Searcharxiv⌕ Search

arXiv subjects

Donghwan Lee

Publications and source records attributed to Donghwan Lee.

At least 19 recordsLinked to original sources

An Augmented Lagrangian Approach to Lifted Bellman Linear Programming

Temporal-difference critics regress onto a bootstrapped target and usually need a target network. Lifted Bellman Linear Programming (LBLP) instead expresses the in-sample optimal values of offline RL as the solution to a linear program over the dataset. Its neural solver ALBUM relaxes the constraints into hinge penalties with fixed coefficients, and its guarantees are conditions on coefficients chosen in advance. We solve the same program with the method of multipliers: the coefficients become multipliers learned by dual ascent, with one penalty parameter $ρ$. In the convex case the fixed points are the KKT pairs of LBLP, and at every stationary point the applied constraint weights balance the objective exactly. The critic can serve a Gaussian actor, a flow actor, and best-of-N sampling, and for the Gaussian actor we propose an advantage-normalized DDPG+BC objective. With a single $Q$ and $V$ and no target network, our AL-LBLP with best-of-N reaches 60.1\% average success on the 50 state-based OGBench tasks. On long-horizon tasks, it reaches 81.5% on humanoidmaze-giant and 79.8% on antmaze-giant. It does not collapse up to batch size 8192.

cs.AI↗

VIGOR: Zero-Shot Visual Generalization via Latent-Space Consistency in Model-Based Reinforcement Learning

Model-based reinforcement learning (MBRL) achieves strong sample efficiency by planning in learned latent dynamics but degrades substantially under unseen visual distractions such as background, lighting, or camera changes. Unlike model-free RL, where encoder perturbations affect only single-step predictions, MBRL suffers from a two-level vulnerability: distractions push encoder outputs out of distribution, and these errors compound through recursive latent rollouts. We propose visual generalization via latent-space consistency in model-based RL (VIGOR), which enables zero-shot generalization to unseen distractions while retaining its backbone's sample efficiency. VIGOR comprises three interdependent components: (i) asymmetric weak-to-strong augmentation, which pairs weak-only and weak-to-strong latent views within a batch; (ii) dynamics-level consistency, which enforces augmentation-invariant transition predictions via direct latent regression; and (iii) encoder-level stabilization, which prevents encoder drift under the cross-augmentation supervision of (ii). On the DeepMind Control Suite (DMC) and Robosuite, VIGOR instantiated on TD-MPC2 outperforms state-of-the-art model-free and model-based baselines in zero-shot generalization, surpassing the second-best baseline by 3.4% on DMC and 43.6% on Robosuite, with sample efficiency comparable to its backbone and fewer environment interactions than the other baselines. Ablations show that VIGOR is augmentation-agnostic: augmentations from distinct perturbation families preserve strong generalization, confirming that latent-space consistency, not the augmentation choice, drives robustness. A second instantiation on DreamerV3, a structurally distinct recurrent state-space model, also improves zero-shot generalization on both benchmarks, indicating backbone-agnostic benefits. Code and videos: https://vigor-gen.github.io/.

cs.AI↗

ReState: Noise-Level-Aware State Correction for Training-Free Object Removal in Few-Step Diffusion Models

Training-free object removal commonly redirects self-attention away from the target region, but on few-step distilled diffusion models this often leaves object-shaped residuals. We identify a temporal mismatch: a structured prediction forms inside the mask at the first forward, and the short trajectory offers few later chances to revise it, while attention rerouting acts mainly in later steps. Across four distilled backbones, magnitude-matched single-forward interventions show that background-directed state correction affects the final result more than attention rerouting. We therefore propose ReState, which corrects the masked clean prediction toward background evidence retrieved from the input using a noise-level-aware weight that vanishes at the final forward. ReState retains object-excluding attention as a complementary constraint and adds no denoiser evaluations. Across two paired benchmarks and two few-step backbones, ReState achieves the best masked PSNR, SSIM, and LPIPS among training-free methods, improving PSNR by 3.17 dB on average over the strongest baseline. On OBER-Test, it also outperforms the 50-step attention-forcing reference on all three metrics at 6.7x lower latency. The same configuration transfers to two additional distilled backbones without retuning. Project page: https://tdrlee.github.io/ReState

cs.CV↗

Depth-to-RGB: Repurposing a Frozen Depth Estimator for Geometry-Guided Compositing

Reference-based object compositing inserts or replaces an object using a background image, a reference image, and a 2D compositing mask. These inputs guide appearance and placement but leave the completed scene's geometry implicit, which can distort object structure or alter the surroundings. Our Depth-to-RGB (D2R) framework predicts composite depth for a scene not yet observed in the RGB inputs. It learns reference-conditioned corrections to a frozen depth estimator using encoder features of paired completed scenes as targets. The unchanged decoder maps the corrected representation to the intended scene's depth, which a separately trained renderer holds fixed during RGB synthesis. Under matched architecture and training, encoder-feature supervision reduces OOD Stage-1 AbsRel by 31.4% relative to decoded-depth supervision. We also introduce AnyInsertion++ with paired in-distribution and category-disjoint splits to evaluate generalization beyond compositing training categories. The complete D2R system leads 12 open-source and 3 closed-source baselines in estimator-derived geometry and photometric quality on both paired splits. On category-disjoint data, D2R reduces AbsRel by 43.7% and improves PSNR by 2.4 dB over the matched RGB baseline. Across three unpaired benchmarks, D2R leads both identity metrics and reduces mean CLIP reference cosine distance by 55% relative to the strongest baseline. Project page: https://shjo-april.github.io/Depth2RGB/

cs.CV↗

Refinement-Based Flow Policy Optimization

Flow-based policies offer an expressive representation for online reinforcement learning, but conventional flow matching requires samples drawn from the distribution to be modeled. This poses a challenge when the desired action distribution is defined only implicitly by a Q-function, since directly sampling actions from the resulting distribution is generally intractable. We propose Refinement-Based Flow Policy Optimization (RFPO), a novel framework for training a flow policy in online reinforcement learning by alternating between Q-guided sample refinement and self-target flow matching. RFPO first generates actions from Gaussian noise using the current flow policy and then uses a finite-step stochastic refinement procedure to move them toward an energy-based distribution induced by the Q-function. Each refined action is then paired with its corresponding initial noise sample and used as a fixed target for flow-matching training. By repeatedly refining its own outputs and learning from the resulting targets, RFPO incorporates Q-guidance into the policy without requiring direct samples from the target distribution, while retaining the capacity to represent multiple action modes. We further provide a theoretical analysis of the distributional dynamics induced by RFPO. Across six continuous-control tasks, RFPO matches or outperforms a standard Gaussian-policy baseline on almost every task. Experiments on six synthetic two-dimensional target distributions with diverse geometries demonstrate that RFPO captures complex multimodal structure without mode collapse.

cs.LG↗

Is Online Interaction Necessary for Recovery? A Minimalist Approach to Robust Planning via Perturbation

Behavior cloning (BC) is vulnerable to covariate shift during closed-loop execution, where small prediction or execution errors can drive the robot toward states poorly covered by the demonstration data. We focus on action-sequence planning, where a policy predicts a finite-horizon sequence of actions as a reference trajectory for robot execution. Existing approaches to covariate shift often rely on collecting additional corrective demonstrations, requiring further environment interaction and access to an expert or reference policy. We propose Perturbation-Augmented Recovery Supervision (PARS), a simple training approach for improving recovery using only existing demonstrations. PARS perturbs the robot's proprioceptive state and anchors only the later portion of the predicted trajectory to the original demonstration, leaving the earlier portion free to generate corrective motion and recover toward the demonstrated behavior within the planning horizon. Unlike conventional input-noise augmentation, which preserves the original supervision target over the entire prediction horizon, PARS explicitly provides trajectory-level supervision for recovery from perturbed states. PARS requires neither additional environment interaction nor expert queries and can be instantiated across different action-sequence policy classes with only minor modifications to their BC objectives. Experiments on 51 RLBench manipulation tasks with flow-based, transformer-based, and diffusion-based policies demonstrate that PARS improves robustness to covariate shift during closed-loop execution.

cs.RO↗

Lifted Bellman Linear Programming for Offline Reinforcement Learning

Offline reinforcement learning (RL) typically trains a critic by minimizing a regression loss against bootstrapped value targets stabilized by target networks with exponential moving average (EMA) updates. Multi-step targets incorporate behavior-policy actions and therefore require off-policy correction. We instead impose in-sample Bellman optimality on the critic through inequality constraints. We formulate the Lifted Bellman Linear Program (LBLP), which lifts the linear programming characterization of Bellman optimality to the joint $(Q,V)$ space so that every constraint involves only state-action pairs in the dataset. Its unique minimizer is the in-sample optimal pair, and constraints along $K$-step segments of dataset trajectories leave this minimizer unchanged for any rollout policy and horizon. Under deterministic dynamics, this minimizer lies between the best dataset return and the optimal value. Relaxing the constraints into hinge penalties recovers the same solution above a finite penalty coefficient in the tabular case. Approximate Lifted Bellman Unconstrained Minimization (ALBUM) implements this relaxation with neural networks and detaches the $K$-step rollout targets by stop gradient. Its objective contains no squared regression onto bootstrapped targets, so it can be trained without target networks or EMA updates. Under deterministic dynamics, the LBLP solution is a stationary point of the detached update under a coefficient condition independent of $γ$ and $K$, and the inequality constraints allow discounted returns along dataset trajectories to serve as lower bounds without off-policy correction or action chunking. On OGBench, ALBUM uses a single critic with a Gaussian policy, matches the average performance of FQL, and is comparable to recent action-chunking methods, while using the fewest parameters and the least peak GPU memory among all compared methods.

cs.LG↗

Taming the Adversary: A Cost-to-Disturbance Ratio Approach to Adversarial Reinforcement Learning

Reinforcement learning (RL) policies trained in simulation often degrade once deployed on real systems, where the controller must reject external disturbances that were never encountered in simulation. Robust RL addresses this by exposing the controller to perturbations while it learns, through domain randomization, adversarial minimax formulations, or probabilistic mixtures of protagonist and adversarial behavior. However, an unregulated disturbance mechanism destabilizes training and often collapses nominal performance relative to standard, non-robust methods. We propose cost-to-disturbance ratio adversarial training (CoDRA), a framework that expresses the controller--adversary trade-off as a ratio of accumulated cost to accumulated squared disturbance norm, and optimizes it through a self-normalized actor--critic update. In this algorithm, each value term is scaled by a stop-gradient normalization constant computed from the current batch. This moderates the adversary's incentive without altering the controller's own update, and requires neither an explicit disturbance penalty nor an auxiliary trade-off parameter. We evaluate CoDRA on two MuJoCo pendulum environments under force and mass sweeps. On InvertedDoublePendulum, CoDRA attains the lowest cost at every force level, including a force outside the range seen during training, and in all but one cell of the mass grid, whereas its advantage is less pronounced on the milder InvertedPendulum.

cs.LG↗

A Smooth Polynomial Lyapunov Certificate for Convergence of Q-Learning and Its Smooth Variants

Classical convergence analyses of Q-learning rely on the $\infty$-norm contraction of Bellman operators, and existing ordinary differential equation (ODE) arguments often use the non-differentiable $\infty$-norm directly. This paper develops a smooth polynomial Lyapunov-function-based stability certificate for convergence of Q-learning by transferring $\infty$-norm contraction to a weighted degree-$2p$ polynomial Lyapunov function induced by a finite $2p$-norm. The framework is conceptual and structural: it avoids non-differentiability, handles preconditioned dynamics arising in Q-learning and its variants, and gives a unified stability argument for standard Q-learning and smooth variants based on log-sum-exp (LSE), mellowmax, and Boltzmann softmax operators. For contractive operators, including the max, LSE, and mellowmax cases, the associated ODEs are globally exponentially stable and, under the stated independent and identically distributed (i.i.d.) sampling model, the stochastic approximation iterates converge almost surely. For the Boltzmann operator, which need not be contractive, the same framework yields convergence to an explicit invariant error set around the optimal Q-function. The resulting theory is not intended as a finite-time bound, but as a clean ODE foundation that unifies and simplifies asymptotic analyses of Q-learning and its smooth variants.

cs.LG↗

Switching Theory for Q-Learning

Q-learning is a fundamental algorithmic primitive in reinforcement learning. This paper develops a new framework for analyzing constant step-size tabular Q-learning from a switching linear system (SLS) viewpoint. In particular, we derive a stochastic SLS representation of the Q-learning error, and a finite-time error analysis through the joint spectral radius (JSR) of the corresponding SLS model, where the JSR is the exact worst-case exponential rate of the associated SLS. To the best of our knowledge, this is the first convergence rate analysis of standard Q-learning whose leading exponential rate is expressed through the JSR. The resulting rate is tied to the intrinsic worst-case exponential rate of the direct SLS representation and can be sharper than row-sum upper bounds when those bounds are conservative. We further prove that the JSR of Q-learning equals the largest spectral radius among the deterministic-policy modes and give an exact linear programming characterization that can be evaluated to any prescribed accuracy.

cs.LG↗

Adaptive Policy Backbone via Shared Network

Reinforcement learning (RL) has achieved impressive results across domains, yet learning an optimal policy typically requires extensive interaction data, limiting practical deployment. A common remedy is to leverage priors, such as pre-collected datasets or reference policies, but their utility degrades under task mismatch between training and deployment. While prior work has sought to address this mismatch, it has largely been restricted to in-distribution settings. To address this challenge, we propose Adaptive Policy Backbone (APB), a meta-transfer RL method that inserts lightweight linear layers before and after a shared backbone, thereby enabling parameter-efficient fine-tuning (PEFT) while preserving prior knowledge during adaptation. Our results show that APB improves sample efficiency over standard RL and adapts to out-of-distribution (OOD) tasks where existing meta-RL baselines typically fail.

cs.LG↗

Explore Beyond the Boundary Using Entropic Information

In reinforcement learning, exploration with sparse and delayed rewards presents a significant challenge due to the limited feedback available for guiding the learning process. Addressing this issue requires extensive exploration in the state space to discover valuable reward signals. In this paper, we propose Entropic Information for Exploration (ENTINEX), a novel method that enhances exploration by incentivizing agents to explore beyond the boundaries of the state distribution. ENTINEX achieves this by assigning intrinsic rewards to these boundaries, leveraging entropic information to identify them effectively. Through extensive experimentation, we demonstrate that ENTINEX consistently improves exploration performance in environments characterized by sparse and delayed rewards. Our experimental results show that ENTINEX outperforms existing exploration methods, highlighting its effectiveness in both sparse and delayed reward scenarios.

cs.LG↗

Spectral Analysis of Heavy-Ball Q-value Iteration

We study the convergence and acceleration of Qvalue iteration (QVI) with heavy-ball momentum. Although acceleration of value iteration has been studied extensively, there has been less work on the acceleration of heavy-ball QVI for control tasks. We analyze heavy-ball QVI from the viewpoint of switched linear system (SLS) theory and the joint spectral radius (JSR). First, we convert heavy-ball QVI into an exact SLS and interpret its convergence through the JSR. We also give JSR-based conditions under which heavy-ball QVI can be faster than standard QVI.

math.OC↗

MATANet: A Multi-context Attention and Taxonomy-Aware Network for Fine-Grained Underwater Recognition of Marine Species

Accurate fine-grained recognition of marine organisms is important for scalable biodiversity monitoring and ecological assessment using underwater imagery. However, existing methods mainly focus on target appearance and make limited use of surrounding environmental cues and biological taxonomy. We propose the Multi-Context Attention and Taxonomy-Aware Network (MATANet) for region-of-interest (ROI)-guided marine organism recognition. MATANet contains two complementary components. The Multi-Context Environmental Attention Module uses the ROI representation as a query to aggregate spatial patch features from ROI-centered contextual views at multiple scales, enabling target-conditioned modeling of the surrounding environment. Level-wise auxiliary classifiers further incorporate higher taxonomic ranks during training, encouraging hierarchically consistent representations without changing the finest-label prediction space or inference procedure. On the official FathomNet 2025 Private test split, MATANet achieves a hierarchical distance of 1.570 with the base backbone and 1.423 with the large backbone, substantially outperforming the strongest benchmark value of 2.603. On FishCLEF2015, MATANet achieves an accuracy of 0.793 and a hierarchical distance of 1.120, outperforming the strongest benchmark values of 0.766 and 1.327, respectively. Ablation studies show that surrounding scene information provides complementary evidence beyond repeated multi-scale observations of the target and that target-conditioned aggregation outperforms direct multi-view concatenation. Additional experiments on FAIR1M v2.0 examine the applicability of the proposed design beyond underwater imagery. In post-detection evaluation, MATANet improves fine-grained classification accuracy on matched detector-generated ROIs from 0.828 to 0.959 supporting its engineering applicability to automated marine monitoring.

cs.CV↗

Spectral Analysis of Dueling Q-Learning

Q-learning is a fundamental algorithm in reinforcement learning (RL) for solving discounted Markov decision processes (MDPs) when the transition kernel is unknown. The deep Q-network (DQN) extends Q-learning by using a deep neural network for Q-function approximation, which makes Q-learning applicable to more practical high-dimensional problems. Dueling Q-learning decomposes the Q-function into a value function and an advantage function and learns the two components jointly, which can improve learning efficiency. However, the theoretical understanding of dueling Q-learning is still limited. Recent work has initiated an analysis of tabular dueling Q-learning, but existing guarantees focus on a regularized formulation and leave the pure tabular update less completely understood. This paper strengthens that line of analysis by adding a direct interpretation of the centered tabular decomposition and by establishing convergence guarantees for the unregularized, unprojected constant step-size recursion. In particular, we derive an exact switching linear system representation for deterministic dueling Q-learning and a finite-time error bound in expectation for the sampled stochastic version. The analysis clarifies how the value and advantage updates act as different gains on the action-common (value function) and action-differential (advantage function) components of the Q-function.

cs.LG↗

Sign-Separated Asymmetric Finite-Time Error Analysis of Q-Learning

Q-learning is known to suffer from overestimation bias: because the Bellman update maximizes noisy or imperfect action-value estimates, positive errors can be selected and propagated, causing learned values to exceed the true optimal values. This bias can slow learning, degrade policy quality, and make value estimates unreliable. Although the convergence of Q-learning has been studied extensively, convergence theory that explicitly reflects this overestimation mechanism remains limited. This paper studies the asymmetric convergence behavior of Q-learning induced by overestimation bias. We decompose the Q-learning error into its componentwise positive and negative parts and derive separate finite-time rates for the two components. The resulting certificates can assign a slower exponential envelope to the positive component than to the negative component. This rate separation provides indirect theoretical evidence for max-induced overestimation: positive errors can be amplified through the maximization step, whereas negative errors admit a sharper comparison with an optimal-policy system. The separation is a difference between upper bounds, so it need not hold for every realized Q-learning trajectory. Nevertheless, we construct examples in which the predicted asymmetry appears in the actual trajectory. The analysis gives deterministic and stochastic constant-step-size bounds and clarifies how overestimation enters the switching-system dynamics of Q-learning.

cs.AI↗

Heavy-Ball Q-Learning with Residual Weighting Correction

This paper proposes a corrected heavy-ball Q-learning method for reinforcement learning (RL) and establishes convergence of its deterministic mean dynamics. It also identifies conditions under which the method is theoretically guaranteed to converge faster than standard Q-learning. The same construction is then extended to Q-learning with linear function approximation, where analogous convergence and acceleration statements are derived for the corresponding corrected fixed point. The sampled stochastic versions are treated through conditional-mean recursions and, in the stated linear-function-approximation setting, finite-time bounds. The analysis is based on a switched linear system (SLS) representation of Q-learning algorithms and on the joint spectral radius (JSR) of the associated switching families. This SLS viewpoint is not commonly used in standard analyses of Q-learning, and it provides a complementary framework and new insight into how heavy-ball momentum can accelerate Q-learning.

cs.LG↗

A Switching System Theory of Q-Learning with Linear Function Approximation

Q-learning is a fundamental algorithmic primitive in reinforcement learning. This paper develops a new framework for analyzing linear Q-learning from a switching linear system (SLS) viewpoint, where linear Q-learning denotes Q-learning with linear function approximation. We derive a stochastic SLS representation of the linear Q-learning error and obtain a finite-time error analysis for linear Q-learning through the joint spectral radius (JSR) of the associated SLS family; the JSR is the exact worst-case exponential rate of the corresponding SLSs. The JSR-based rate is tied to the intrinsic worst-case exponential rate of the SLS representation. Moreover, we provide a JSR-based certificate for convergence of linear Q-learning, which can be less conservative than one-step norm bounds.

cs.LG↗