arXiv subjects

Wenpin Tang

Publications and source records attributed to Wenpin Tang.

At least 37 records · Page 2Linked to original sources

arXiv2026

Improved techniques for fine-tuning flow models via adjoint matching: a deterministic control pipeline

We propose a deterministic adjoint matching framework that formulates human preference alignment for flow-based generative models as an optimal control problem over velocity fields. One can directly regress the control toward a value-gradient-induced target under the current policy, leading to a simple and stable training objective. Building on this perspective, we introduce a truncated adjoint scheme that focuses computation on the terminal portion of the trajectory, where reward-relevant signals concentrate, which yields substantial computational savings while preserving alignment quality. We further generalize the framework beyond standard KL-based regularization, allowing more flexible trade-offs between alignment strength and distributional preservation. Experiments on SiT-XL/2 and FLUX.2-Klein-4B demonstrate consistent gains across multiple alignment metrics, along with substantially improved diversity and mode preservation.

cs.AI↗

arXiv2026

RPO: Fine-Tuning Visual Generative Models via Rich Vision-Language Preferences

Traditional preference tuning methods for LLMs/Visual Generative Models often rely solely on reward model labeling, which can be opaque, offer limited insights into the rationale behind preferences, and are prone to issues such as reward hacking or overfitting. We introduce Rich Preference Optimization (RPO), a novel pipeline that leverages rich feedback signals from Vision Language Models (VLMs) to improve the curation of preference pairs for fine-tuning visual generative models like text-to-image diffusion models. Our approach begins with prompting VLMs to generate detailed critiques of synthesized images, from which we further prompt VLMs to extract reliable and actionable image editing instructions. By implementing these instructions, we create refined images, resulting in synthetic, informative preference pairs that serve as enhanced tuning datasets. We demonstrate the effectiveness of our pipeline and the resulting datasets in fine-tuning state-of-the-art diffusion models.

cs.LG↗

arXiv2026

DiFFPO: Training Diffusion LLMs to Reason Fast and Furious via Reinforcement Learning

We propose DiFFPO, Diffusion Fast and Furious Policy Optimization, a unified framework for training masked diffusion large language models (dLLMs) to reason not only better (furious), but also faster via reinforcement learning (RL). We first unify the existing baseline approach such as d1 by proposing to train surrogate policies via off-policy RL, whose likelihood is much more tractable as an approximation to the true dLLM policy. This naturally motivates a more accurate and informative two-stage likelihood approximation combined with importance sampling correction, which leads to generalized RL algorithms with better sample efficiency and superior task performance. Second, we propose a new direction of joint training efficient samplers/controllers of dLLMs policy. Via RL, we incentivize dLLMs' natural multi-token prediction capabilities by letting the model learn to adaptively allocate an inference threshold for each prompt. By jointly training the sampler, we yield better accuracies with lower number of function evaluations (NFEs) compared to training the model only, obtaining the best performance in improving the Pareto frontier of the inference-time compute of dLLMs. We showcase the effectiveness of our pipeline by training open source large diffusion language models over benchmark math and planning tasks.

cs.LG↗

arXiv2025

Understanding Sampler Stochasticity in Training Diffusion Models for RLHF

Reinforcement Learning from Human Feedback (RLHF) is increasingly used to fine-tune diffusion models, but a key challenge arises from the mismatch between stochastic samplers used during training and deterministic samplers used during inference. In practice, models are fine-tuned using stochastic SDE samplers to encourage exploration, while inference typically relies on deterministic ODE samplers for efficiency and stability. This discrepancy induces a reward gap, raising concerns about whether high-quality outputs can be expected during inference. In this paper, we theoretically characterize this reward gap and provide non-vacuous bounds for general diffusion models, along with sharper convergence rates for Variance Exploding (VE) and Variance Preserving (VP) Gaussian models. Methodologically, we adopt the generalized denoising diffusion implicit models (gDDIM) framework to support arbitrarily high levels of stochasticity, preserving data marginals throughout. Empirically, our findings through large-scale experiments on text-to-image models using denoising diffusion policy optimization (DDPO) and mixed group relative policy optimization (MixGRPO) validate that reward gaps consistently narrow over training, and ODE sampling quality improves when models are updated using higher-stochasticity SDE training.

cs.LG↗

arXiv2025

Analysis of the Order Flow Auction under Proposer-Builder Separation on Blockchain

We study the impact of the order flow auction (OFA) in the context of the proposer-builder separation (PBS) mechanism in blockchains through a game-theoretic perspective. The OFA is designed to improve user welfare by redistributing maximal extractable value (MEV) to the users, in which two sequential auctions take place: the order flow auction and the block-building auction. We formulate the OFA as a multiplayer game, and establish the existence of a Nash equilibrium, and in the two-player case derive a closed-form solution (and prove its uniqueness) via a quartic equation. Our result shows that the builder with a competitive advantage pays a lower cost, leading to a higher revenue, and adding to centralization in the builder space. In contrast, the proposer's shares evolve as a martingale process, which implies decentralization in the proposer/validator space. Our analyses rely on various tools from stochastic processes, convex optimization, and polynomial equations. We also conduct numerical studies to corroborate our findings, and to bring out other features of the OFA under the PBS mechanism.

econ.TH↗

arXiv2025

Optimal Decisions for Liquid Staking: Allocation and Exit Timing

In this paper, we study an investor's optimal entry and exit decisions in a liquid staking protocol (LSP) and an automated market maker (AMM), primarily from the standpoint of the investor. Our analysis focuses on two key investor actions: the initial allocation decision at time $t=0$, and the optimal timing of exit. First, we derive an optimal allocation strategy that enables the investor to distribute risk across the LSP, AMM, and direct holding. Our results also offer insights for LSP and AMM designers, identifying the necessary and sufficient conditions under which the investor is incentivized to stake through an LSP, and further, to provide liquidity in addition to staking. These conditions include a lower bound on the transaction fee, for which we propose a fee mechanism that attains the bound. Second, given a fixed protocol design, we model the optimal exit timing of an individual investor using Laplace transforms and free-boundary techniques. We analyze scenarios with and without transaction fees. In the absence of fees, we decompose the investor's payoff into impermanent loss and opportunity cost, and provide theoretical results characterizing the investor's payoff and the optimal exit threshold. With transaction fees, we conduct numerical analyses to examine how fee accumulation influences exit strategies. Our results reveal that in both settings, a stop-loss strategy often maximizes the investor's expected payoff, driven by opportunity gains and the accumulation of fees where fees are present. Our analyses rely on various tools from stochastic processes and control theory, as well as convex optimization and analysis. We further support our theoretical insights with numerical experiments and explore additional properties of the investor's value function and optimal behavior.

q-fin.MF↗

arXiv2025

SOCRATES: Simulation Optimization with Correlated Replicas and Adaptive Trajectory Evaluations

The field of simulation optimization (SO) encompasses various methods developed to optimize complex, expensive-to-sample stochastic systems. Established methods include, but are not limited to, ranking-and-selection for finite alternatives and surrogate-based methods for continuous domains, with broad applications in engineering and operations management. The recent advent of large language models (LLMs) offers a new paradigm for exploiting system structure and automating the strategic selection and composition of these established SO methods into a tailored optimization procedure. This work introduces SOCRATES (Simulation Optimization with Correlated Replicas and Adaptive Trajectory Evaluations), a novel two-stage procedure that leverages LLMs to automate the design of tailored SO algorithms. The first stage constructs an ensemble of digital replicas of the real system. An LLM is employed to implement causal discovery from a textual description of the system, generating a structural `skeleton' that guides the sample-efficient learning of the replicas. In the second stage, this replica ensemble is used as an inexpensive testbed to evaluate a set of baseline SO algorithms. An LLM then acts as a meta-optimizer, analyzing the performance trajectories of these algorithms to iteratively revise and compose a final, hybrid optimization schedule. This schedule is designed to be adaptive, with the ability to be updated during the final execution on the real system when the optimization performance deviates from expectations. By integrating LLM-driven reasoning with LLM-assisted trajectory-aware meta-optimization, SOCRATES creates an effective and sample-efficient solution for complex SO optimization problems.

stat.ML↗

arXiv2025

The football model, stochastic ordering and martingale transport

Tournaments are competitions between a number of teams, the outcome of which determines the relative strength or rank of each team. In many cases, the strength of a team in the tournament is given by a score. Perhaps, the most striking mathematical result on the tournament is Moon's theorem, which provides a necessary and sufficient condition for a feasible score sequence via majorization. To give a probabilistic interpretation of Moon's result, Aldous and Kolesnik introduced the football model, the existence of which gives a short proof of Moon's theorem. However, the existence proof of Aldous and Kolesnik is nonconstructive, leading to the question of a ``canonical'' construction of the football model. The purpose of this paper is to provide explicit constructions of the football model with an additional stochastic ordering constraint, which can be formulated by martingale transport. Two solutions are given: one is by solving an entropy optimization problem via Sinkhorn's algorithm, and the other relies on the idea of shadow couplings. It turns out that both constructions yield the property of strong stochastic transitivity. The nontransitive situations of the football model are also considered.

math.PR↗

arXiv2025

Diffusion Generative Models Meet Compressed Sensing, with Applications to Imaging and Finance

In this study we develop dimension-reduction techniques to accelerate diffusion model inference in the context of synthetic data generation. The idea is to integrate compressed sensing into diffusion models (hence, CSDM): First, compress the dataset into a latent space (from an ambient space), and train a diffusion model in the latent space; next, apply a compressed sensing algorithm to the samples generated in the latent space for decoding back to the original space; and the goal is to facilitate the efficiency of both model training and inference. Under certain sparsity assumptions on data, our proposed approach achieves provably faster convergence, via combining diffusion model inference with sparse recovery. It also sheds light on the best choice of the latent space dimension. To illustrate the effectiveness of this approach, we run numerical experiments on a range of datasets, including handwritten digits, medical and climate images, and financial time series for stress testing.

stat.ML↗

arXiv2025

Bayesian Geostatistics Using Predictive Stacking

We develop Bayesian predictive stacking for geostatistical models, where the primary inferential objective is to provide inference on the latent spatial random field and conduct spatial predictions at arbitrary locations. We exploit analytically tractable posterior distributions for regression coefficients of predictors and the realizations of the spatial process conditional upon process parameters. We subsequently combine such inference by stacking these models across the range of values of the hyper-parameters. We devise stacking of means and posterior densities in a manner that is computationally efficient without resorting to iterative algorithms such as Markov chain Monte Carlo (MCMC) and can exploit the benefits of parallel computations. We offer novel theoretical insights into the resulting inference within an infill asymptotic paradigm and through empirical results showing that stacked inference is comparable to full sampling-based Bayesian inference at a significantly lower computational cost.

stat.ME↗

arXiv2025

Fine-tuning of diffusion models via stochastic control: entropy regularization and beyond

This paper aims to develop and provide a rigorous treatment to the problem of entropy regularized fine-tuning in the context of continuous-time diffusion models, which was recently proposed by Uehara et al. (arXiv:2402.15194, 2024). The idea is to use stochastic control for sample generation, where the entropy regularizer is introduced to mitigate reward collapse. We also show how the analysis can be extended to fine-tuning with a general $f$-divergence regularizer. Numerical experiments on large-scale text-to-image models--Stable Diffusion v1.5 are conducted to validate our approach.

math.OC↗

arXiv2025

Score as Action: Fine-Tuning Diffusion Generative Models by Continuous-time Reinforcement Learning

Reinforcement learning from human feedback (RLHF), which aligns a diffusion model with input prompt, has become a crucial step in building reliable generative AI models. Most works in this area use a discrete-time formulation, which is prone to induced discretization errors, and often not applicable to models with higher-order/black-box solvers. The objective of this study is to develop a disciplined approach to fine-tune diffusion models using continuous-time RL, formulated as a stochastic control problem with a reward function that aligns the end result (terminal state) with input prompt. The key idea is to treat score matching as controls or actions, and thereby making connections to policy optimization and regularization in continuous-time RL. To carry out this idea, we lay out a new policy optimization framework for continuous-time RL, and illustrate its potential in enhancing the value networks design space via leveraging the structural property of diffusion models. We validate the advantages of our method by experiments in downstream tasks of fine-tuning large-scale Text2Image models of Stable Diffusion v1.5.

cs.LG↗

arXiv2025

Score-based Diffusion Models via Stochastic Differential Equations -- a Technical Tutorial

This is an expository article on the score-based diffusion models, with a particular focus on the formulation via stochastic differential equations (SDE). After a gentle introduction, we discuss the two pillars in the diffusion modeling -- sampling and score matching, which encompass the SDE/ODE sampling, score matching efficiency, the consistency models, and reinforcement learning. Short proofs are given to illustrate the main idea of the stated results. The article is primarily a technical introduction to the field, and practitioners may also find some analysis useful in designing new models or algorithms.

cs.LG↗

arXiv2025

MallowsPO: Fine-Tune Your LLM with Preference Dispersions

Direct Preference Optimization (DPO) has recently emerged as a popular approach to improve reinforcement learning with human feedback (RLHF), leading to better techniques to fine-tune large language models (LLM). A weakness of DPO, however, lies in its lack of capability to characterize the diversity of human preferences. Inspired by Mallows' theory of preference ranking, we develop in this paper a new approach, the MallowsPO. A distinct feature of this approach is a dispersion index, which reflects the dispersion of human preference to prompts. We show that existing DPO models can be reduced to special cases of this dispersion index, thus unified with MallowsPO. More importantly, we demonstrate (empirically) how to use this dispersion index to enhance the performance of DPO in a broad array of benchmark tasks, from synthetic bandit selection to controllable generations and dialogues, while maintaining great generalization capabilities. MallowsPO is also compatible with other SOTA offline preference optimization methods, boosting nearly 2\% extra LC win rate when used as a plugin for fine-tuning Llama3-Instruct.

cs.LG↗

arXiv2025

Policy iteration for the deterministic control problems -- a viscosity approach

This paper is concerned with the convergence rate of policy iteration for (deterministic) optimal control problems in continuous time. To overcome the problem of ill-posedness due to lack of regularity, we consider a semi-discrete scheme by adding a viscosity term via finite differences in space. We prove that PI for the semi-discrete scheme converges exponentially fast, and provide a bound on the error induced by the semi-discrete scheme. We also consider the discrete space-time scheme, where both space and time are discretized. Convergence rate of PI and the discretization error are studied.

math.OC↗

arXiv2025

RainbowPO: A Unified Framework for Combining Improvements in Preference Optimization

Recently, numerous preference optimization algorithms have been introduced as extensions to the Direct Preference Optimization (DPO) family. While these methods have successfully aligned models with human preferences, there is a lack of understanding regarding the contributions of their additional components. Moreover, fair and consistent comparisons are scarce, making it difficult to discern which components genuinely enhance downstream performance. In this work, we propose RainbowPO, a unified framework that demystifies the effectiveness of existing DPO methods by categorizing their key components into seven broad directions. We integrate these components into a single cohesive objective, enhancing the performance of each individual element. Through extensive experiments, we demonstrate that RainbowPO outperforms existing DPO variants. Additionally, we provide insights to guide researchers in developing new DPO methods and assist practitioners in their implementations.

cs.AI↗

arXiv2024

Preference Tuning with Human Feedback on Language, Speech, and Vision Tasks: A Survey

Preference tuning is a crucial process for aligning deep generative models with human preferences. This survey offers a thorough overview of recent advancements in preference tuning and the integration of human feedback. The paper is organized into three main sections: 1) introduction and preliminaries: an introduction to reinforcement learning frameworks, preference tuning tasks, models, and datasets across various modalities: language, speech, and vision, as well as different policy approaches, 2) in-depth exploration of each preference tuning approach: a detailed analysis of the methods used in preference tuning, and 3) applications, discussion, and future directions: an exploration of the applications of preference tuning in downstream tasks, including evaluation methods for different modalities, and an outlook on future research directions. Our objective is to present the latest methodologies in preference tuning and model alignment, enhancing the understanding of this field for researchers and practitioners. We hope to encourage further engagement and innovation in this area.

cs.CL↗

arXiv2024

Contractive Diffusion Probabilistic Models

Diffusion probabilistic models (DPMs) have emerged as a promising technique in generative modeling. The success of DPMs relies on two ingredients: time reversal of diffusion processes and score matching. In view of possibly unguaranteed score matching, we propose a new criterion -- the contraction property of backward sampling in the design of DPMs, leading to a novel class of contractive DPMs (CDPMs). Our key insight is that, the contraction property can provably narrow score matching errors and discretization errors, thus our proposed CDPMs are robust to both sources of error. For practical use, we show that CDPM can leverage weights of pretrained DPMs by a simple transformation, and does not need retraining. We corroborated our approach by experiments on synthetic 1-dim examples, Swiss Roll, MNIST, CIFAR-10 32$\times$32 and AFHQ 64$\times$64 dataset. Notably, CDPM steadily improves the performance of baseline score-based diffusion models.

cs.LG↗