Enhancing Online Reinforcement Learning with Meta-Learned Objective from Offline Data

A major challenge in Reinforcement Learning (RL) is the difficulty of learning an optimal policy from sparse rewards. Prior works enhance online RL with conventional Imitation Learning (IL) via a handcrafted auxiliary objective, at the cost of restricting the RL policy to be sub-optimal when the offline data is generated by a non-expert policy. Instead, to better leverage valuable information in offline data, we develop Generalized Imitation Learning from Demonstration (GILD), which meta-learns an objective that distills knowledge from offline data and instills intrinsic motivation towards the optimal policy. Distinct from prior works that are exclusive to a specific RL algorithm, GILD is a flexible module intended for diverse vanilla off-policy RL algorithms. In addition, GILD introduces no domain-specific hyperparameter and minimal increase in computational cost. In four challenging MuJoCo tasks with sparse rewards, we show that three RL algorithms enhanced with GILD significantly outperform state-of-the-art methods.

Towards Pattern-aware Data Augmentation for Temporal Knowledge Graph Completion

Predicting missing facts for temporal knowledge graphs (TKGs) is a fundamental task, called temporal knowledge graph completion (TKGC). One key challenge in this task is the imbalance in data distribution, where facts are unevenly spread across entities and timestamps. This imbalance can lead to poor completion performance or long-tail entities and timestamps, and unstable training due to the introduction of false negative samples. Unfortunately, few previous studies have investigated how to mitigate these effects. Moreover, for the first time, we found that existing methods suffer from model preferences, revealing that entities with specific properties (e.g., recently active) are favored by different models. Such preferences will lead to error accumulation and further exacerbate the effects of imbalanced data distribution, but are overlooked by previous studies. To alleviate the impacts of imbalanced data and model preferences, we introduce Booster, the first data augmentation strategy for TKGs. The unique requirements here lie in generating new samples that fit the complex semantic and temporal patterns within TKGs, and identifying hard-learning samples specific to models. Therefore, we propose a hierarchical scoring algorithm based on triadic closures within TKGs. By incorporating both global semantic patterns and local time-aware structures, the algorithm enables pattern-aware validation for new samples. Meanwhile, we propose a two-stage training approach to identify samples that deviate from the model's preferred patterns. With a well-designed frequency-based filtering strategy, this approach also helps to avoid the misleading of false negatives. Experiments justify that Booster can seamlessly adapt to existing TKGC models and achieve up to an 8.7% performance improvement.

Imitation Learning from Suboptimal Demonstrations via Meta-Learning An Action Ranker

A major bottleneck in imitation learning is the requirement of a large number of expert demonstrations, which can be expensive or inaccessible. Learning from supplementary demonstrations without strict quality requirements has emerged as a powerful paradigm to address this challenge. However, previous methods often fail to fully utilize their potential by discarding non-expert data. Our key insight is that even demonstrations that fall outside the expert distribution but outperform the learned policy can enhance policy performance. To utilize this potential, we propose a novel approach named imitation learning via meta-learning an action ranker (ILMAR). ILMAR implements weighted behavior cloning (weighted BC) on a limited set of expert demonstrations along with supplementary demonstrations. It utilizes the functional of the advantage function to selectively integrate knowledge from the supplementary demonstrations. To make more effective use of supplementary demonstrations, we introduce meta-goal in ILMAR to optimize the functional of the advantage function by explicitly minimizing the distance between the current policy and the expert policy. Comprehensive experiments using extensive tasks demonstrate that ILMAR significantly outperforms previous methods in handling suboptimal demonstrations. Code is available at https://github.com/F-GOD6/ILMAR.

SynerGen-VL: Towards Synergistic Image Understanding and Generation with Vision Experts and Token Folding

The remarkable success of Large Language Models (LLMs) has extended to the multimodal domain, achieving outstanding performance in image understanding and generation. Recent efforts to develop unified Multimodal Large Language Models (MLLMs) that integrate these capabilities have shown promising results. However, existing approaches often involve complex designs in model architecture or training pipeline, increasing the difficulty of model training and scaling. In this paper, we propose SynerGen-VL, a simple yet powerful encoder-free MLLM capable of both image understanding and generation. To address challenges identified in existing encoder-free unified MLLMs, we introduce the token folding mechanism and the vision-expert-based progressive alignment pretraining strategy, which effectively support high-resolution image understanding while reducing training complexity. After being trained on large-scale mixed image-text data with a unified next-token prediction objective, SynerGen-VL achieves or surpasses the performance of existing encoder-free unified MLLMs with comparable or smaller parameter sizes, and narrows the gap with task-specific state-of-the-art models, highlighting a promising path toward future unified MLLMs. Our code and models shall be released.

CreatiLayout: Siamese Multimodal Diffusion Transformer for Creative Layout-to-Image Generation

Diffusion models have been recognized for their ability to generate images that are not only visually appealing but also of high artistic quality. As a result, Layout-to-Image (L2I) generation has been proposed to leverage region-specific positions and descriptions to enable more precise and controllable generation. However, previous methods primarily focus on UNet-based models (e.g., SD1.5 and SDXL), and limited effort has explored Multimodal Diffusion Transformers (MM-DiTs), which have demonstrated powerful image generation capabilities. Enabling MM-DiT for layout-to-image generation seems straightforward but is challenging due to the complexity of how layout is introduced, integrated, and balanced among multiple modalities. To this end, we explore various network variants to efficiently incorporate layout guidance into MM-DiT, and ultimately present SiamLayout. To Inherit the advantages of MM-DiT, we use a separate set of network weights to process the layout, treating it as equally important as the image and text modalities. Meanwhile, to alleviate the competition among modalities, we decouple the image-layout interaction into a siamese branch alongside the image-text one and fuse them in the later stage. Moreover, we contribute a large-scale layout dataset, named LayoutSAM, which includes 2.7 million image-text pairs and 10.7 million entities. Each entity is annotated with a bounding box and a detailed description. We further construct the LayoutSAM-Eval benchmark as a comprehensive tool for evaluating the L2I generation quality. Finally, we introduce the Layout Designer, which taps into the potential of large language models in layout planning, transforming them into experts in layout generation and optimization. Our code, model, and dataset will be available at https://creatilayout.github.io.

Quantization without Tears

Deep neural networks, while achieving remarkable success across diverse tasks, demand significant resources, including computation, GPU memory, bandwidth, storage, and energy. Network quantization, as a standard compression and acceleration technique, reduces storage costs and enables potential inference acceleration by discretizing network weights and activations into a finite set of integer values. However, current quantization methods are often complex and sensitive, requiring extensive task-specific hyperparameters, where even a single misconfiguration can impair model performance, limiting generality across different models and tasks. In this paper, we propose Quantization without Tears (QwT), a method that simultaneously achieves quantization speed, accuracy, simplicity, and generality. The key insight of QwT is to incorporate a lightweight additional structure into the quantized network to mitigate information loss during quantization. This structure consists solely of a small set of linear layers, keeping the method simple and efficient. More importantly, it provides a closed-form solution, allowing us to improve accuracy effortlessly under 2 minutes. Extensive experiments across various vision, language, and multimodal tasks demonstrate that QwT is both highly effective and versatile. In fact, our approach offers a robust solution for network quantization that combines simplicity, accuracy, and adaptability, which provides new insights for the design of novel quantization paradigms.

Diffusion Product Quantization

In this work, we explore the quantization of diffusion models in extreme compression regimes to reduce model size while maintaining performance. We begin by investigating classical vector quantization but find that diffusion models are particularly susceptible to quantization error, with the codebook size limiting generation quality. To address this, we introduce product quantization, which offers improved reconstruction precision and larger capacity -- crucial for preserving the generative capabilities of diffusion models. Furthermore, we propose a method to compress the codebook by evaluating the importance of each vector and removing redundancy, ensuring the model size remaining within the desired range. We also introduce an end-to-end calibration approach that adjusts assignments during the forward pass and optimizes the codebook using the DDPM loss. By compressing the model to as low as 1 bit (resulting in over 24 times reduction in model size), we achieve a balance between compression and quality. We apply our compression method to the DiT model on ImageNet and consistently outperform other quantization approaches, demonstrating competitive generative performance.

Online Detection of Anomalies in Temporal Knowledge Graphs with Interpretability

Temporal knowledge graphs (TKGs) are valuable resources for capturing evolving relationships among entities, yet they are often plagued by noise, necessitating robust anomaly detection mechanisms. Existing dynamic graph anomaly detection approaches struggle to capture the rich semantics introduced by node and edge categories within TKGs, while TKG embedding methods lack interpretability, undermining the credibility of anomaly detection. Moreover, these methods falter in adapting to pattern changes and semantic drifts resulting from knowledge updates. To tackle these challenges, we introduce AnoT, an efficient TKG summarization method tailored for interpretable online anomaly detection in TKGs. AnoT begins by summarizing a TKG into a novel rule graph, enabling flexible inference of complex patterns in TKGs. When new knowledge emerges, AnoT maps it onto a node in the rule graph and traverses the rule graph recursively to derive the anomaly score of the knowledge. The traversal yields reachable nodes that furnish interpretable evidence for the validity or the anomalous of the new knowledge. Overall, AnoT embodies a detector-updater-monitor architecture, encompassing a detector for offline TKG summarization and online scoring, an updater for real-time rule graph updates based on emerging knowledge, and a monitor for estimating the approximation error of the rule graph. Experimental results on four real-world datasets demonstrate that AnoT surpasses existing methods significantly in terms of accuracy and interoperability. All of the raw datasets and the implementation of AnoT are provided in https://github.com/zjs123/ANoT.

Test-time adaptation for geospatial point cloud semantic segmentation with distinct domain shifts

Domain adaptation (DA) techniques help deep learning models generalize across data shifts for point cloud semantic segmentation (PCSS). Test-time adaptation (TTA) allows direct adaptation of a pre-trained model to unlabeled data during inference stage without access to source data or additional training, avoiding privacy issues and large computational resources. We address TTA for geospatial PCSS by introducing three domain shift paradigms: photogrammetric to airborne LiDAR, airborne to mobile LiDAR, and synthetic to mobile laser scanning. We propose a TTA method that progressively updates batch normalization (BN) statistics with each testing batch. Additionally, a self-supervised learning module optimizes learnable BN affine parameters. Information maximization and reliability-constrained pseudo-labeling improve prediction confidence and supply supervisory signals. Experimental results show our method improves classification accuracy by up to 20\% mIoU, outperforming other methods. For photogrammetric (SensatUrban) to airborne (Hessigheim 3D) adaptation at the inference stage, our method achieves 59.46\% mIoU and 85.97\% OA without retraining or fine-turning.

Style Alignment based Dynamic Observation Method for UAV-View Geo-localization

The task of UAV-view geo-localization is to estimate the localization of a query satellite/drone image by matching it against a reference dataset consisting of drone/satellite images. Though tremendous strides have been made in feature alignment between satellite and drone views, vast differences in both inter and intra-class due to changes in viewpoint, altitude, and lighting remain a huge challenge. In this paper, a style alignment based dynamic observation method for UAV-view geo-localization is proposed to meet the above challenges from two perspectives: visual style transformation and surrounding noise control. Specifically, we introduce a style alignment strategy to transfrom the diverse visual style of drone-view images into a unified satellite images visual style. Then a dynamic observation module is designed to evaluate the spatial distribution of images by mimicking human observation habits. It is featured by the hierarchical attention block (HAB) with a dual-square-ring stream structure, to reduce surrounding noise and geographical deformation. In addition, we propose a deconstruction loss to push away features of different geo-tags and squeeze knowledge from unmatched images by correlation calculation. The experimental results demonstrate the state-of-the-art performance of our model on benchmarked datasets. In particular, when compared to the prior art on University-1652, our results surpass the best of them (FSRA), while only requiring 2x fewer parameters. Code will be released at https://github.com/Xcco1/SA\_DOM