward functions that are intrinsic to the agent. Curiosity is a type of intrins…
This work improves MARL exploration efficiency using predictive guidance and prioritized experience replay.
problem Challenges in multi-agent reinforcement learning, especially in exploration efficiency and reward informativeness.
method A predictive network forecasts reward, guiding MARL to accelerate exploration. Improved prioritized experience replay utilizes TD error effectively.
result The proposed algorithm outperforms existing methods in cooperative multi-agent environments.
A network of spiking agents learns complex tasks using global reward signals.
problem Solving complex reinforcement learning tasks.
method A hierarchical network of GLM spiking agents, each modulating its firing policy based on local and global reward signals.
result A network of spiking agents can learn complex action representations to solve RL tasks.
The study categorizes reward errors in reinforcement learning, finding some can be beneficial.
problem Training language models with imperfect proxy rewards.
method Theoretical analysis of policy gradient optimization and categorization of reward errors.
result Reward errors can be benign or even beneficial, preventing policy from stalling.
Learning reward functions can lead to poor policy performance despite low error.
problem Low error in learned reward functions does not guarantee low regret in policy performance.
method Mathematical analysis of reward learning and policy optimization.
result A low expected test error of the reward model guarantees low worst-case regret, but error-regret mismatch can occur with certain data distributions.
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.
STAR framework reduces OPE variance by distilling complex problems into discrete ARPs.
problem High variance and bias in off-policy evaluation methods.
method STAR framework that includes various OPE estimators and leverages state abstraction.
result Predictions from ARPs estimated from off-policy data are asymptotically correct.
No real-world reward function is perfect. Sensory errors and software bugs may result in RL agents observing higher (or lower) rewards than they should. For example, a reinforcement learning agent may prefer states where a sensory error gives it the maximum reward, but where the true reward is actually small. We formal…
In reinforcement learning, a decision needs to be made at some point as to whether it is worthwhile to carry on with the learning process or to terminate it. In many such situations, stochastic elements are often present which govern the occurrence of rewards, with the sequential occurrences of positive rewards randoml…
Develops a new model to predict training dynamics of large language models.
problem Lack of mechanistic understanding of training dynamics in large language models.
method A first-principles reduced-order model of training dynamics, predicting group-size invariance and stability thresholds.
result Closed-form model predicts training dynamics with high accuracy and provides new diagnostics.
In many real-world scenarios, rewards extrinsic to the agent are extremely sparse, or absent altogether. In such cases, curiosity can serve as an intrinsic reward signal to enable the agent to explore its environment and learn skills that might be useful later in its life. We formulate curiosity as the error in an agen…
Post-training optimizes model performance beyond base model limits.
problem Optimizing sequence prediction models beyond the base model's support.
method Policy gradient (PG) and adaptive learning rate (LR) techniques.
result Post-training with PG can achieve near-optimal performance beyond the base model's support.
Theory for RLHF generalization under reward shift and clipped KL.
problem Theoretical understanding of RLHF generalization, especially with reward shift and clipped KL.
method Developed generalization theory for RLHF, accounting for reward shift and clipped KL.
result Presented generalization bounds for RLHF, suggesting generalization error from sampling, reward shift, and KL clipping.
New algorithm optimally identifies best arm in both stochastic and adversarial settings.
problem Best arm identification in stochastic and adversarial reward scenarios.
method Parameter-free algorithm designed to be robust to adversarial rewards and optimal in stochastic problems.
result Algorithm's error rate matches optimal bounds in stochastic problems and is robust to adversarial rewards.
Proposes a method to boost deep reinforcement learning with sparse rewards.
problem Challenges in learning complex behaviors with long horizons and sparse rewards.
method Predictive coding for reward shaping.
result Achieves better learning by providing reward signals that understand environment dynamics and emphasize useful features.
We introduce Recurrent Predictive State Policy (RPSP) networks, a recurrent architecture that brings insights from predictive state representations to reinforcement learning in partially observable environments. Predictive state policy networks consist of a recursive filter, which keeps track of a belief about the stat…
This work shows how approximate reward models can significantly improve inference-time scaling.
problem Improving the efficiency of inference for large language models.
method Identifying the Bellman error of approximate reward models and using Sequential Monte Carlo (SMC) for inference.
result Approximate reward models can reduce computational complexity from exponential to polynomial in T. TD learning reduces prediction error in Markov chain problems.
problem Estimating value functions in Markov chains with temporal inconsistency.
method Temporal difference learning minimizes temporal inconsistency between successive estimates.
result TD learning can significantly reduce mean-squared error in value estimates.
AutoDIME automates design of multi-agent environments for RL.
problem Designing multi-agent environments for reinforcement learning is challenging.
method Developed intrinsic teacher rewards for multi-agent settings and evaluated them in various tasks.
result Value disagreement was found to be most consistent and effective across tasks.
New algorithm for multi-player bandits with collision-dependent rewards.
problem Stochastic multi-player multi-armed bandits with collision-dependent reward distributions.
method Error-Correction Collision Communication (EC3) algorithm.
result EC3 algorithm achieves optimal regret approaching centralized MP-MAB regret.
Self-supervised reward prediction improves RL in sparse reward settings.
problem Data efficiency and sparse reward signals in reinforcement learning.
method Learning a state representation for reward prediction and using it to shape rewards.
result Self-supervised reward prediction enhances RL algorithms in single-goal environments.
Method predicts future rewards from past actions in a linear Gaussian system.
problem Maximizing cumulative reward in a stochastic multi-armed bandit with linear Gaussian dynamics.
method Proposes a method using a modified Kalman filter to predict future rewards based on past rewards.
result Reward from any action can be used to predict another action's future reward.
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.
The study analyzes how neural reward models learn features for policy optimization in a Gaussian single-index model.
problem Reward modeling in policy optimization and its impact on downstream value.
method Two-stage neural reward model: first learns hidden direction, then fits readout layer.
result For any feature-learning temperature above a dimension-free threshold, a constant fraction of neurons recover the hidden direction.
Machine learning is often used in competitive scenarios: Participants learn and fit static models, and those models compete in a shared platform. The common assumption is that in order to win a competition one has to have the best predictive model, i.e., the model with the smallest out-sample error. Is that necessarily…
Reward augmented maximum likelihood (RAML), a simple and effective learning framework to directly optimize towards the reward function in structured prediction tasks, has led to a number of impressive empirical successes. RAML incorporates task-specific reward by performing maximum-likelihood updates on candidate outpu…
New approach for reward-free exploration reduces estimation error.
problem Reward-free exploration in reinforcement learning.
method Adaptive approach reducing MDP estimation error.
result Reward-free UCRL algorithm improves sample complexity.
Research predicts cryptocurrency staking rewards with high accuracy.
problem Predicting cryptocurrency staking rewards.
method Two predictive methodologies: sliding-window average and linear regression models.
result ETH staking rewards can be forecasted with RMSE within 0.7% and 1.1% of the mean value for 1-day and 7-day look-aheads respectively.
New metrics improve scRNA-seq perturbation modeling by reducing mode collapse.
problem Outperformed by simple mean prediction in scRNA-seq perturbation modeling.
method Introduce DEG-aware metrics (WMSE, Rw2(Δ)) and negative/positive baselines. result WMSE loss function reduces mode collapse and improves model performance.
Privacy-preserving reinforcement learning from human feedback using decoupled reward modeling.
problem Training large language models with sensitive user information while preserving privacy.
method Proposes a privacy-preserving framework that imposes differential privacy on reward learning only.
result Privacy contributes an additional additive term to the suboptimality gap, and the upper bound is rate-optimal up to logarithmic factors.
Proposes learning latent reward model for planning from rewards.
problem Planning in high-dimensional state spaces with limited reward information.
method Directly learns a latent dynamics model from rewards, planning in latent state-space.
result Successfully learns accurate latent reward prediction model, achieving strong performance and high sample efficiency.
Recent studies have shown that reinforcement learning (RL) models are vulnerable in various noisy scenarios. For instance, the observed reward channel is often subject to noise in practice (e.g., when rewards are collected through sensors), and is therefore not credible. In addition, for applications such as robotics, …
A novel approach uses an ensemble of Gaussian processes for robust and adaptive reinforcement learning.
problem Adaptive reinforcement learning in large or continuous state spaces.
method Online scalable (OS) approach with a weighted ensemble of Gaussian processes.
result The ensemble approach improves performance in adversarial settings.
Robust Q-learning algorithm resists corrupted rewards.
problem Non-ideal environments with corrupted rewards.
method Developed a novel robust Q-learning algorithm using historical data.
result Finite-time convergence rate matches state-of-the-art bounds.
Many practical environments contain catastrophic states that an optimal agent would visit infrequently or never. Even on toy problems, Deep Reinforcement Learning (DRL) agents tend to periodically revisit these states upon forgetting their existence under a new policy. We introduce intrinsic fear (IF), a learned reward…
A smart method predicts and optimizes decisions online with resource constraints.
problem Online decision-making with resource constraints.
method Combines prediction and optimization with dual update using mirror descent.
result Regret bounds and convergence rates for general convex feasible regions.
Paper addresses reward learning issues in RL, improving both under- and over-estimation.
problem Reward learning from data can lead to reward delusions or underestimation, causing unintended behaviors.
method Connects reward learning to positive-unlabeled (PU) learning and applies a large-scale PU learning algorithm.
result Improves both GAIL and supervised reward learning without additional assumptions.
New method predicts and optimizes test-time scaling for LLMs.
problem Lack of principled guidance on scaling LLMs efficiently.
method Tail-guided search to predict and allocate compute.
result SLG Search achieves higher rewards with less compute.
TLRS improves predictive power of mined formulaic alpha factors.
problem Sparse rewards in RL for mining formulaic alpha factors.
method Trajectory-level Reward Shaping (TLRS) with reward centering.
result TLRS boosts predictive power by 9.29% over existing methods.
New algorithm uses machine learning to predict rewards for decision-making problems.
problem Sequential decision-making under uncertainty with scarce online data.
method Machine Learning-Assisted Upper Confidence Bound (MLA-UCB) algorithm.
result Proves to improve cumulative regret even with biased surrogate rewards.
Kernel-based methods improve policy evaluation in MRP models.
problem Estimating value functions in infinite-horizon discounted MRP models.
method Kernel-based temporal difference methods using reproducing kernel Hilbert spaces.
result Optimal error bounds derived for the kernel-based LSTD estimate.
Resolves spurious correlations in causal models via intervention design.
problem Spurious correlations lead to incorrect causal models in reinforcement learning environments.
method Proposes a method to design interventions that improve causal models by incentivizing agents to find errors.
result Experimental results show improved causal models compared to baselines.
We present an adversarial active exploration for inverse dynamics model learning, a simple yet effective learning scheme that incentivizes exploration in an environment without any human intervention. Our framework consists of a deep reinforcement learning (DRL) agent and an inverse dynamics model contesting with each …
In structured output prediction tasks, labeling ground-truth training output is often expensive. However, for many tasks, even when the true output is unknown, we can evaluate predictions using a scalar reward function, which may be easily assembled from human knowledge or non-differentiable pipelines. But searching th…