Billion-Scale Graph Deep Learning Framework for Ads Recommendation

Temporal graph learning focuses on graph deep learning in real-world dynamic scenarios, which uses interaction sequence instead of adjacency matrix to observe the graph dynamic changes more microscopically from the perspective of time evolution. However, current temporal graph methods only focus on extra dynamic information, ignoring the large amount of multi-modal information contained in the real world. These information can reflect the rich changes in the real world from different perspectives. Ignoring them means that temporal graph learning still lacks the ability to restore and mine more complex real-world data. We argue that the main challenges causing the above phenomenon in temporal graph learning are the lack of multi-modal architecture and public multi-modal datasets. To solve the above challenges, we propose ModalTGL, which enhances the computational efficiency of the model in complex dynamic scenarios by introducing the dictionary graph network, and achieves multi-modal fusion by embedding tuning. In addition, we also discuss the effects of different time encoding functions on dynamic information preservation. At the data level, we build several multi-modal temporal graph datasets from different areas, and compare with multiple SOTA methods on these datasets. The experimental results verify the effectiveness of the ModalTGL method, achieving the performance improvement of up to 18.48%. Code and data can be obtained from https://github.com/MGitHubL/ModalTGL.

Three-dimensional (3D) medical images, such as Computed Tomography (CT) and Magnetic Resonance Imaging (MRI), are essential for clinical applications. However, the need for diverse and comprehensive representations is particularly pronounced when considering the variability across different organs, diagnostic tasks, and imaging modalities. How to effectively interpret the intricate contextual information and extract meaningful insights from these images remains an open challenge to the community. While current self-supervised learning methods have shown potential, they often consider an image as a whole thereby overlooking the extensive, complex relationships among local regions from one or multiple images. In this work, we introduce a pioneering method for learning 3D medical image representations through an autoregressive pre-training framework. Our approach sequences various 3D medical images based on spatial, contrast, and semantic correlations, treating them as interconnected visual tokens within a token sequence. By employing an autoregressive sequence modeling task, we predict the next visual token in the sequence, which allows our model to deeply understand and integrate the contextual information inherent in 3D medical images. Additionally, we implement a random startup strategy to avoid overestimating token relationships and to enhance the robustness of learning. The effectiveness of our approach is demonstrated by the superior performance over others on nine downstream tasks in public datasets.

Democratizing artificial intelligence (AI) enables model development by clinicians with a lack of coding expertise, powerful computing resources, and large, well-labeled data sets. To determine whether resource-constrained clinicians can use self-training via automated machine learning (ML) and public data sets to design high-performing diabetic retinopathy classification models. This diagnostic quality improvement study was conducted from January 1, 2021, to December 31, 2021. A self-training method without coding was used on 2 public data sets with retinal images from patients in France (Messidor-2 [n = 1748]) and the UK and US (EyePACS [n = 58 689]) and externally validated on 1 data set with retinal images from patients of a private Egyptian medical retina clinic (Egypt [n = 210]). An AI model was trained to classify referable diabetic retinopathy as an exemplar use case. Messidor-2 images were assigned adjudicated labels available on Kaggle; 4 images were deemed ungradable and excluded, leaving 1744 images. A total of 300 images randomly selected from the EyePACS data set were independently relabeled by 3 blinded retina specialists using the International Classification of Diabetic Retinopathy protocol for diabetic retinopathy grade and diabetic macular edema presence; 19 images were deemed ungradable, leaving 281 images. Data analysis was performed from February 1 to February 28, 2021. Using public data sets, a teacher model was trained with labeled images using supervised learning. Next, the resulting predictions, termed pseudolabels, were used on an unlabeled public data set. Finally, a student model was trained with the existing labeled images and the additional pseudolabeled images. The analyzed metrics for the models included the area under the receiver operating characteristic curve (AUROC), accuracy, sensitivity, specificity, and F1 score. The Fisher exact test was performed, and 2-tailed P values were calculated for failure case analysis. For the internal validation data sets, AUROC values for performance ranged from 0.886 to 0.939 for the teacher model and from 0.916 to 0.951 for the student model. For external validation of automated ML model performance, AUROC values and accuracy were 0.964 and 93.3% for the teacher model, 0.950 and 96.7% for the student model, and 0.890 and 94.3% for the manually coded bespoke model, respectively. These findings suggest that self-training using automated ML is an effective method to increase both model performance and generalizability while decreasing the need for costly expert labeling. This approach advances the democratization of AI by enabling clinicians without coding expertise or access to large, well-labeled private data sets to develop their own AI models.

To optimize surgical procedures and prevent retained surgical instruments, precise identification of required instruments during surgical treatment is essential. However, establishing ground truth data can be a labor-intensive barrier for researchers. Therefore, we developed and evaluated a novel system for detecting laparoscopic surgical instruments during laparoscopic cholecystectomy through virtual image creation. Virtual images were created by synthesizing laparoscopic instrument photos with surgical video backgrounds. The 311 instrument images and 1610 background images from 52 patients were augmented through random brightness, contrast, crop, rotation, scaling, flipping, and perspective transformations, resulting in 6023 composite images. These data were split into training, tuning, and internal test sets. Based on synthetic data, we developed a system comprising two-step processes. The first model is a unified instrument localization model that detects surgical instruments, and the second model is an instrument-type classification model that categorizes the detected surgical instruments. External and public datasets were used to evaluate generalizability. The unified instrument localization model achieved average precision (AP) values with intersection over union (IoU) of 0.5 of 0.981, 0.882, and 0.689 for internal, external, and public datasets, respectively. The instrument-type classification model demonstrated area under the curve (AUC) values of 0.959 for seven instrument types in the external dataset and 0.749 for four instrument types in the public dataset. The final two-step instrument detection model demonstrated an AUC of 0.848 for the external dataset and 0.688 for the public dataset, which showed significantly superior performance compared to conventional multi-class instrument models. This validated deep learning model using synthetically generated data provides a reliable framework for surgical instrument detection. Our approach demonstrates strong performance and generalizability, suggesting its potential for improving operative workflow efficiency and surgical education across various minimally invasive procedures.

Deep neural networks have demonstrated promising performance in screening mammography with recent studies reporting performance at or above the level of trained radiologists on internal datasets. However, it remains unclear whether the performance of these trained models is robust and replicates across external datasets. In this study, we evaluate four state-of-the-art publicly available models using four publicly available mammography datasets (CBIS-DDSM, INbreast, CMMD, OMI-DB). Where test data was available, published results were replicated. The best-performing model, which achieved an area under the ROC curve (AUC) of 0.88 on internal data from NYU, achieved here an AUC of 0.9 on the external CMMD dataset (N = 826 exams). On the larger OMI-DB dataset (N = 11,440 exams), it achieved an AUC of 0.84 but did not match the performance of individual radiologists (at a specificity of 0.92, the sensitivity was 0.97 for the radiologist and 0.53 for the network for a 1-year follow-up). The network showed higher performance for in situ cancers, as opposed to invasive cancers. Among invasive cancers, it was relatively weaker at identifying asymmetries and was relatively stronger at identifying masses. The three other trained models that we evaluated all performed poorly on external datasets. Independent validation of trained models is an essential step to ensure safe and reliable use. Future progress in AI for mammography may depend on a concerted effort to make larger datasets publicly available that span multiple clinical sites.

Top-$N$ sequential recommendation models each user as a sequence of items interacted in the past and aims to predict top-$N$ ranked items that a user will likely interact in a `near future'. The order of interaction implies that sequential patterns play an important role where more recent items in a sequence have a larger impact on the next item. In this paper, we propose a Convolutional Sequence Embedding Recommendation Model (\emph{Caser}) as a solution to address this requirement. The idea is to embed a sequence of recent items into an `image' in the time and latent spaces and learn sequential patterns as local features of the image using convolutional filters. This approach provides a unified and flexible network structure for capturing both general preferences and sequential patterns. The experiments on public datasets demonstrated that Caser consistently outperforms state-of-the-art sequential recommendation methods on a variety of common evaluation metrics.

Temporal Graph Clustering (TGC) is a new task with little attention, focusing on node clustering in temporal graphs. Compared with existing static graph clustering, it can find the balance between time requirement and space requirement (Time-Space Balance) through the interaction sequence-based batch-processing pattern. However, there are two major challenges that hinder the development of TGC, i.e., inapplicable clustering techniques and inapplicable datasets. To address these challenges, we propose a comprehensive benchmark, called BenchTGC. Specially, we design a BenchTGC Framework to illustrate the paradigm of temporal graph clustering and improve existing clustering techniques to fit temporal graphs. In addition, we also discuss problems with public temporal graph datasets and develop multiple datasets suitable for TGC task, called BenchTGC Datasets. According to extensive experiments, we not only verify the advantages of BenchTGC, but also demonstrate the necessity and importance of TGC task. We wish to point out that the dynamically changing and complex scenarios in real world are the foundation of temporal graph clustering.

In recent years, recommender systems have advanced rapidly, where embedding learning for users and items plays a critical role. A standard method learns a unique embedding vector for each user and item. However, such a method has two important limitations in real-world applications: 1) it is hard to learn embeddings that generalize well for users and items with rare interactions; and 2) it may incur unbearably high memory costs when the number of users and items scales up. Existing approaches either can only address one of the limitations or have flawed overall performances. In this paper, we propose Clustered Embedding Learning (CEL) as an integrated solution to these two problems. CEL is a plug-and-play embedding learning framework that can be combined with any differentiable feature interaction model. It is capable of achieving improved performance, especially for cold users and items, with reduced memory cost. CEL enables automatic and dynamic clustering of users and items in a top-down fashion, where clustered entities jointly learn a shared embedding. The accelerated version of CEL has an optimal time complexity, which supports efficient online updates. Theoretically, we prove the identifiability and the existence of a unique optimal number of clusters for CEL in the context of nonnegative matrix factorization. Empirically, we validate the effectiveness of CEL on three public datasets and one business dataset, showing its consistently superior performance against current state-of-the-art methods. In particular, when incorporating CEL into the business model, it brings an improvement of in AUC, which translates into a significant revenue gain; meanwhile, the size of the embedding table gets 2650 times smaller.1

Automated, anatomically coherent retinal layer segmentation in optical coherence tomography (OCT) is one of the most important components of retinal disease management. However, current methods rely on large amounts of labeled data, which can be difficult and expensive to obtain. In addition, these systems tend often propose anatomically impossible results, which undermines their clinical reliability. This study introduces a semi-supervised approach to retinal layer segmentation that leverages large amounts of unlabeled data and anatomical prior knowledge related to the structure of the retina. During training, we use a novel topological engine that converts inferred retinal layer boundaries into pixel-wise structured segmentations. These compose a set of anatomically valid disentangled representations which, together with predicted style factors, are used to reconstruct the input image. At training time, the retinal layer boundaries and pixel-wise predictions are both guided by reference annotations, where available, but more importantly by innovatively exploiting anatomical priors that improve the performance, robustness and coherence of the method even if only a small amount of labeled data is available. Exhaustive experiments with respect to label efficiency, contribution of unsupervised data and robustness to different acquisition settings were conducted. The proposed method showed state of-the-art performance on all the studied public and internal datasets, specially in low annotated data regimes. Additionally, the model was able to make use of unlabeled data from a different domain with only a small performance drop in comparison to a fully-supervised setting. A novel, robust, label-efficient retinal layer segmentation method was proposed. The approach has shown state-of-the-art layer segmentation performance with a fraction of the training data available, while at the same time, its robustness against domain shift was also shown.

Accurate segmentation of the Clinical Target Volume (CTV) and Organs At Risk (OARs) is a prerequisite for effective cervical cancer radiotherapy. However, this task remains challenging due to the low tissue contrast of the CTV and the complex, tubular geometry of gastrointestinal organs. Existing methods often struggle to delineate indistinct boundaries, leading to suboptimal treatment planning. To address this, we propose BA-UNet, a boundary-augmented framework designed to enforce geometric consistency during both feature encoding and optimization. Specifically, the proposed architecture integrates a Boundary-Infused Feature Aggregation (BIFA) module to inject deterministic edge priors into the multi-scale encoder, explicitly preserving high-frequency boundary information that is often lost in deep feature abstraction. Concurrently, we introduce a novel Boundary-Aware Curvature (BAC) loss that utilizes the Hessian matrix to estimate 3D curvature. This loss function penalizes geometric deviations in complex regions, compelling the network to focus on sharp boundary transitions. Validation on our private dataset and the public WORD dataset demonstrates the superiority of our method. BA-UNet achieved a mean Dice Similarity Coefficient (DSC) of 85.41% and a 95% Hausdorff Distance (HD95) of 11.07 mm on the internal dataset, significantly outperforming state-of-the-art methods, particularly in challenging anatomical structures.

Search systems in online content platforms are typically biased toward a minority of highly consumed items, reflecting the most common user behavior of navigating toward content that is already familiar and popular. Query suggestions are a powerful tool to support query formulation and to encourage exploratory search and content discovery. However, classic approaches for query suggestions typically rely either on semantic similarity, which lacks diversity and does not reflect user searching behavior, or on a collaborative similarity measure mined from search logs, which suffers from data sparsity and is biased by highly popular queries. In this work, we argue that the task of query suggestion can be modelled as a link prediction task on a heterogeneous graph including queries and documents, enabling Graph Learning methods to effectively generate query suggestions encompassing both semantic and collaborative information. We perform an offline evaluation on an internal Spotify dataset of search logs and on two public datasets, showing that node2vec leads to an accurate and diversified set of results, especially on the large scale real-world data. We then describe the implementation in an instant search scenario and discuss a set of additional challenges tied to the specific production environment. Finally, we report the results of a large scale A/B test involving millions of users and prove that node2vec query suggestions lead to an increase in online metrics such as coverage (+1.42% shown search results pages with suggestions) and engagement (+1.21% clicks), with a specifically notable boost in the number of clicks on exploratory search queries (+9.37%).

ObjectiveTo assess the performance of the current foundation models in histopathology. Patients and MethodsThe assessment involves a comprehensive evaluation of some foundation models, such as the CLIP derivatives, namely PLIP and BiomedCLIP, which were fine-tuned on data scraped from the internet. The comparison is performed against simpler and nonfoundational histology models that are trained on well-curated data, eg, the cancer genome atlas. All models are evaluated on 8 datasets, 4 of which are internal histology datasets collected and curated at Mayo Clinic, and 4 well-known public datasets: PANDA, BRACS, CAMELYON16, and DigestPath. Evaluation metrics include accuracy and macro-averaged F1 score, using a majority vote among top-k (eg, MV@5) at the whole slide image/patch levels. Moreover, all models are evaluated in classification settings. This detailed analysis allows for a deep understanding of each model’s performance across various datasets. ResultsIn various evaluation tasks, domain-specific (and nonfoundational) models like DinoSSLPath and KimiaNet outperform general-purpose foundation models. The DinoSSLPath excels in whole slide image-level retrieval for internal colorectal cancer and liver datasets with MV@5 macro-averaged F1 scores of 63% and 74%, respectively. The KimiaNet leads in breast and skin cancer tasks with respective Top-1 and MV@5 scores of 56% and 70%, respectively and scores 75% on the public CAMELYON16 dataset. Similar trends are observed in patch-level metrics, highlighting the advantage of using specialized datasets like the cancer genome atlas for histopathological analysis. ConclusionTo enable effective vision-language foundation models in biomedicine, high-quality, multi-modal medical datasets are essential. These datasets serve as the substrate for training models capable of translating research into clinical practice. Of importance, the alignment (correspondence) between textual and visual data—often diagnostic—is critical and requires validation by domain experts. Thus, advancing foundation models in this field necessitates collaborative efforts in data curation and validation.

Deep learning models for human activity recognition (HAR) based on sensor data have been heavily studied recently. However, the generalization ability of deep models on complex real-world HAR data is limited by the availability of high-quality labeled activity data, which are hard to obtain. In this article, we design a similarity embedding neural network that maps input sensor signals onto real vectors through carefully designed convolutional and Long Short-Term Memory (LSTM) layers. The embedding network is trained with a pairwise similarity loss, encouraging the clustering of samples from the same class in the embedded real space, and can be effectively trained on a small dataset and even on a noisy dataset with mislabeled samples. Based on the learned embeddings, we further propose both nonparametric and parametric approaches for activity recognition. Extensive evaluation based on two public datasets has shown that the proposed similarity embedding network significantly outperforms state-of-the-art deep models on HAR classification tasks, is robust to mislabeled samples in the training set, and can also be used to effectively denoise a noisy dataset.

Analyzing the temporal behavior of nodes in time-varying graphs is useful for many applications such as targeted advertising, community evolution and outlier detection. In this paper, we present a novel approach, STWalk, for learning trajectory representations of nodes in temporal graphs. The proposed framework makes use of structural properties of graphs at current and previous time-steps to learn effective node trajectory representations. STWalk performs random walks on a graph at a given time step (called space-walk) as well as on graphs from past time-steps (called time-walk) to capture the spatio-temporal behavior of nodes. We propose two variants of STWalk to learn trajectory representations. In one algorithm, we perform space-walk and time-walk as part of a single step. In the other variant, we perform space-walk and time-walk separately and combine the learned representations to get the final trajectory embedding. Extensive experiments on three real-world temporal graph datasets validate the effectiveness of the learned representations when compared to three baseline methods. We also show the goodness of the learned trajectory embeddings for change point detection, as well as demonstrate that arithmetic operations on these trajectory representations yield interesting and interpretable results.

Recently, representation learning on temporal graphs has drawn increasing attention, which aims at learning temporal patterns to characterize the evolving nature of dynamic graphs in real-world applications. Despite effectiveness, these methods commonly ignore the individual- and combinatorial-level patterns derived from different types of interactions (e.g.,user-item), which are at the heart of the representation learning on temporal graphs. To fill this gap, we propose MERIT, a novel multi-level graph attention network for inductive representation learning on temporal graphs.We adaptively embed the original timestamps to a higher, continuous dimensional space for learn-ing individual-level periodicity through Personalized Time Encoding (PTE) module. Furthermore, we equip MERIT with Continuous time and Con-text aware Attention (Coco-Attention) mechanism which chronologically locates most relevant neighbors by jointly capturing multi-level context on temporal graphs. Finally, MERIT performs multiple aggregations and propagations to explore and exploit high-order structural information for down-stream tasks. Extensive experiments on four public datasets demonstrate the effectiveness of MERITon both (inductive / transductive) link prediction and node classification task.