arXiv subjects

Jiang Bian

Publications and source records attributed to Jiang Bian.

At least 127 records · Page 7Linked to original sources

arXiv2025

Policy Filtration for RLHF to Mitigate Noise in Reward Models

While direct policy optimization methods exist, pioneering LLMs are fine-tuned with reinforcement learning from human feedback (RLHF) to generate better responses under the supervision of a reward model learned from preference data. One major challenge of RLHF is the inaccuracy of the intermediate reward model, especially in the tasks that requires complex reasoning for the reward model to score a response. We find that the reliability of the reward model varies across responses assigned with different rewards. This motivates us to filter the samples whose rewards may be unreliable to improve the signal-to-noise ratio during policy learning, resulting in Policy Filtration for Proximal Policy Optimization (PF-PPO). To choose a proper policy filtering strategy, we use the coefficient of determination (R2) between the rewards and actual scores on filtered samples as the metrics to help us find promising strategies since it measures how well the rewards filtered by PF-PPO indicate real performance. We provide extensive experiments to validate the effectiveness of PF-PPO in code generation and math reasoning tasks. In code generation, PF-PPO achieves the state-of-the-art performance of 7-billion-parameter models on HumanEval (+7.9%), MBPP (+0.7%), and LeetCode Contest (+10.0%) which is a more challenging benchmark created by us. In math reasoning, PF-PPO yields performance increase using different reward models and benchmarks (Ape210K and CMATH). Code is available on https://github.com/DtYXs/verl/tree/pf-ppo.

cs.LG↗

arXiv2025

Unveiling Markov Heads in Pretrained Language Models for Offline Reinforcement Learning

Recently, incorporating knowledge from pretrained language models (PLMs) into decision transformers (DTs) has generated significant attention in offline reinforcement learning (RL). These PLMs perform well in RL tasks, raising an intriguing question: what kind of knowledge from PLMs has been transferred to RL to achieve such good results? This work first dives into this problem by analyzing each head quantitatively and points out Markov head, a crucial component that exists in the attention heads of PLMs. It leads to extreme attention on the last-input token and performs well only in short-term environments. Furthermore, we prove that this extreme attention cannot be changed by re-training embedding layer or fine-tuning. Inspired by our analysis, we propose a general method GPT2-DTMA, which equips a pretrained DT with Mixture of Attention (MoA), to accommodate diverse attention requirements during fine-tuning. Extensive experiments corroborate our theorems and demonstrate the effectiveness of GPT2-DTMA: it achieves comparable performance in short-term environments while significantly narrowing the performance gap in long-term environments.

cs.LG↗

arXiv2025

ELEVATE-GenAI: Reporting Guidelines for the Use of Large Language Models in Health Economics and Outcomes Research: an ISPOR Working Group on Generative AI Report

Introduction: Generative artificial intelligence (AI), particularly large language models (LLMs), holds significant promise for Health Economics and Outcomes Research (HEOR). However, standardized reporting guidance for LLM-assisted research is lacking. This article introduces the ELEVATE GenAI framework and checklist - reporting guidelines specifically designed for HEOR studies involving LLMs. Methods: The framework was developed through a targeted literature review of existing reporting guidelines, AI evaluation frameworks, and expert input from the ISPOR Working Group on Generative AI. It comprises ten domains, including model characteristics, accuracy, reproducibility, and fairness and bias. The accompanying checklist translates the framework into actionable reporting items. To illustrate its use, the framework was applied to two published HEOR studies: one focused on systematic literature review tasks and the other on economic modeling. Results: The ELEVATE GenAI framework offers a comprehensive structure for reporting LLM-assisted HEOR research, while the checklist facilitates practical implementation. Its application to the two case studies demonstrates its relevance and usability across different HEOR contexts. Limitations: Although the framework provides robust reporting guidance, further empirical testing is needed to assess its validity, completeness, usability, as well as its generalizability across diverse HEOR use cases. Conclusion: The ELEVATE GenAI framework and checklist address a critical gap by offering structured guidance for transparent, accurate, and reproducible reporting of LLM-assisted HEOR research. Future work will focus on extensive testing and validation to support broader adoption and refinement.

cs.CY↗

arXiv2025

Text-to-CAD Generation Through Infusing Visual Feedback in Large Language Models

Creating Computer-Aided Design (CAD) models requires significant expertise and effort. Text-to-CAD, which converts textual descriptions into CAD parametric sequences, is crucial in streamlining this process. Recent studies have utilized ground-truth parametric sequences, known as sequential signals, as supervision to achieve this goal. However, CAD models are inherently multimodal, comprising parametric sequences and corresponding rendered visual objects. Besides,the rendering process from parametric sequences to visual objects is many-to-one. Therefore, both sequential and visual signals are critical for effective training. In this work, we introduce CADFusion, a framework that uses Large Language Models (LLMs) as the backbone and alternates between two training stages: the sequential learning (SL) stage and the visual feedback (VF) stage. In the SL stage, we train LLMs using ground-truth parametric sequences, enabling the generation of logically coherent parametric sequences. In the VF stage, we reward parametric sequences that render into visually preferred objects and penalize those that do not, allowing LLMs to learn how rendered visual objects are perceived and evaluated. These two stages alternate throughout the training, ensuring balanced learning and preserving benefits of both signals. Experiments demonstrate that CADFusion significantly improves performance, both qualitatively and quantitatively.

cs.CV↗

arXiv2025

AdaptiveStep: Automatically Dividing Reasoning Step through Model Confidence

Current approaches for training Process Reward Models (PRMs) often involve breaking down responses into multiple reasoning steps using rule-based techniques, such as using predefined placeholder tokens or setting the reasoning step's length into a fixed size. These approaches overlook the fact that specific words do not typically mark true decision points in a text. To address this, we propose AdaptiveStep, a method that divides reasoning steps based on the model's confidence in predicting the next word. This division method provides more decision-making information at each step, enhancing downstream tasks, such as reward model learning. Moreover, our method does not require manual annotation. We demonstrate its effectiveness through experiments with AdaptiveStep-trained PRMs in mathematical reasoning and code generation tasks. Experimental results indicate that the outcome PRM achieves state-of-the-art Best-of-N performance, surpassing greedy search strategy with token-level value-guided decoding, while also reducing construction costs by over 30% compared to existing open-source PRMs. In addition, we provide a thorough analysis and case study on the PRM's performance, transferability, and generalization capabilities.

cs.AI↗

arXiv2025

Learning to Select In-Context Demonstration Preferred by Large Language Model

In-context learning (ICL) enables large language models (LLMs) to adapt to new tasks during inference using only a few demonstrations. However, ICL performance is highly dependent on the selection of these demonstrations. Recent work explores retrieval-based methods for selecting query-specific demonstrations, but these approaches often rely on surrogate objectives such as metric learning, failing to directly optimize ICL performance. Consequently, they struggle to identify truly beneficial demonstrations. Moreover, their discriminative retrieval paradigm is ineffective when the candidate pool lacks sufficient high-quality demonstrations. To address these challenges, we propose GenICL, a novel generative preference learning framework that leverages LLM feedback to directly optimize demonstration selection for ICL. Experiments on 19 datasets across 11 task categories demonstrate that GenICL achieves superior performance than existing methods in selecting the most effective demonstrations, leading to better ICL performance.

cs.LG↗

arXiv2025

DPO Meets PPO: Reinforced Token Optimization for RLHF

In the classical Reinforcement Learning from Human Feedback (RLHF) framework, Proximal Policy Optimization (PPO) is employed to learn from sparse, sentence-level rewards -- a challenging scenario in traditional deep reinforcement learning. Despite the great successes of PPO in the alignment of large language models, its open-source implementation is still largely sub-optimal. To address these issues, we introduce a framework that models RLHF problems as a Markov decision process (MDP), enabling the capture of fine-grained token-wise information. Under this framework, we introduce an algorithm Reinforced Token Optimization (\texttt{RTO}), which learns the token-wise reward function from preference data and performs policy optimization based on this learned token-wise reward signal. Theoretically, \texttt{RTO} is proven to have the capability of finding the near-optimal policy sample-efficiently. For its practical implementation, \texttt{RTO} innovatively integrates Direct Preference Optimization (DPO) and PPO. DPO, originally derived from sparse sentence rewards, surprisingly provides us with a token-wise characterization of response quality, which is seamlessly incorporated into our subsequent PPO training stage. Extensive experiments demonstrate that \texttt{RTO} performs better than PPO and other direct preference learning algorithms. In particular, RTO outperforms PPO by 7.5 points on the AlpacaEval 2 benchmark and by 4.1 points on Arena-Hard. Our code and models are available at \href{https://github.com/zkshan2002/RTO}{https://github.com/zkshan2002/RTO}.

cs.LG↗

arXiv2025

C-MORL: Multi-Objective Reinforcement Learning through Efficient Discovery of Pareto Front

Multi-objective reinforcement learning (MORL) excels at handling rapidly changing preferences in tasks that involve multiple criteria, even for unseen preferences. However, previous dominating MORL methods typically generate a fixed policy set or preference-conditioned policy through multiple training iterations exclusively for sampled preference vectors, and cannot ensure the efficient discovery of the Pareto front. Furthermore, integrating preferences into the input of policy or value functions presents scalability challenges, in particular as the dimension of the state and preference space grow, which can complicate the learning process and hinder the algorithm's performance on more complex tasks. To address these issues, we propose a two-stage Pareto front discovery algorithm called Constrained MORL (C-MORL), which serves as a seamless bridge between constrained policy optimization and MORL. Concretely, a set of policies is trained in parallel in the initialization stage, with each optimized towards its individual preference over the multiple objectives. Then, to fill the remaining vacancies in the Pareto front, the constrained optimization steps are employed to maximize one objective while constraining the other objectives to exceed a predefined threshold. Empirically, compared to recent advancements in MORL methods, our algorithm achieves more consistent and superior performances in terms of hypervolume, expected utility, and sparsity on both discrete and continuous control tasks, especially with numerous objectives (up to nine objectives in our experiments).

cs.LG↗

arXiv2025

EvoPrompt: Connecting LLMs with Evolutionary Algorithms Yields Powerful Prompt Optimizers

Large Language Models (LLMs) excel in various tasks, but they rely on carefully crafted prompts that often demand substantial human effort. To automate this process, in this paper, we propose a novel framework for discrete prompt optimization, called EvoPrompt, which borrows the idea of evolutionary algorithms (EAs) as they exhibit good performance and fast convergence. To enable EAs to work on discrete prompts, which are natural language expressions that need to be coherent and human-readable, we connect LLMs with EAs. This approach allows us to simultaneously leverage the powerful language processing capabilities of LLMs and the efficient optimization performance of EAs. Specifically, abstaining from any gradients or parameters, EvoPrompt starts from a population of prompts and iteratively generates new prompts with LLMs based on the evolutionary operators, improving the population based on the development set. We optimize prompts for both closed- and open-source LLMs including GPT-3.5 and Alpaca, on 31 datasets covering language understanding, generation tasks, as well as BIG-Bench Hard (BBH) tasks. EvoPrompt significantly outperforms human-engineered prompts and existing methods for automatic prompt generation (e.g., up to 25% on BBH). Furthermore, EvoPrompt demonstrates that connecting LLMs with EAs creates synergies, which could inspire further research on the combination of LLMs and conventional algorithms.

cs.CL↗

arXiv2025

MAP: Low-compute Model Merging with Amortized Pareto Fronts via Quadratic Approximation

Model merging has emerged as an effective approach to combine multiple single-task models into a multitask model. This process typically involves computing a weighted average of the model parameters without any additional training. Existing model-merging methods focus on enhancing average task accuracy. However, interference and conflicts between the objectives of different tasks can lead to trade-offs during the merging process. In real-world applications, a set of solutions with various trade-offs can be more informative, helping practitioners make decisions based on diverse preferences. In this paper, we introduce a novel and low-compute algorithm, Model Merging with Amortized Pareto Front (MAP). MAP efficiently identifies a Pareto set of scaling coefficients for merging multiple models, reflecting the trade-offs involved. It amortizes the substantial computational cost of evaluations needed to estimate the Pareto front by using quadratic approximation surrogate models derived from a pre-selected set of scaling coefficients. Experimental results on vision and natural language processing tasks demonstrate that MAP can accurately identify the Pareto front, providing practitioners with flexible solutions to balance competing task objectives. We also introduce Bayesian MAP for scenarios with a relatively low number of tasks and Nested MAP for situations with a high number of tasks, further reducing the computational cost of evaluation.

cs.LG↗

arXiv2025

TarDiff: Target-Oriented Diffusion Guidance for Synthetic Electronic Health Record Time Series Generation

Synthetic Electronic Health Record (EHR) time-series generation is crucial for advancing clinical machine learning models, as it helps address data scarcity by providing more training data. However, most existing approaches focus primarily on replicating statistical distributions and temporal dependencies of real-world data. We argue that fidelity to observed data alone does not guarantee better model performance, as common patterns may dominate, limiting the representation of rare but important conditions. This highlights the need for generate synthetic samples to improve performance of specific clinical models to fulfill their target outcomes. To address this, we propose TarDiff, a novel target-oriented diffusion framework that integrates task-specific influence guidance into the synthetic data generation process. Unlike conventional approaches that mimic training data distributions, TarDiff optimizes synthetic samples by quantifying their expected contribution to improving downstream model performance through influence functions. Specifically, we measure the reduction in task-specific loss induced by synthetic samples and embed this influence gradient into the reverse diffusion process, thereby steering the generation towards utility-optimized data. Evaluated on six publicly available EHR datasets, TarDiff achieves state-of-the-art performance, outperforming existing methods by up to 20.4% in AUPRC and 18.4% in AUROC. Our results demonstrate that TarDiff not only preserves temporal fidelity but also enhances downstream model performance, offering a robust solution to data scarcity and class imbalance in healthcare analytics.

cs.LG↗

arXiv2025

Low latency global carbon budget reveals a continuous decline of the land carbon sink during the 2023/24 El Nino event

The high growth rate of atmospheric CO2 in 2023 was found to be caused by a severe reduction of the global net land carbon sink. Here we update the global CO2 budget from January 1st to July 1st 2024, during which El Niño drought conditions continued to prevail in the Tropics but ceased by March 2024. We used three dynamic global vegetation models (DGVMs), machine learning emulators of ocean models, three atmospheric inversions driven by observations from the second Orbiting Carbon Observatory (OCO-2) satellite, and near-real-time fossil CO2 emissions estimates. In a one-year period from July 2023 to July 2024 covering the El Niño 2023/24 event, we found a record-high CO2 growth rate of 3.66~$\pm$~0.09 ppm~yr$^{-1}$ ($\pm$~1 standard deviation) since 1979. Yet, the CO2 growth rate anomaly obtained after removing the long term trend is 1.1 ppm~yr$^{-1}$, which is marginally smaller than the July--July growth rate anomalies of the two major previous El Niño events in 1997/98 and 2015/16. The atmospheric CO2 growth rate anomaly was primarily driven by a 2.24 GtC~yr$^{-1}$ reduction in the net land sink including 0.3 GtC~yr$^{-1}$ of fire emissions, partly offset by a 0.38 GtC~yr$^{-1}$ increase in the ocean sink relative to the 2015--2022 July--July mean. The tropics accounted for 97.5\% of the land CO2 flux anomaly, led by the Amazon (50.6\%), central Africa (34\%), and Southeast Asia (8.2\%), with extra-tropical sources in South Africa and southern Brazil during April--July 2024. Our three DGVMs suggest greater tropical CO2 losses in 2023/2024 than during the two previous large El Niño in 1997/98 and 2015/16, whereas inversions indicate losses more comparable to 2015/16. Overall, this update of the low latency budget highlights the impact of recent El Niño droughts in explaining the high CO2 growth rate until July 2024.

physics.ao-ph↗

arXiv2025

MineWorld: a Real-Time and Open-Source Interactive World Model on Minecraft

World modeling is a crucial task for enabling intelligent agents to effectively interact with humans and operate in dynamic environments. In this work, we propose MineWorld, a real-time interactive world model on Minecraft, an open-ended sandbox game which has been utilized as a common testbed for world modeling. MineWorld is driven by a visual-action autoregressive Transformer, which takes paired game scenes and corresponding actions as input, and generates consequent new scenes following the actions. Specifically, by transforming visual game scenes and actions into discrete token ids with an image tokenizer and an action tokenizer correspondingly, we consist the model input with the concatenation of the two kinds of ids interleaved. The model is then trained with next token prediction to learn rich representations of game states as well as the conditions between states and actions simultaneously. In inference, we develop a novel parallel decoding algorithm that predicts the spatial redundant tokens in each frame at the same time, letting models in different scales generate $4$ to $7$ frames per second and enabling real-time interactions with game players. In evaluation, we propose new metrics to assess not only visual quality but also the action following capacity when generating new scenes, which is crucial for a world model. Our comprehensive evaluation shows the efficacy of MineWorld, outperforming SoTA open-sourced diffusion based world models significantly. The code and model have been released.

cs.CV↗

arXiv2025

DAE-Talker: High Fidelity Speech-Driven Talking Face Generation with Diffusion Autoencoder

While recent research has made significant progress in speech-driven talking face generation, the quality of the generated video still lags behind that of real recordings. One reason for this is the use of handcrafted intermediate representations like facial landmarks and 3DMM coefficients, which are designed based on human knowledge and are insufficient to precisely describe facial movements. Additionally, these methods require an external pretrained model for extracting these representations, whose performance sets an upper bound on talking face generation. To address these limitations, we propose a novel method called DAE-Talker that leverages data-driven latent representations obtained from a diffusion autoencoder (DAE). DAE contains an image encoder that encodes an image into a latent vector and a DDIM image decoder that reconstructs the image from it. We train our DAE on talking face video frames and then extract their latent representations as the training target for a Conformer-based speech2latent model. This allows DAE-Talker to synthesize full video frames and produce natural head movements that align with the content of speech, rather than relying on a predetermined head pose from a template video. We also introduce pose modelling in speech2latent for pose controllability. Additionally, we propose a novel method for generating continuous video frames with the DDIM image decoder trained on individual frames, eliminating the need for modelling the joint distribution of consecutive frames directly. Our experiments show that DAE-Talker outperforms existing popular methods in lip-sync, video fidelity, and pose naturalness. We also conduct ablation studies to analyze the effectiveness of the proposed techniques and demonstrate the pose controllability of DAE-Talker.

cs.CV↗

arXiv2025

Diffusion Models in 3D Vision: A Survey

In recent years, 3D vision has become a crucial field within computer vision, powering a wide range of applications such as autonomous driving, robotics, augmented reality, and medical imaging. This field relies on accurate perception, understanding, and reconstruction of 3D scenes from 2D images or text data sources. Diffusion models, originally designed for 2D generative tasks, offer the potential for more flexible, probabilistic methods that can better capture the variability and uncertainty present in real-world 3D data. In this paper, we review the state-of-the-art methods that use diffusion models for 3D visual tasks, including but not limited to 3D object generation, shape completion, point-cloud reconstruction, and scene construction. We provide an in-depth discussion of the underlying mathematical principles of diffusion models, outlining their forward and reverse processes, as well as the various architectural advancements that enable these models to work with 3D datasets. We also discuss the key challenges in applying diffusion models to 3D vision, such as handling occlusions and varying point densities, and the computational demands of high-dimensional data. Finally, we discuss potential solutions, including improving computational efficiency, enhancing multimodal fusion, and exploring the use of large-scale pretraining for better generalization across 3D tasks. This paper serves as a foundation for future exploration and development in this rapidly evolving field.

cs.CV↗

arXiv2025

VidTwin: Video VAE with Decoupled Structure and Dynamics

Recent advancements in video autoencoders (Video AEs) have significantly improved the quality and efficiency of video generation. In this paper, we propose a novel and compact video autoencoder, VidTwin, that decouples video into two distinct latent spaces: Structure latent vectors, which capture overall content and global movement, and Dynamics latent vectors, which represent fine-grained details and rapid movements. Specifically, our approach leverages an Encoder-Decoder backbone, augmented with two submodules for extracting these latent spaces, respectively. The first submodule employs a Q-Former to extract low-frequency motion trends, followed by downsampling blocks to remove redundant content details. The second averages the latent vectors along the spatial dimension to capture rapid motion. Extensive experiments show that VidTwin achieves a high compression rate of 0.20% with high reconstruction quality (PSNR of 28.14 on the MCL-JCV dataset), and performs efficiently and effectively in downstream generative tasks. Moreover, our model demonstrates explainability and scalability, paving the way for future research in video latent representation and generation. Check our project page for more details: https://vidtwin.github.io/.

cs.CV↗

arXiv2025

Chemistry-aware battery degradation prediction under simulated real-world cyclic protocols

Battery degradation is governed by complex and randomized cyclic conditions, yet existing modeling and prediction frameworks usually rely on rigid, unchanging protocols that fail to capture real-world dynamics. The stochastic electrical signals make such prediction extremely challenging, while, on the other hand, they provide abundant additional information, such as voltage fluctuations, which may probe the degradation mechanisms. Here, we present chemistry-aware battery degradation prediction under dynamic conditions with machine learning, which integrates hidden Markov processes for realistic power simulations, an automated batch-testing system that generates a large electrochemical dataset under randomized conditions, an interfacial chemistry database derived from high-throughput X-ray photoelectron spectroscopy for mechanistic probing, and a machine learning model for prediction. By automatically constructing a polynomial-scale feature space from irregular electrochemical curves, our model accurately predicts both battery life and critical knee points. This feature space also predicts the composition of the solid electrolyte interphase, revealing six distinct failure mechanisms-demonstrating a viable approach to use electrical signals to infer interfacial chemistry. This work establishes a scalable and adaptive framework for integrating chemical engineering and data science to advance noninvasive diagnostics and optimize processes for more durable and sustainable energy storage technologies.

eess.SP↗

arXiv2025

Chain of Functions: A Programmatic Pipeline for Fine-Grained Chart Reasoning Data

Visual reasoning is crucial for multimodal large language models (MLLMs) to address complex chart queries, yet high-quality rationale data remains scarce. Existing methods leveraged (M)LLMs for data generation, but direct prompting often yields limited precision and diversity. In this paper, we propose \textit{Chain of Functions (CoF)}, a novel programmatic reasoning data generation pipeline that utilizes freely-explored reasoning paths as supervision to ensure data precision and diversity. Specifically, it starts with human-free exploration among the atomic functions (e.g., maximum data and arithmetic operations) to generate diverse function chains, which are then translated into linguistic rationales and questions with only a moderate open-sourced LLM. \textit{CoF} provides multiple benefits: 1) Precision: function-governed generation reduces hallucinations compared to freeform generation; 2) Diversity: enumerating function chains enables varied question taxonomies; 3) Explainability: function chains serve as built-in rationales, allowing fine-grained evaluation beyond overall accuracy; 4) Practicality: eliminating reliance on extremely large models. Employing \textit{CoF}, we construct the \textit{ChartCoF} dataset, with 1.4k complex reasoning Q\&A for fine-grained analysis and 50k Q\&A for reasoning enhancement. The fine-grained evaluation on \textit{ChartCoF} reveals varying performance across question taxonomies for each MLLM, and the experiments also show that finetuning with \textit{ChartCoF} achieves state-of-the-art performance among same-scale MLLMs on widely used benchmarks. Furthermore, the novel paradigm of function-governed rationale generation in \textit{CoF} could inspire broader applications beyond charts.

cs.CV↗