Trade-R1 bridges verifiable rewards to stochastic financial markets via process-level reasoning verification.
problem Extending RL to financial markets where rewards are verifiable but noisy.
method A verification method that transforms reasoning over financial documents into a structured RAG task, using a triangular consistency metric.
result DSR achieves superior cross-market generalization while maintaining reasoning consistency.
GRPO optimizes LLMs with verifiable rewards, amplifying policy success.
problem Improving LLMs' reasoning under verifiable binary rewards.
method Introduces GRPO, analyzes variants of reward normalization and regularization.
result GRPO amplifies policy success, converging to a fixed point exceeding the reference.
Extends reinforcement learning alignment to scalar rewards, improving math reasoning.
problem Designing reinforcement learning algorithms for general LLM alignment.
method Introduces f-GRPO and f-HAL, estimating f-divergences between reward-aligned and unaligned distributions.
result Improves math reasoning RLVR tasks and mitigates reward hacking.
RLVR training dynamics reveal an implicit curriculum that shapes learning progression.
problem Understanding how RLVR overcomes the long-horizon barrier.
method Developed a theory of training dynamics for RLVR on transformers, using Fourier analysis on finite groups.
result Mixed-difficulty training naturally follows an implicit curriculum, shaping the learning progression from easy to hard.
RLVR maintains safety while improving reasoning capabilities in LLMs.
problem Safety-capability tradeoff in fine-tuning LLMs.
method Reinforcement Learning with Verifiable Rewards (RLVR) and theoretical analysis.
result RLVR can enhance reasoning while maintaining safety guardrails.
SAGE enhances reinforcement learning by injecting hints to prevent model stagnation.
problem Sparse rewards cause large language models to stall under relative policy optimization.
method SAGE injects privileged hints during training to increase within-group outcome diversity.
result SAGE consistently outperforms GRPO on 6 benchmarks with LLMs, achieving significant improvements.
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.
Sparse reward is one of the biggest challenges in reinforcement learning (RL). In this paper, we propose a novel method called Generative Exploration and Exploitation (GENE) to overcome sparse reward. GENE automatically generates start states to encourage the agent to explore the environment and to exploit received rew…
Improves generative models by optimizing rewards and sample editing.
problem Efficiently generating high-reward samples with structural constraints.
method Introduces MDM-VGB, a discrete diffusion sampler that augments unmasking generation with reward-guided remasking.
result MDM-VGB achieves quadratic complexity and robustness to noise, outperforming heuristics like best-of-N. The paper analyzes RLVR's training dynamics, proving convergence depends on aligning update direction with Gradient Gap.
problem Understanding why RLVR works and its limitations.
method Analysis of RLVR's training process at trajectory and token levels, introducing Gradient Gap.
result Convergence depends on aligning update direction with Gradient Gap, with a sharp step-size threshold.
The paper explores a Multi-Objective RL approach for trading that generalizes reward functions.
problem Improving performance in single-asset trading through adaptive reward functions.
method Developed a Multi-Objective Deep Reinforcement Learning algorithm to generalize reward functions and discount factors.
result The Multi-Objective algorithm demonstrates increased predictive stability and better performance in sparse reward scenarios.
New method uses correlated auxiliary feedback to reduce regret in parameterized bandits.
problem Reducing regret in parameterized bandits with correlated auxiliary feedback.
method Develops a reward estimator using auxiliary feedback with tight confidence bounds.
result Shows significant reduction in regret compared to standard methods.
Language models perform worse with implicit reward models than explicit ones.
problem Understanding why implicit reward models generalize worse than explicit ones.
method Investigated the root cause of the generalization gap between IM-RMs and EX-RMs.
result Implicit reward models rely more on superficial token-level cues, leading to worse generalization.
New algorithm reduces regret in delayed feedback generalised linear bandits.
problem Regret in delayed feedback generalised linear bandits.
method Adaptation of optimistic algorithm to delayed feedback.
result Achieves a regret bound independent of the horizon's delay penalty.
FUSE improves verification quality without ground truth labels.
problem Verification of model outputs using imperfect judges and reward models.
method Ensembling verifiers without ground truth labels using spectral algorithms.
result FUSE matches or improves upon semi-supervised alternatives in test-time scaling experiments.
Policy gradient algorithms typically combine discounted future rewards with an estimated value function, to compute the direction and magnitude of parameter updates. However, for most Reinforcement Learning tasks, humans can provide additional insight to constrain the policy learning. We introduce a general method to i…
EORM boosts LLM accuracy with a lightweight, energy-based verifier.
problem Efficiently verifying mathematical reasoning in large language models.
method Energy-based framework to rank Chain-of-Thought solutions using simple outcome labels.
result EORM boosts LLM accuracy to 90.7% on GSM8k and 63.7% on MATH with only 55M parameters.
In a linear stochastic bandit model, each arm is a vector in an Euclidean space and the observed return at each time step is an unknown linear function of the chosen arm at that time step. In this paper, we investigate the problem of learning the best arm in a linear stochastic bandit model, where each arm's expected r…
Adapts GRPO for off-policy RL, improving reward.
problem Improving training stability and efficiency in RL.
method Adapts GRPO to off-policy setting, uses clipped surrogate objectives.
result Off-policy GRPO outperforms on-policy GRPO in empirical tests.
New RLHF framework handles general preference oracles without reward functions.
problem Handling general preference oracles without assuming a reward function.
method Developed a minimax game between two LLMs for RLHF under a general preference oracle, focusing on KL-regularized preference.
result Proposed algorithms for efficient offline and online RLHF learning.
A scalable algorithm for sampling and fine-tuning models using Tilt Matching.
problem Efficient sampling and fine-tuning of generative models.
method Tilt Matching, arising from a dynamical equation, minimizes variance and inherits regularity from stochastic interpolants.
result Empirically verified to be efficient and highly scalable, providing state-of-the-art results.
Study on when RLVR can learn compositional problems.
problem Understanding when RLVR can learn compositional problems.
method Theoretical analysis of task-advantage ratio to characterize learnability.
result Identified conditions for learnability of compositional problems.
New method optimizes language model performance for test-time strategies.
problem Mismatch between training objectives and test-time deployment of large language models.
method Tail-Extrapolated estimators to approximate best-of-N performance from limited training rollouts.
result Improved performance of best-of-N deployment across various models and datasets.
Unified approach to RLHF tackles uncertainty in reward function.
problem Uncertainty in reward function learned from human feedback.
method Value-incentivized preference optimization (VPO) that regularizes the reward function with value function.
result Theoretical and practical guarantees for both online and offline RLHF settings.
The paper develops a theory for iterative self-improvement of models, proving conditions for better performance with easy-to-hard curricula.
problem Lack of theoretical foundation for iterative self-improvement in practical settings.
method Modeling self-improvement as maximum-likelihood fine-tuning on reward-filtered distributions and proving finite-sample guarantees.
result Explicit feedback loop and conditions for better performance with easy-to-hard curricula.
RLHF uses human feedback to train AI models, posing statistical challenges.
problem Aligning AI models with human preferences using noisy, subjective feedback.
method Supervised fine-tuning, reward modeling, policy optimization, statistical ideas.
result Statistical methods for reward function learning and policy optimization.
E-valuator converts verifier scores into reliable decision rules.
problem Ensuring the correctness of agent trajectories based on heuristic scores.
method Sequential hypothesis testing framework for online monitoring of agent trajectories.
result E-valuator provides better false alarm rate control and statistical power than other strategies.
In this paper, we study multi-armed bandit problems in explore-then-commit setting. In our proposed explore-then-commit setting, the goal is to identify the best arm after a pure experimentation (exploration) phase and exploit it once or for a given finite number of times. We identify that although the arm with the hig…
A new algorithm for personalized recommendations adapts to changing user interests.
problem Adapting to time-varying user interests in recommendation systems.
method Contextual bandit approach with models for disjoint and hybrid payoffs.
result Sublinear regret in time length T for abrupt reward changes.
Study online learning with off-policy feedback in adversarial bandit problems.
problem Learning with limited direct feedback in sequential decision making.
method Proposed algorithms that adapt pessimistic reward estimators to handle unknown behavior policy.
result Guaranteed regret bounds scaling with policy mismatch, improving performance against well-covered comparators.
New GFlowNet training framework using policy gradients for combinatorial object generation.
problem Training efficiency and robustness in GFlowNet models.
method Policy-dependent rewards and coupled training strategy for forward and backward policies.
result Advanced RL perspectives for robust gradient estimation improve GFlowNet performance.
New algorithm prevents strategic replication in multi-armed bandit problems.
problem Strategic replication by agents can exploit bandit algorithms' balance.
method Designs Hierarchical UCB (H-UCB) and Robust Hierarchical UCB (RH-UCB) algorithms.
result Achieves O(lnT)-regret and sublinear regret in realistic scenarios. 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 MAB model incentivizes user arm-pulling with self-reinforcing preferences.
problem Balancing exploration and exploitation in recommender systems with incentivized user preferences.
method Proposes a new MAB model with random arm selection and two policies: At-Least-n Explore-Then-Commit and UCB-List. result Achieves O(logT) expected regret and O(logT) expected payment over a time horizon T. New approach combines semi-supervised learning and bandits for better predictions.
problem Online semi-supervised learning with bandit feedback for applications like clinical trials and ad recommendations.
method Adjusted Graph Convolutional Network (GCN) for contextual bandits, with semi-supervised missing rewards imputation.
result Developed multi-GCN embedded contextual bandit algorithms verified on real-world datasets.
A new estimator reduces variance in slate bandit OPE.
problem Large action spaces in slate bandits cause high variance in OPE.
method Develops Latent IPS (LIPS) to optimize slate abstractions for low variance and bias.
result LIPS substantially outperforms existing estimators in scenarios with non-linear rewards and large slate spaces.
New framework improves restless bandit policies for large numbers of arms.
problem Efficiently compute policies for large numbers of arms in restless bandit problems.
method Follow-the-Virtual-Advice framework, converting single-armed policies to N-armed policies.
result Achieves an O(1/\sqrt{N}) optimality gap in both discrete and continuous settings.
Study on indexability of restless multi-armed bandits and rollout policy performance.
problem Maximizing discounted rewards in finite state restless multi-armed bandit problems.
method Decouple the problem into single-armed restless bandits, analyze using value iteration, and compare with Whittle index policy.
result Demonstrates conditions for indexability and compares performance of index policy and rollout policy.
This study explains RL training dynamics in LLMs, focusing on token-level optimization and reasoning pattern reshaping.
problem Understanding the training dynamics of RL in LLMs to improve their reasoning capabilities.
method Empirical analysis and theoretical modeling of RL training process, focusing on reasoning patterns and token optimization.
result RL primarily optimizes a sparse subset of critical tokens, reshaping reasoning pattern distributions and affecting model performance.
New algorithm robust to probabilistic unbounded adversarial attacks in bandit problems.
problem Powerful adversaries that can catastrophically perturb the revealed reward in bandit problems.
method Proposes med-E-UCB and med-ε-greedy algorithms based on sample median for robustness. result Achieves O(logT) pseudo-regret under arbitrary and unbounded reward perturbation. With the breakthrough of computational power and deep neural networks, many areas that we haven't explore with various techniques that was researched rigorously in past is feasible. In this paper, we will walk through possible concepts to achieve robo-like trading or advising. In order to accomplish similar level of pe…
New meta-learning approach for bandit policies that achieve high average reward.
problem Designing bandit policies that balance between worst-case and Bayesian assumptions.
method Differentiable parameterized policies optimized using policy gradients.
result Proposed algorithm achieves low regret and is practical for various bandit problems.
Formalizes weak and strong verification for LLMs, controlling errors without assumptions.
problem Balancing cost and reliability in reasoning with LLMs.
method Formalizes weak-strong verification policies, introduces metrics, develops online algorithm.
result Optimal policies admit a two-threshold structure, and calibration and sharpness govern value of weak verifiers.
Transformers learn sparse Boolean functions through RL and SFT, revealing distinct learning behaviors.
problem Learning sparse Boolean functions with Transformers.
method Reinforcement Learning (RL) with process rewards and Supervised Fine-Tuning (SFT).
result RL learns the whole CoT chain simultaneously, while SFT learns step by step.
This paper proposes a new AED framework for multi-metric experiments with fixed budget.
problem Statistical power challenges in testing multiple metrics simultaneously.
method Two-phase structure: adaptive exploration followed by validation. SHRVar algorithm with relative-variance-based sampling.
result Achieves provable error probability that decreases exponentially.
Inference-Time Scaling can be extended to domains prone to systematic failure using intrinsic statistics.
problem Scaling inference time in domains prone to systematic failure
method Intrinsic Selection (iS), Intrinsic Particle Filtering (iPF), and Particle Distillation (dPF)
result Intrinsic Selection improves engineering design selection by 20% and pass@1 by 6.1 points on average.
Polynomial chaos surrogates quantify epistemic uncertainty in AI-driven scientific models.
problem Uncertainty in reward estimates hinders interpretability in sequential generative models.
method Fit polynomial chaos expansions to trained models to propagate epistemic uncertainty and quantify sensitivity.
result Interpretable decomposition of reward components driving generative decisions.
Counterfactual thinking describes a psychological phenomenon that people re-infer the possible results with different solutions about things that have already happened. It helps people to gain more experience from mistakes and thus to perform better in similar future tasks. This paper investigates the counterfactual th…