Haoxiang Wang

Niket Agarwal, Arslan Ali, Maciej Bala et al. · nvidia

Physical AI needs to be trained digitally first. It needs a digital twin of itself, the policy model, and a digital twin of the world, the world model. In this paper, we present the Cosmos World Foundation Model Platform to help developers build customized world models for their Physical AI setups. We position a world foundation model as a general-purpose world model that can be fine-tuned into customized world models for downstream applications. Our platform covers a video curation pipeline, pre-trained world foundation models, examples of post-training of pre-trained world foundation models, and video tokenizers. To help Physical AI builders solve the most critical problems of our society, we make Cosmos open-source and our models open-weight with permissive licenses available via https://github.com/nvidia-cosmos/cosmos-predict1.

Alisson Azzolini, Junjie Bai, Hannah Brandon et al. · nvidia

Physical AI systems need to perceive, understand, and perform complex actions in the physical world. In this paper, we present the Cosmos-Reason1 models that can understand the physical world and generate appropriate embodied decisions (e.g., next step action) in natural language through long chain-of-thought reasoning processes. We begin by defining key capabilities for Physical AI reasoning, with a focus on physical common sense and embodied reasoning. To represent physical common sense, we use a hierarchical ontology that captures fundamental knowledge about space, time, and physics. For embodied reasoning, we rely on a two-dimensional ontology that generalizes across different physical embodiments. Building on these capabilities, we develop two multimodal large language models, Cosmos-Reason1-7B and Cosmos-Reason1-56B. We curate data and train our models in two stages: Physical AI supervised fine-tuning (SFT) and Physical AI reinforcement learning (RL). To evaluate our models, we build comprehensive benchmarks for physical common sense and embodied reasoning according to our ontologies. Evaluation results show that Physical AI SFT and RL bring significant improvements. To facilitate the development of Physical AI, we make our code and pre-trained models available under the NVIDIA Open Model License at https://github.com/nvidia-cosmos/cosmos-reason1.

Yong Lin, Hangyu Lin, Wei Xiong et al.

LLMs acquire a wide range of abilities during pre-training, but aligning LLMs under Reinforcement Learning with Human Feedback (RLHF) can lead to forgetting pretrained abilities, which is also known as the alignment tax. To investigate alignment tax, we conducted experiments with existing RLHF algorithms using OpenLLaMA-3B, which revealed a pronounced alignment tax in NLP tasks. Whereas, despite various techniques to mitigate forgetting, they are often at odds with the RLHF performance, leading to a trade-off between alignment performance and forgetting mitigation, leading to an alignment-forgetting trade-off. In this paper we show that model averaging, which simply interpolates between pre and post RLHF model weights, surprisingly achieves the most strongest alignment-forgetting Pareto front among a wide range of competing methods. To understand its effectiveness, we offer theoretical insights into model averaging, revealing that it enhances performance Pareto front by increasing feature diversity on the layers where tasks share overlapped feature spaces. Empirical evidence corroborates our analysis by showing the benefits of averaging low-level transformer layers. Building on the analysis and the observation that averaging different layers of the transformer leads to significantly different alignment-forgetting trade-offs, we propose Heterogeneous Model Averaging (HMA) to Heterogeneously find various combination ratios of model layers. HMA seeks to maximize the alignment performance while incurring minimal alignment tax. Moreover, we validate HMA's performance across a range of RLHF algorithms over OpenLLaMA-3B and further extend our findings to Mistral-7B which is evaluated by open-sourced preference model and GPT4. Code available here: https://github.com/avalonstrel/Mitigating-the-Alignment-Tax-of-RLHF.git.

Arslan Ali, Junjie Bai, Maciej Bala et al. · nvidia

We introduce [Cosmos-Predict2.5], the latest generation of the Cosmos World Foundation Models for Physical AI. Built on a flow-based architecture, [Cosmos-Predict2.5] unifies Text2World, Image2World, and Video2World generation in a single model and leverages [Cosmos-Reason1], a Physical AI vision-language model, to provide richer text grounding and finer control of world simulation. Trained on 200M curated video clips and refined with reinforcement learning-based post-training, [Cosmos-Predict2.5] achieves substantial improvements over [Cosmos-Predict1] in video quality and instruction alignment, with models released at 2B and 14B scales. These capabilities enable more reliable synthetic data generation, policy evaluation, and closed-loop simulation for robotics and autonomous systems. We further extend the family with [Cosmos-Transfer2.5], a control-net style framework for Sim2Real and Real2Real world translation. Despite being 3.5$\times$ smaller than [Cosmos-Transfer1], it delivers higher fidelity and robust long-horizon video generation. Together, these advances establish [Cosmos-Predict2.5] and [Cosmos-Transfer2.5] as versatile tools for scaling embodied intelligence. To accelerate research and deployment in Physical AI, we release source code, pretrained checkpoints, and curated benchmarks under the NVIDIA Open Model License at https://github.com/nvidia-cosmos/cosmos-predict2.5 and https://github.com/nvidia-cosmos/cosmos-transfer2.5. We hope these open resources lower the barrier to adoption and foster innovation in building the next generation of embodied intelligence.

Aditi, Niket Agarwal, Arslan Ali et al.

We introduce Cosmos 3, a family of omnimodal world models designed to jointly process and generate language, image, video, audio, and action sequences within a unified mixture-of-transformers architecture. By supporting highly flexible input-output configurations, Cosmos 3 seamlessly unifies critical modalities for Physical AI -- effectively subsuming vision-language models, video generators, world simulators, and world-action models into a single framework. Our evaluation demonstrates that Cosmos 3 establishes a new state-of-the-art across a diverse suite of understanding and generation tasks, demonstrating omnimodal world models as scalable, general-purpose backbones for embodied agents. Our post-trained Cosmos 3 models were ranked as the best open-source Text-to-Image and Image-to-Video models by Artificial Analysis, and the best policy model by RoboArena at the time the technical report was written. To accelerate open research and deployment in Physical AI, we make our code, model checkpoints, curated synthetic datasets, and evaluation benchmark available under the Linux Foundation's OpenMDW-1.1 https://openmdw.ai/license/1-1/ License at https://github.com/nvidia/cosmos}{github.com/nvidia/cosmos and https://huggingface.co/collections/nvidia/cosmos3 . The project website is available at https://research.nvidia.com/labs/cosmos-lab/cosmos3 .

Haoxiang Wang, Pavan Kumar Anasosalu Vasu, Fartash Faghri et al. · utoronto

The landscape of publicly available vision foundation models (VFMs), such as CLIP and Segment Anything Model (SAM), is expanding rapidly. VFMs are endowed with distinct capabilities stemming from their pre-training objectives. For instance, CLIP excels in semantic understanding, while SAM specializes in spatial understanding for segmentation. In this work, we introduce a simple recipe to efficiently merge VFMs into a unified model that absorbs their expertise. Our method integrates techniques of multi-task learning, continual learning, and distillation. Further, it demands significantly less computational cost compared to traditional multi-task training from scratch, and it only needs a small fraction of the pre-training datasets that were initially used to train individual models. By applying our method to SAM and CLIP, we obtain SAM-CLIP: a unified model that combines the capabilities of SAM and CLIP into a single vision transformer. Compared with deploying SAM and CLIP independently, our merged model, SAM-CLIP, reduces storage and compute costs for inference, making it well-suited for edge device applications. We show that SAM-CLIP not only retains the foundational strengths of SAM and CLIP, but also introduces synergistic functionalities, notably in zero-shot semantic segmentation, where SAM-CLIP establishes new state-of-the-art results on 5 benchmarks. It outperforms previous models that are specifically designed for this task by a large margin, including +6.8% and +5.9% mean IoU improvement on Pascal-VOC and COCO-Stuff datasets, respectively.

Haoxiang Wang, Bo Li, Han Zhao

The vast majority of existing algorithms for unsupervised domain adaptation (UDA) focus on adapting from a labeled source domain to an unlabeled target domain directly in a one-off way. Gradual domain adaptation (GDA), on the other hand, assumes a path of $(T-1)$ unlabeled intermediate domains bridging the source and target, and aims to provide better generalization in the target domain by leveraging the intermediate ones. Under certain assumptions, Kumar et al. (2020) proposed a simple algorithm, Gradual Self-Training, along with a generalization bound in the order of $e^{O(T)} \left(\varepsilon_0+O\left(\sqrt{log(T)/n}\right)\right)$ for the target domain error, where $\varepsilon_0$ is the source domain error and $n$ is the data size of each domain. Due to the exponential factor, this upper bound becomes vacuous when $T$ is only moderately large. In this work, we analyze gradual self-training under more general and relaxed assumptions, and prove a significantly improved generalization bound as $\varepsilon_0+ O \left(TΔ+ T/\sqrt{n}\right) + \widetilde{O}\left(1/\sqrt{nT}\right)$, where $Δ$ is the average distributional distance between consecutive domains. Compared with the existing bound with an exponential dependency on $T$ as a multiplicative factor, our bound only depends on $T$ linearly and additively. Perhaps more interestingly, our result implies the existence of an optimal choice of $T$ that minimizes the generalization error, and it also naturally suggests an optimal way to construct the path of intermediate domains so as to minimize the accumulative path length $TΔ$ between the source and target. To corroborate the implications of our theory, we examine gradual self-training on multiple semi-synthetic and real datasets, which confirms our findings. We believe our insights provide a path forward toward the design of future GDA algorithms.

Haoxiang Wang, Yite Wang, Ruoyu Sun et al.

Model-agnostic meta-learning (MAML) and its variants have become popular approaches for few-shot learning. However, due to the non-convexity of deep neural nets (DNNs) and the bi-level formulation of MAML, the theoretical properties of MAML with DNNs remain largely unknown. In this paper, we first prove that MAML with over-parameterized DNNs is guaranteed to converge to global optima at a linear rate. Our convergence analysis indicates that MAML with over-parameterized DNNs is equivalent to kernel regression with a novel class of kernels, which we name as Meta Neural Tangent Kernels (MetaNTK). Then, we propose MetaNTK-NAS, a new training-free neural architecture search (NAS) method for few-shot learning that uses MetaNTK to rank and select architectures. Empirically, we compare our MetaNTK-NAS with previous NAS methods on two popular few-shot learning benchmarks, miniImageNet, and tieredImageNet. We show that the performance of MetaNTK-NAS is comparable or better than the state-of-the-art NAS method designed for few-shot learning while enjoying more than 100x speedup. We believe the efficiency of MetaNTK-NAS makes itself more practical for many real-world tasks.

Yifei He, Haoxiang Wang, Bo Li et al.

Unsupervised domain adaptation (UDA) adapts a model from a labeled source domain to an unlabeled target domain in a one-off way. Though widely applied, UDA faces a great challenge whenever the distribution shift between the source and the target is large. Gradual domain adaptation (GDA) mitigates this limitation by using intermediate domains to gradually adapt from the source to the target domain. In this work, we first theoretically analyze gradual self-training, a popular GDA algorithm, and provide a significantly improved generalization bound compared with Kumar et al. (2020). Our theoretical analysis leads to an interesting insight: to minimize the generalization error on the target domain, the sequence of intermediate domains should be placed uniformly along the Wasserstein geodesic between the source and target domains. The insight is particularly useful under the situation where intermediate domains are missing or scarce, which is often the case in real-world applications. Based on the insight, we propose $\textbf{G}$enerative Gradual D$\textbf{O}$main $\textbf{A}$daptation with Optimal $\textbf{T}$ransport (GOAT), an algorithmic framework that can generate intermediate domains in a data-dependent way. More concretely, we first generate intermediate domains along the Wasserstein geodesic between two given consecutive domains in a feature space, then apply gradual self-training to adapt the source-trained classifier to the target along the sequence of intermediate domains. Empirically, we demonstrate that our GOAT framework can improve the performance of standard GDA when the given intermediate domains are scarce, significantly broadening the real-world application scenarios of GDA. Our code is available at https://github.com/uiuctml/GOAT.

Mao Ye, Ruichen Jiang, Haoxiang Wang et al.

One of the key challenges of learning an online recommendation model is the temporal domain shift, which causes the mismatch between the training and testing data distribution and hence domain generalization error. To overcome, we propose to learn a meta future gradient generator that forecasts the gradient information of the future data distribution for training so that the recommendation model can be trained as if we were able to look ahead at the future of its deployment. Compared with Batch Update, a widely used paradigm, our theory suggests that the proposed algorithm achieves smaller temporal domain generalization error measured by a gradient variation term in a local regret. We demonstrate the empirical advantage by comparing with various representative baselines.

Yifei He, Haoxiang Wang, Ziyan Jiang et al. · amazon-science

Reward models (RM) capture the values and preferences of humans and play a central role in Reinforcement Learning with Human Feedback (RLHF) to align pretrained large language models (LLMs). Traditionally, training these models relies on extensive human-annotated preference data, which poses significant challenges in terms of scalability and cost. To overcome these limitations, we propose Semi-Supervised Reward Modeling (SSRM), an approach that enhances RM training using unlabeled data. Given an unlabeled dataset, SSRM involves three key iterative steps: pseudo-labeling unlabeled examples, selecting high-confidence examples through a confidence threshold, and supervised finetuning on the refined dataset. Across extensive experiments on various model configurations, we demonstrate that SSRM significantly improves reward models without incurring additional labeling costs. Notably, SSRM can achieve performance comparable to models trained entirely on labeled data of equivalent volumes. Overall, SSRM substantially reduces the dependency on large volumes of human-annotated data, thereby decreasing the overall cost and time involved in training effective reward models.

Haoxiang Wang, Zhanhong Jiang, Chao Liu et al.

In the context of distributed deep learning, the issue of stale weights or gradients could result in poor algorithmic performance. This issue is usually tackled by delay tolerant algorithms with some mild assumptions on the objective functions and step sizes. In this paper, we propose a different approach to develop a new algorithm, called $\textbf{P}$redicting $\textbf{C}$lipping $\textbf{A}$synchronous $\textbf{S}$tochastic $\textbf{G}$radient $\textbf{D}$escent (aka, PC-ASGD). Specifically, PC-ASGD has two steps - the $\textit{predicting step}$ leverages the gradient prediction using Taylor expansion to reduce the staleness of the outdated weights while the $\textit{clipping step}$ selectively drops the outdated weights to alleviate their negative effects. A tradeoff parameter is introduced to balance the effects between these two steps. Theoretically, we present the convergence rate considering the effects of delay of the proposed algorithm with constant step size when the smooth objective functions are weakly strongly-convex and nonconvex. One practical variant of PC-ASGD is also proposed by adopting a condition to help with the determination of the tradeoff parameter. For empirical validation, we demonstrate the performance of the algorithm with two deep neural network architectures on two benchmark datasets.

Haoxiang Wang, Maurice Weber, Josh Izaac et al.

Machine learning has emerged recently as a powerful tool for predicting properties of quantum many-body systems. For many ground states of gapped Hamiltonians, generative models can learn from measurements of a single quantum state to reconstruct the state accurately enough to predict local observables. Alternatively, classification and regression models can predict local observables by learning from measurements on different but related states. In this work, we combine the benefits of both approaches and propose the use of conditional generative models to simultaneously represent a family of states, learning shared structures of different quantum states from measurements. The trained model enables us to predict arbitrary local properties of ground states, even for states not included in the training data, without necessitating further training for new observables. We first numerically validate our approach on 2D random Heisenberg models using simulations of up to 45 qubits. Furthermore, we conduct quantum simulations on a neutral-atom quantum computer and demonstrate that our method can accurately predict the quantum phases of square lattices of 13$\times$13 Rydberg atoms.

Hanze Dong, Wei Xiong, Bo Pang et al.

We present the workflow of Online Iterative Reinforcement Learning from Human Feedback (RLHF) in this technical report, which is widely reported to outperform its offline counterpart by a large margin in the recent large language model (LLM) literature. However, existing open-source RLHF projects are still largely confined to the offline learning setting. In this technical report, we aim to fill in this gap and provide a detailed recipe that is easy to reproduce for online iterative RLHF. In particular, since online human feedback is usually infeasible for open-source communities with limited resources, we start by constructing preference models using a diverse set of open-source datasets and use the constructed proxy preference model to approximate human feedback. Then, we discuss the theoretical insights and algorithmic principles behind online iterative RLHF, followed by a detailed practical implementation. Our trained LLM achieves impressive performance on LLM chatbot benchmarks, including AlpacaEval-2, Arena-Hard, and MT-Bench, as well as other academic benchmarks such as HumanEval and TruthfulQA. We have shown that supervised fine-tuning (SFT) and iterative RLHF can obtain state-of-the-art performance with fully open-source datasets. Further, we have made our models, curated datasets, and comprehensive step-by-step code guidebooks publicly available. Please refer to https://github.com/RLHFlow/RLHF-Reward-Modeling and https://github.com/RLHFlow/Online-RLHF for more detailed information.

Yunheng Shen, Haoxiang Wang, Hairong Lv

Federated learning aims to learn a global model collaboratively while the training data belongs to different clients and is not allowed to be exchanged. However, the statistical heterogeneity challenge on non-IID data, such as class imbalance in classification, will cause client drift and significantly reduce the performance of the global model. This paper proposes a simple and effective approach named FedShift which adds the shift on the classifier output during the local training phase to alleviate the negative impact of class imbalance. We theoretically prove that the classifier shift in FedShift can make the local optimum consistent with the global optimum and ensure the convergence of the algorithm. Moreover, our experiments indicate that FedShift significantly outperforms the other state-of-the-art federated learning approaches on various datasets regarding accuracy and communication efficiency.

Haotian Ye, Qiyuan He, Jiaqi Han et al.

Accurate and efficient discrete video tokenization is essential for long video sequences processing. Yet, the inherent complexity and variable information density of videos present a significant bottleneck for current tokenizers, which rigidly compress all content at a fixed rate, leading to redundancy or information loss. Drawing inspiration from Shannon's information theory, this paper introduces InfoTok, a principled framework for adaptive video tokenization. We rigorously prove that existing data-agnostic training methods are suboptimal in representation length, and present a novel evidence lower bound (ELBO)-based algorithm that approaches theoretical optimality. Leveraging this framework, we develop a transformer-based adaptive compressor that enables adaptive tokenization. Empirical results demonstrate state-of-the-art compression performance, saving 20% tokens without influence on performance, and achieving 2.3x compression rates while still outperforming prior heuristic adaptive approaches. By allocating tokens according to informational richness, InfoTok enables a more compressed yet accurate tokenization for video representation, offering valuable insights for future research.

Haoxiang Wang, Gargi Balasubramaniam, Haozhe Si et al.

Domain generalization asks for models trained over a set of training environments to generalize well in unseen test environments. Recently, a series of algorithms such as Invariant Risk Minimization (IRM) have been proposed for domain generalization. However, Rosenfeld et al. (2021) shows that in a simple linear data model, even if non-convexity issues are ignored, IRM and its extensions cannot generalize to unseen environments with less than $d_s+1$ training environments, where $d_s$ is the dimension of the spurious-feature subspace. In this work, we propose Invariant-feature Subspace Recovery (ISR): a new class of algorithms to achieve provable domain generalization across the settings of classification and regression problems. First, in the binary classification setup of Rosenfeld et al. (2021), we show that our first algorithm, ISR-Mean, can identify the subspace spanned by invariant features from the first-order moments of the class-conditional distributions, and achieve provable domain generalization with $d_s+1$ training environments. Our second algorithm, ISR-Cov, further reduces the required number of training environments to $O(1)$ using the information of second-order moments. Notably, unlike IRM, our algorithms bypass non-convexity issues and enjoy global convergence guarantees. Next, we extend ISR-Mean to the more general setting of multi-class classification and propose ISR-Multiclass, which leverages class information and provably recovers the invariant-feature subspace with $\lceil d_s/k\rceil+1$ training environments for $k$-class classification. Finally, for regression problems, we propose ISR-Regression that can identify the invariant-feature subspace with $d_s+1$ training environments. Empirically, we demonstrate the superior performance of our ISRs on synthetic benchmarks. Further, ISR can be used as post-processing methods for feature extractors such as neural nets.

Mingze Sun, Haoxiang Wang, Wei Yao et al.

Recent studies have found that pain in infancy has a significant impact on infant development, including psychological problems, possible brain injury, and pain sensitivity in adulthood. However, due to the lack of specialists and the fact that infants are unable to express verbally their experience of pain, it is difficult to assess infant pain. Most existing infant pain assessment systems directly apply adult methods to infants ignoring the differences between infant expressions and adult expressions. Meanwhile, as the study of facial action coding system continues to advance, the use of action units (AUs) opens up new possibilities for expression recognition and pain assessment. In this paper, a novel AuE-IPA method is proposed for assessing infant pain by leveraging different engagement levels of AUs. First, different engagement levels of AUs in infant pain are revealed, by analyzing the class activation map of an end-to-end pain assessment model. The intensities of top-engaged AUs are then used in a regression model for achieving automatic infant pain assessment. The model proposed is trained and experimented on YouTube Immunization dataset, YouTube Blood Test dataset, and iCOPEVid dataset. The experimental results show that our AuE-IPA method is more applicable to infants and possesses stronger generalization ability than end-to-end assessment model and the classic PSPI metric.

Haoxiang Wang, Yong Lin, Wei Xiong et al.

Fine-grained control over large language models (LLMs) remains a significant challenge, hindering their adaptability to diverse user needs. While Reinforcement Learning from Human Feedback (RLHF) shows promise in aligning LLMs, its reliance on scalar rewards often limits its ability to capture diverse user preferences in real-world applications. To address this limitation, we introduce the Directional Preference Alignment (DPA) framework. Unlike the scalar-reward RLHF, DPA incorporates multi-objective reward modeling to represent diverse preference profiles. Additionally, DPA models user preferences as directions (i.e., unit vectors) in the reward space to achieve user-dependent preference control. Our method involves training a multi-objective reward model and then fine-tuning the LLM with a preference-conditioned variant of Rejection Sampling Finetuning (RSF), an RLHF method adopted by Llama 2. This method enjoys a better performance trade-off across various reward objectives. In comparison with the scalar-reward RLHF, DPA offers users intuitive control over LLM generation: they can arithmetically specify their desired trade-offs (e.g., more helpfulness with less verbosity). We also validate the effectiveness of DPA with real-world alignment experiments on Mistral-7B. Our method provides straightforward arithmetic control over the trade-off between helpfulness and verbosity while maintaining competitive performance with strong baselines such as Direct Preference Optimization (DPO).

Yao Wei, Haoxiang Wang, Mingze Sun et al.

Facial action unit (AU) recognition is essential to facial expression analysis. Since there are highly positive or negative correlations between AUs, some existing AU recognition works have focused on modeling AU relations. However, previous relationship-based approaches typically embed predefined rules into their models and ignore the impact of various AU relations in different crowds. In this paper, we propose a novel Attention Based Relation Network (ABRNet) for AU recognition, which can automatically capture AU relations without unnecessary or even disturbing predefined rules. ABRNet uses several relation learning layers to automatically capture different AU relations. The learned AU relation features are then fed into a self-attention fusion module, which aims to refine individual AU features with attention weights to enhance the feature robustness. Furthermore, we propose an AU relation dropout strategy and AU relation loss (AUR-Loss) to better model AU relations, which can further improve AU recognition. Extensive experiments show that our approach achieves state-of-the-art performance on the DISFA and DISFA+ datasets.

Haoxiang Wang, Da Yu, Huishuai Zhang

Evaluating large language models (LLMs) today rests on fixed benchmarks that apply the same set of items to any model, producing ceiling and floor effects that mask capability gaps. We argue that the most informative evaluation signal lies at the boundary, where the per-prompt pass probability is near $0.5$ under random-sampling decoding, and propose Dynamic Boundary Evaluation (DBE), which actively locates each model's boundary and places it on a globally comparable difficulty scale. DBE delivers three artifacts: (i) a calibrated item bank covering safety, capability, and truthfulness, with per-item difficulty labels validated across $9$ reference LLMs; (ii) Skill-Guided Boundary Search (SGBS), a search algorithm that finds boundary items for a given target LLM using only API-level query access; and (iii) an evaluation protocol that places a new LLM on a unified ability scale and grows the evaluation set adaptively when the target falls outside the bank's coverage. We instantiate DBE on four categories spanning safety (harmful request refusal and over-refusal), capability (constrained instruction following), and truthfulness (multi-turn sycophancy resistance). The resulting evaluation covers a broader model spectrum without saturation while remaining compatible with existing datasets.

Wei Li, Hanxiao Sun, Tao Huang et al.

Participating media are a pervasive and intriguing visual effect in virtual environments. Unfortunately, rendering such phenomena in real-time is notoriously difficult due to the computational expense of estimating the volume rendering equation. While the six-way lightmaps technique has been widely used in video games to render smoke with a camera-oriented billboard and approximate lighting effects using six precomputed lightmaps, achieving a balance between realism and efficiency, it is limited to pre-simulated animation sequences and is ignorant of camera movement. In this work, we propose a neural six-way lightmaps method to strike a long-sought balance between dynamics and visual realism. Our approach first generates a guiding map from the camera view using ray marching with a large sampling distance to approximate smoke scattering and silhouette. Then, given a guiding map, we train a neural network to predict the corresponding six-way lightmaps. The resulting lightmaps can be seamlessly used in existing game engine pipelines. This approach supports visually appealing rendering effects while enabling real-time user interactivity, including smoke-obstacle interaction, camera movement, and light change. By conducting a series of comprehensive benchmarks, we demonstrate that our method is well-suited for real-time applications, such as games and VR/AR.

Huayu Chen, Kaiwen Zheng, Qinsheng Zhang et al.

Reinforcement Learning (RL) has played a central role in the recent surge of LLMs' math abilities by enabling self-improvement through binary verifier signals. In contrast, Supervised Learning (SL) is rarely considered for such verification-driven training, largely due to its heavy reliance on reference answers and inability to reflect on mistakes. In this work, we challenge the prevailing notion that self-improvement is exclusive to RL and propose Negative-aware Fine-Tuning (NFT) -- a supervised approach that enables LLMs to reflect on their failures and improve autonomously with no external teachers. In online training, instead of throwing away self-generated negative answers, NFT constructs an implicit negative policy to model them. This implicit policy is parameterized with the same positive LLM we target to optimize on positive data, enabling direct policy optimization on all LLMs' generations. We conduct experiments on 7B and 32B models in math reasoning tasks. Results consistently show that through the additional leverage of negative feedback, NFT significantly improves over SL baselines like Rejection sampling Fine-Tuning, matching or even surpassing leading RL algorithms like GRPO and DAPO. Furthermore, we demonstrate that NFT and GRPO are actually equivalent in strict-on-policy training, even though they originate from entirely different theoretical foundations. Our experiments and theoretical findings bridge the gap between SL and RL methods in binary-feedback learning systems.

Kaiwen Zheng, Huayu Chen, Haotian Ye et al.

Online reinforcement learning (RL) has been central to post-training language models, but its extension to diffusion models remains challenging due to intractable likelihoods. Recent works discretize the reverse sampling process to enable GRPO-style training, yet they inherit fundamental drawbacks, including solver restrictions, forward-reverse inconsistency, and complicated integration with classifier-free guidance (CFG). We introduce Diffusion Negative-aware FineTuning (DiffusionNFT), a new online RL paradigm that optimizes diffusion models directly on the forward process via flow matching. DiffusionNFT contrasts positive and negative generations to define an implicit policy improvement direction, naturally incorporating reinforcement signals into the supervised learning objective. This formulation enables training with arbitrary black-box solvers, eliminates the need for likelihood estimation, and requires only clean images rather than sampling trajectories for policy optimization. DiffusionNFT is up to $25\times$ more efficient than FlowGRPO in head-to-head comparisons, while being CFG-free. For instance, DiffusionNFT improves the GenEval score from 0.24 to 0.98 within 1k steps, while FlowGRPO achieves 0.95 with over 5k steps and additional CFG employment. By leveraging multiple reward models, DiffusionNFT significantly boosts the performance of SD3.5-Medium in every benchmark tested.

Haoxiang Wang, Haozhe Si, Huajie Shao et al.

Real-world applications of machine learning models often confront data distribution shifts, wherein discrepancies exist between the training and test data distributions. In the common multi-domain multi-class setup, as the number of classes and domains scales up, it becomes infeasible to gather training data for every domain-class combination. This challenge naturally leads the quest for models with Compositional Generalization (CG) ability, where models can generalize to unseen domain-class combinations. To delve into the CG challenge, we develop CG-Bench, a suite of CG benchmarks derived from existing real-world image datasets, and observe that the prevalent pretraining-finetuning paradigm on foundational models, such as CLIP and DINOv2, struggles with the challenge. To address this challenge, we propose Compositional Feature Alignment (CFA), a simple two-stage finetuning technique that i) learns two orthogonal linear heads on a pretrained encoder with respect to class and domain labels, and ii) fine-tunes the encoder with the newly learned head frozen. We theoretically and empirically justify that CFA encourages compositional feature learning of pretrained models. We further conduct extensive experiments on CG-Bench for CLIP and DINOv2, two powerful pretrained vision foundation models. Experiment results show that CFA outperforms common finetuning techniques in compositional generalization, corroborating CFA's efficacy in compositional feature learning.

Haoxiang Wang, Zinan Lin, Da Yu et al.

Generating high fidelity, differentially private (DP) synthetic images offers a promising route to share and analyze sensitive visual data without compromising individual privacy. However, existing DP image synthesis methods struggle to produce high resolution outputs that faithfully capture the structure of the original data. In this paper, we introduce a novel method, referred to as Synthesis via Private Textual Intermediaries (SPTI), that can generate high resolution DP images with easy adoption. The key idea is to shift the challenge of DP image synthesis from the image domain to the text domain by leveraging state of the art DP text generation methods. SPTI first summarizes each private image into a concise textual description using image to text models, then applies a modified Private Evolution algorithm to generate DP text, and finally reconstructs images using text to image models. Notably, SPTI requires no model training, only inference with off the shelf models. Given a private dataset, SPTI produces synthetic images of substantially higher quality than prior DP approaches. On the LSUN Bedroom dataset, SPTI attains an FID equal to 26.71 under epsilon equal to 1.0, improving over Private Evolution FID of 40.36. Similarly, on MM CelebA HQ, SPTI achieves an FID equal to 33.27 at epsilon equal to 1.0, compared to 57.01 from DP fine tuning baselines. Overall, our results demonstrate that Synthesis via Private Textual Intermediaries provides a resource efficient and proprietary model compatible framework for generating high resolution DP synthetic images, greatly expanding access to private visual datasets.

Zhengxian Yang, Shengqi Wang, Shi Pan et al.

Fully immersive experiences that tightly integrate 6-DoF visual and auditory interaction are essential for virtual and augmented reality. While such experiences can be achieved through computer-generated content, constructing them directly from real-world captured videos remains largely unexplored. We introduce Immersive Volumetric Videos, a new volumetric media format designed to provide large 6-DoF interaction spaces, audiovisual feedback, and high-resolution, high-frame-rate dynamic content. To support IVV construction, we present ImViD, a multi-view, multi-modal dataset built upon a space-oriented capture philosophy. Our custom capture rig enables synchronized multi-view video-audio acquisition during motion, facilitating efficient capture of complex indoor and outdoor scenes with rich foreground--background interactions and challenging dynamics. The dataset provides 5K-resolution videos at 60 FPS with durations of 1-5 minutes, offering richer spatial, temporal, and multimodal coverage than existing benchmarks. Leveraging this dataset, we develop a dynamic light field reconstruction framework built upon a Gaussian-based spatio-temporal representation, incorporating flow-guided sparse initialization, joint camera temporal calibration, and multi-term spatio-temporal supervision for robust and accurate modeling of complex motion. We further propose, to our knowledge, the first method for sound field reconstruction from such multi-view audiovisual data. Together, these components form a unified pipeline for immersive volumetric video production. Extensive benchmarks and immersive VR experiments demonstrate that our pipeline generates high-quality, temporally stable audiovisual volumetric content with large 6-DoF interaction spaces. This work provides both a foundational definition and a practical construction methodology for immersive volumetric videos.

Xiaohang Yu, Zhengxian Yang, Shi Pan et al.

We have built a custom mobile multi-camera large-space dense light field capture system, which provides a series of high-quality and sufficiently dense light field images for various scenarios. Our aim is to contribute to the development of popular 3D scene reconstruction algorithms such as IBRnet, NeRF, and 3D Gaussian splitting. More importantly, the collected dataset, which is much denser than existing datasets, may also inspire space-oriented light field reconstruction, which is potentially different from object-centric 3D reconstruction, for immersive VR/AR experiences. We utilized a total of 40 GoPro 10 cameras, capturing images of 5k resolution. The number of photos captured for each scene is no less than 1000, and the average density (view number within a unit sphere) is 134.68. It is also worth noting that our system is capable of efficiently capturing large outdoor scenes. Addressing the current lack of large-space and dense light field datasets, we made efforts to include elements such as sky, reflections, lights and shadows that are of interest to researchers in the field of 3D reconstruction during the data capture process. Finally, we validated the effectiveness of our provided dataset on three popular algorithms and also integrated the reconstructed 3DGS results into the Unity engine, demonstrating the potential of utilizing our datasets to enhance the realism of virtual reality (VR) and create feasible interactive spaces. The dataset is available at our project website.

Hao Yang, Haoxuan Li, Luyu Chen et al.

Hidden confounding remains a central challenge in estimating treatment effects from observational data, as unobserved variables can lead to biased causal estimates. While recent work has explored the use of large language models (LLMs) for causal inference, most approaches still rely on the unconfoundedness assumption. In this paper, we make the first attempt to mitigate hidden confounding using LLMs. We propose ProCI (Progressive Confounder Imputation), a framework that elicits the semantic and world knowledge of LLMs to iteratively generate, impute, and validate hidden confounders. ProCI leverages two key capabilities of LLMs: their strong semantic reasoning ability, which enables the discovery of plausible confounders from both structured and unstructured inputs, and their embedded world knowledge, which supports counterfactual reasoning under latent confounding. To improve robustness, ProCI adopts a distributional reasoning strategy instead of direct value imputation to prevent the collapsed outputs. Extensive experiments demonstrate that ProCI uncovers meaningful confounders and significantly improves treatment effect estimation across various datasets and LLMs.

Zhengxian Yang, Shi Pan, Shengqi Wang et al.

User engagement is greatly enhanced by fully immersive multi-modal experiences that combine visual and auditory stimuli. Consequently, the next frontier in VR/AR technologies lies in immersive volumetric videos with complete scene capture, large 6-DoF interaction space, multi-modal feedback, and high resolution & frame-rate contents. To stimulate the reconstruction of immersive volumetric videos, we introduce ImViD, a multi-view, multi-modal dataset featuring complete space-oriented data capture and various indoor/outdoor scenarios. Our capture rig supports multi-view video-audio capture while on the move, a capability absent in existing datasets, significantly enhancing the completeness, flexibility, and efficiency of data capture. The captured multi-view videos (with synchronized audios) are in 5K resolution at 60FPS, lasting from 1-5 minutes, and include rich foreground-background elements, and complex dynamics. We benchmark existing methods using our dataset and establish a base pipeline for constructing immersive volumetric videos from multi-view audiovisual inputs for 6-DoF multi-modal immersive VR experiences. The benchmark and the reconstruction and interaction results demonstrate the effectiveness of our dataset and baseline method, which we believe will stimulate future research on immersive volumetric video production.

Haoxiang Wang, Wei Xiong, Tengyang Xie et al.

Reinforcement learning from human feedback (RLHF) has emerged as the primary method for aligning large language models (LLMs) with human preferences. The RLHF process typically starts by training a reward model (RM) using human preference data. Conventional RMs are trained on pairwise responses to the same user request, with relative ratings indicating which response humans prefer. The trained RM serves as a proxy for human preferences. However, due to the black-box nature of RMs, their outputs lack interpretability, as humans cannot intuitively understand why an RM thinks a response is good or not. As RMs act as human preference proxies, we believe they should be human-interpretable to ensure that their internal decision processes are consistent with human preferences and to prevent reward hacking in LLM alignment. To build RMs with interpretable preferences, we propose a two-stage approach: i) train an Absolute-Rating Multi-Objective Reward Model (ArmoRM) with multi-dimensional absolute-rating data, each dimension corresponding to a human-interpretable objective (e.g., honesty, verbosity, safety); ii) employ a Mixture-of-Experts (MoE) strategy with a gating network that automatically selects the most suitable reward objectives based on the context. We efficiently trained an ArmoRM with Llama-3 8B and a gating network consisting of a shallow MLP on top of the ArmoRM. Our trained model, ArmoRM-Llama3-8B, obtains state-of-the-art performance on RewardBench, a benchmark evaluating RMs for language modeling. Notably, the performance of our model surpasses the LLM-as-a-judge method with GPT-4 judges by a margin, and approaches the performance of the much larger Nemotron-4 340B reward model.

Xiaohang Yu, Haoxiang Wang, Yuqi Han et al.

This paper proposes a hybrid radiance field representation for unbounded immersive light field reconstruction which supports high-quality rendering and aggressive view extrapolation. The key idea is to first formally separate the foreground and the background and then adaptively balance learning of them during the training process. To fulfill this goal, we represent the foreground and background as two separate radiance fields with two different spatial mapping strategies. We further propose an adaptive sampling strategy and a segmentation regularizer for more clear segmentation and robust convergence. Finally, we contribute a novel immersive light field dataset, named THUImmersive, with the potential to achieve much larger space 6DoF immersive rendering effects compared with existing datasets, by capturing multiple neighboring viewpoints for the same scene, to stimulate the research and AR/VR applications in the immersive light field domain. Extensive experiments demonstrate the strong performance of our method for unbounded immersive light field reconstruction.

Haoxiang Wang, Haozhe Si, Bo Li et al.

Domain generalization asks for models trained over a set of training environments to perform well in unseen test environments. Recently, a series of algorithms such as Invariant Risk Minimization (IRM) has been proposed for domain generalization. However, Rosenfeld et al. (2021) shows that in a simple linear data model, even if non-convexity issues are ignored, IRM and its extensions cannot generalize to unseen environments with less than $d_s+1$ training environments, where $d_s$ is the dimension of the spurious-feature subspace. In this paper, we propose to achieve domain generalization with Invariant-feature Subspace Recovery (ISR). Our first algorithm, ISR-Mean, can identify the subspace spanned by invariant features from the first-order moments of the class-conditional distributions, and achieve provable domain generalization with $d_s+1$ training environments under the data model of Rosenfeld et al. (2021). Our second algorithm, ISR-Cov, further reduces the required number of training environments to $O(1)$ using the information of second-order moments. Notably, unlike IRM, our algorithms bypass non-convexity issues and enjoy global convergence guarantees. Empirically, our ISRs can obtain superior performance compared with IRM on synthetic benchmarks. In addition, on three real-world image and text datasets, we show that both ISRs can be used as simple yet effective post-processing methods to improve the worst-case accuracy of (pre-)trained models against spurious correlations and group shifts.

Haoxiang Wang, Han Zhao, Bo Li

Multi-task learning (MTL) aims to improve the generalization of several related tasks by learning them jointly. As a comparison, in addition to the joint training scheme, modern meta-learning allows unseen tasks with limited labels during the test phase, in the hope of fast adaptation over them. Despite the subtle difference between MTL and meta-learning in the problem formulation, both learning paradigms share the same insight that the shared structure between existing training tasks could lead to better generalization and adaptation. In this paper, we take one important step further to understand the close connection between these two learning paradigms, through both theoretical analysis and empirical investigation. Theoretically, we first demonstrate that MTL shares the same optimization formulation with a class of gradient-based meta-learning (GBML) algorithms. We then prove that for over-parameterized neural networks with sufficient depth, the learned predictive functions of MTL and GBML are close. In particular, this result implies that the predictions given by these two models are similar over the same unseen task. Empirically, we corroborate our theoretical findings by showing that, with proper implementation, MTL is competitive against state-of-the-art GBML algorithms on a set of few-shot image classification benchmarks. Since existing GBML algorithms often involve costly second-order bi-level optimization, our first-order MTL method is an order of magnitude faster on large-scale datasets such as mini-ImageNet. We believe this work could help bridge the gap between these two learning paradigms, and provide a computationally efficient alternative to GBML that also supports fast task adaptation.

Haoxiang Wang, Jiasheng Zhang, Chenbei Lu et al.

Smart meter devices enable a better understanding of the demand at the potential risk of private information leakage. One promising solution to mitigating such risk is to inject noises into the meter data to achieve a certain level of differential privacy. In this paper, we cast one-shot non-intrusive load monitoring (NILM) in the compressive sensing framework, and bridge the gap between theoretical accuracy of NILM inference and differential privacy's parameters. We then derive the valid theoretical bounds to offer insights on how the differential privacy parameters affect the NILM performance. Moreover, we generalize our conclusions by proposing the hierarchical framework to solve the multi-shot NILM problem. Numerical experiments verify our analytical results and offer better physical insights of differential privacy in various practical scenarios. This also demonstrates the significance of our work for the general privacy preserving mechanism design.

Haoxiang Wang, Ruoyu Sun, Bo Li

Gradient-based meta-learning (GBML) with deep neural nets (DNNs) has become a popular approach for few-shot learning. However, due to the non-convexity of DNNs and the bi-level optimization in GBML, the theoretical properties of GBML with DNNs remain largely unknown. In this paper, we first aim to answer the following question: Does GBML with DNNs have global convergence guarantees? We provide a positive answer to this question by proving that GBML with over-parameterized DNNs is guaranteed to converge to global optima at a linear rate. The second question we aim to address is: How does GBML achieve fast adaption to new tasks with prior experience on past tasks? To answer it, we theoretically show that GBML is equivalent to a functional gradient descent operation that explicitly propagates experience from the past tasks to new ones, and then we prove a generalization error bound of GBML with over-parameterized DNNs.

Jingshi Cui, Haoxiang Wang, Chenye Wu et al.

To enable an efficient electricity market, a good pricing scheme is of vital importance. Among many practical schemes, customized pricing is commonly believed to be able to best exploit the flexibility in the demand side. However, due to the large volume of consumers in the electricity sector, such task is simply too overwhelming. In this paper, we first compare two data driven schemes: one based on load profile and the other based on user's marginal system cost. Vulnerability analysis shows that the former approach may lead to loopholes in the electricity market while the latter one is able to guarantee the robustness, which yields our robust data-driven pricing scheme. Although k-means clustering is in general NP-hard, surprisingly, by exploiting the structure of our problem, we design an efficient yet optimal k-means clustering algorithm to implement our proposed scheme.

Jingshi Cui, Haoxiang Wang, Chenye Wu et al.

Data analytics and machine learning techniques are being rapidly adopted into the power system, including power system control as well as electricity market design. In this paper, from an adversarial machine learning point of view, we examine the vulnerability of data-driven electricity market design. More precisely, we follow the idea that consumer's load profile should uniquely determine its electricity rate, which yields a clustering oriented pricing scheme. We first identify the strategic behaviors of malicious users by defining a notion of disguising. Based on this notion, we characterize the sensitivity zones to evaluate the percentage of malicious users in each cluster. Based on a thorough cost benefit analysis, we conclude with the vulnerability analysis.

Rongcui Dong, Haoxiang Wang

This thesis describes a study to perform change detection on Very High Resolution satellite images using image fusion based on 2D Discrete Wavelet Transform and Fuzzy C-Means clustering algorithm. Multiple other methods are also quantitatively and qualitatively compared in this study.

Zeling Wu, Haoxiang Wang

In this article, we propose a super-resolution method to resolve the problem of image low spatial because of the limitation of imaging devices. We make use of the strong non-linearity mapped ability of the back-propagation neural networks(BPNN). Training sample images are got by undersampled method. The elements chose as the inputs of the BPNN are pixels referred to Non-local means(NL-Means). Making use of the self-similarity of the images, those inputs are the pixels which are pixels gained from modified NL-means which is specific for super-resolution. Besides, small change on core function of NL-means has been applied in the method we use in this article so that we can have a clearer edge in the shrunk image. Experimental results gained from the Peak Signal to Noise Ratio(PSNR) and the Equivalent Number of Look(ENL), indicate that adding the similar pixels as inputs will increase the results than not taking them into consideration.

Ru-Ze Liang, Lihui Shi, Haoxiang Wang et al.

In this paper we study the problem of content-based image retrieval. In this problem, the most popular performance measure is the top precision measure, and the most important component of a retrieval system is the similarity function used to compare a query image against a database image. However, up to now, there is no existing similarity learning method proposed to optimize the top precision measure. To fill this gap, in this paper, we propose a novel similarity learning method to maximize the top precision measure. We model this problem as a minimization problem with an objective function as the combination of the losses of the relevant images ranked behind the top-ranked irrelevant image, and the squared Frobenius norm of the similarity function parameter. This minimization problem is solved as a quadratic programming problem. The experiments over two benchmark data sets show the advantages of the proposed method over other similarity learning methods when the top precision is used as the performance measure.

Jingbin Wang, Haoxiang Wang, Yihua Zhou et al.

In this paper, we propose a multi-kernel classifier learning algorithm to optimize a given nonlinear and nonsmoonth multivariate classifier performance measure. Moreover, to solve the problem of kernel function selection and kernel parameter tuning, we proposed to construct an optimal kernel by weighted linear combination of some candidate kernels. The learning of the classifier parameter and the kernel weight are unified in a single objective function considering to minimize the upper boundary of the given multivariate performance measure. The objective function is optimized with regard to classifier parameter and kernel weight alternately in an iterative algorithm by using cutting plane algorithm. The developed algorithm is evaluated on two different pattern classification methods with regard to various multivariate performance measure optimization problems. The experiment results show the proposed algorithm outperforms the competing methods.

Jingyan Wang, Yihua Zhou, Haoxiang Wang et al.

The problem of tag completion is to learn the missing tags of an image. In this paper, we propose to learn a tag scoring vector for each image by local linear learning. A local linear function is used in the neighborhood of each image to predict the tag scoring vectors of its neighboring images. We construct a unified objective function for the learning of both tag scoring vectors and local linear function parame- ters. In the objective, we impose the learned tag scoring vectors to be consistent with the known associations to the tags of each image, and also minimize the prediction error of each local linear function, while reducing the complexity of each local function. The objective function is optimized by an alternate optimization strategy and gradient descent methods in an iterative algorithm. We compare the proposed algorithm against different state-of-the-art tag completion methods, and the results show its advantages.