Abstract:Sequential recommendation systems leveraging transformer architectures have demonstrated exceptional capabilities in capturing user behavior patterns. At the core of these systems lies the critical challenge of constructing effective item representations. Traditional approaches employ feature fusion through simple concatenation or basic neural architectures to create uniform representation sequences. However, these conventional methods fail to address the intrinsic diversity of item attributes, thereby constraining the transformer's capacity to discern fine-grained patterns and hindering model extensibility. Although recent research has begun incorporating user-related heterogeneous features into item sequences, the equally crucial item-side heterogeneous feature continue to be neglected. To bridge this methodological gap, we present HeterRec - an innovative framework featuring two novel components: the Heterogeneous Token Flattening Layer (HTFL) and Hierarchical Causal Transformer (HCT). HTFL pioneers a sophisticated tokenization mechanism that decomposes items into multi-dimensional token sets and structures them into heterogeneous sequences, enabling scalable performance enhancement through model expansion. The HCT architecture further enhances pattern discovery through token-level and item-level attention mechanisms. furthermore, we develop a Listwise Multi-step Prediction (LMP) objective function to optimize learning process. Rigorous validation, including real-world industrial platforms, confirms HeterRec's state-of-the-art performance in both effective and efficiency.
Abstract:Online continual learning (OCL), which enables AI systems to adaptively learn from non-stationary data streams, is commonly achieved using experience replay (ER)-based methods that retain knowledge by replaying stored past during training. However, these methods face challenges of prediction bias, stemming from deviations in parameter update directions during task transitions. This paper identifies parameter variation imbalance as a critical factor contributing to prediction bias in ER-based OCL. Specifically, using the proposed parameter variation evaluation method, we highlight two types of imbalance: correlation-induced imbalance, where certain parameters are disproportionately updated across tasks, and layer-wise imbalance, where output layer parameters update faster than those in preceding layers. To mitigate the above imbalances, we propose the Parameter Variation Balancing Framework (PVBF), which incorporates: 1) a novel method to compute parameter correlations with previous tasks based on parameter variations, 2) an encourage-and-consolidate (E&C) method utilizing parameter correlations to perform gradient adjustments across all parameters during training, 3) a dual-layer copy weights with reinit (D-CWR) strategy to slowly update output layer parameters for frequently occuring sample categories. Experiments on short and long task sequences demonstrate that PVBF significantly reduces prediction bias and improves OCL performance, achieving up to 47\% higher accuracy compared to existing ER-based methods.
Abstract:Currently, there are few effective methods for synthesizing a mass of high-resolution rainy images in complex illumination conditions. However, these methods are essential for synthesizing large-scale high-quality paired rainy-clean image datasets, which can train deep learning-based single image rain removal models capable of generalizing to various illumination conditions. Therefore, we propose a practical two-stage learning-from-rendering pipeline for high-resolution rainy image synthesis. The pipeline combines the benefits of the realism of rendering-based methods and the high-efficiency of learning-based methods, providing the possibility of creating large-scale high-quality paired rainy-clean image datasets. In the rendering stage, we use a rendering-based method to create a High-resolution Rainy Image (HRI) dataset, which contains realistic high-resolution paired rainy-clean images of multiple scenes and various illumination conditions. In the learning stage, to learn illumination information from background images for high-resolution rainy image generation, we propose a High-resolution Rainy Image Generation Network (HRIGNet). HRIGNet is designed to introduce a guiding diffusion model in the Latent Diffusion Model, which provides additional guidance information for high-resolution image synthesis. In our experiments, HRIGNet is able to synthesize high-resolution rainy images up to 2048x1024 resolution. Rain removal experiments on real dataset validate that our method can help improve the robustness of deep derainers to real rainy images. To make our work reproducible, source codes and the dataset have been released at https://kb824999404.github.io/HRIG/.
Abstract:Industrial recommendation systems typically involve a two-stage process: retrieval and ranking, which aims to match users with millions of items. In the retrieval stage, classic embedding-based retrieval (EBR) methods depend on effective negative sampling techniques to enhance both performance and efficiency. However, existing techniques often suffer from false negatives, high cost for ensuring sampling quality and semantic information deficiency. To address these limitations, we propose Effective and Semantic-Aware Negative Sampling (ESANS), which integrates two key components: Effective Dense Interpolation Strategy (EDIS) and Multimodal Semantic-Aware Clustering (MSAC). EDIS generates virtual samples within the low-dimensional embedding space to improve the diversity and density of the sampling distribution while minimizing computational costs. MSAC refines the negative sampling distribution by hierarchically clustering item representations based on multimodal information (visual, textual, behavioral), ensuring semantic consistency and reducing false negatives. Extensive offline and online experiments demonstrate the superior efficiency and performance of ESANS.
Abstract:Recent advancements in text-to-video (T2V) generation have leveraged diffusion models to enhance the visual coherence of videos generated from textual descriptions. However, most research has primarily focused on object motion, with limited attention given to cinematic language in videos, which is crucial for cinematographers to convey emotion and narrative pacing. To address this limitation, we propose a threefold approach to enhance the ability of T2V models to generate controllable cinematic language. Specifically, we introduce a cinematic language dataset that encompasses shot framing, angle, and camera movement, enabling models to learn diverse cinematic styles. Building on this, to facilitate robust cinematic alignment evaluation, we present CameraCLIP, a model fine-tuned on the proposed dataset that excels in understanding complex cinematic language in generated videos and can further provide valuable guidance in the multi-shot composition process. Finally, we propose CLIPLoRA, a cost-guided dynamic LoRA composition method that facilitates smooth transitions and realistic blending of cinematic language by dynamically fusing multiple pre-trained cinematic LoRAs within a single video. Our experiments demonstrate that CameraCLIP outperforms existing models in assessing the alignment between cinematic language and video, achieving an R@1 score of 0.81. Additionally, CLIPLoRA improves the ability for multi-shot composition, potentially bridging the gap between automatically generated videos and those shot by professional cinematographers.
Abstract:Micro-expressions (MEs) are subtle facial movements that occur spontaneously when people try to conceal the real emotions. Micro-expression recognition (MER) is crucial in many fields, including criminal analysis and psychotherapy. However, MER is challenging since MEs have low intensity and ME datasets are small in size. To this end, a three-stream temporal-shift attention network based on self-knowledge distillation (SKD-TSTSAN) is proposed in this paper. Firstly, to address the low intensity of ME muscle movements, we utilize learning-based motion magnification modules to enhance the intensity of ME muscle movements. Secondly, we employ efficient channel attention (ECA) modules in the local-spatial stream to make the network focus on facial regions that are highly relevant to MEs. In addition, temporal shift modules (TSMs) are used in the dynamic-temporal stream, which enables temporal modeling with no additional parameters by mixing ME motion information from two different temporal domains. Furthermore, we introduce self-knowledge distillation (SKD) into the MER task by introducing auxiliary classifiers and using the deepest section of the network for supervision, encouraging all blocks to fully explore the features of the training set. Finally, extensive experiments are conducted on four ME datasets: CASME II, SAMM, MMEW, and CAS(ME)3. The experimental results demonstrate that our SKD-TSTSAN outperforms other existing methods and achieves new state-of-the-art performance. Our code will be available at https://github.com/GuanghaoZhu663/SKD-TSTSAN.
Abstract:Trajectory prediction plays a vital role in automotive radar systems, facilitating precise tracking and decision-making in autonomous driving. Generative adversarial networks with the ability to learn a distribution over future trajectories tend to predict out-of-distribution samples, which typically occurs when the distribution of forthcoming paths comprises a blend of various manifolds that may be disconnected. To address this issue, we propose a trajectory prediction framework, which can capture the social interaction variations and model disconnected manifolds of pedestrian trajectories. Our framework is based on a fused spatiotemporal graph to better model the complex interactions of pedestrians in a scene, and a multi-generator architecture that incorporates a flexible generator selector network on generated trajectories to learn a distribution over multiple generators. We show that our framework achieves state-of-the-art performance compared with several baselines on different challenging datasets.
Abstract:This study presents a novel multimodal fusion model for three-dimensional mineral prospectivity mapping (3D MPM), effectively integrating structural and fluid information through a deep network architecture. Leveraging Convolutional Neural Networks (CNN) and Multilayer Perceptrons (MLP), the model employs canonical correlation analysis (CCA) to align and fuse multimodal features. Rigorous evaluation on the Jiaojia gold deposit dataset demonstrates the model's superior performance in distinguishing ore-bearing instances and predicting mineral prospectivity, outperforming other models in result analyses. Ablation studies further reveal the benefits of joint feature utilization and CCA incorporation. This research not only advances mineral prospectivity modeling but also highlights the pivotal role of data integration and feature alignment for enhanced exploration decision-making.
Abstract:As a team studying the predictors of complications after lung surgery, we have encountered high missingness of data on one-lung ventilation (OLV) start and end times due to high clinical workload and cognitive overload during surgery. Such missing data limit the precision and clinical applicability of our findings. We hypothesized that available intraoperative mechanical ventilation and physiological time-series data combined with other clinical events could be used to accurately predict missing start and end times of OLV. Such a predictive model can recover existing miss-documented records and relieves the documentation burden by deploying it in clinical settings. To this end, we develop a deep learning model to predict the occurrence and timing of OLV based on routinely collected intraoperative data. Our approach combines the variables' spatial and frequency domain features, using Transformer encoders to model the temporal evolution and convolutional neural network to abstract frequency-of-interest from wavelet spectrum images. The performance of the proposed method is evaluated on a benchmark dataset curated from Massachusetts General Hospital (MGH) and Brigham and Women's Hospital (BWH). Experiments show our approach outperforms baseline methods significantly and produces a satisfactory accuracy for clinical use.
Abstract:The three-dimensional (3D) geological models are the typical and key data source in the 3D mineral prospecitivity modeling. Identifying prospectivity-informative predictor variables from the 3D geological models is a challenging and tedious task. Motivated by the ability of convolutional neural networks (CNNs) to learn the intrinsic features, in this paper, we present a novel method that leverages CNNs to learn 3D mineral prospectivity from the 3D geological models. By exploiting the learning ability of CNNs, the presented method allows for disentangling complex correlation to the mineralization and thus opens a door to circumvent the tedious work for designing the predictor variables. Specifically, to explore the unstructured 3D geological models with the CNNs whose input should be structured, we develop a 2D CNN framework in which the geometry of geological boundary is compiled and reorganized into multi-channel images and fed into the CNN. This ensures an effective and efficient training of CNNs while allowing the prospective model to approximate the ore-forming process. The presented method is applied to a typical structure-controlled hydrothermal deposit, the Dayingezhuang gold deposit, eastern China, in which the presented method was compared with the prospectivity modeling methods using hand-designed predictor variables. The results demonstrate the presented method capacitates a performance boost of the 3D prospectivity modeling and empowers us to decrease work-load and prospecting risk in prediction of deep-seated orebodies.