Multi-domain learning (MDL) has become a prominent topic in enhancing the quality of personalized services. It's critical to learn commonalities between domains and preserve the distinct characteristics of each domain. However, this leads to a challenging dilemma in MDL. On the one hand, a model needs to leverage domain-aware modules such as experts or embeddings to preserve each domain's distinctiveness. On the other hand, real-world datasets often exhibit long-tailed distributions across domains, where some domains may lack sufficient samples to effectively train their specific modules. Unfortunately, nearly all existing work falls short of resolving this dilemma. To this end, we propose a novel Cross-experts Covariance Loss for Disentangled Learning model (Crocodile), which employs multiple embedding tables to make the model domain-aware at the embeddings which consist most parameters in the model, and a covariance loss upon these embeddings to disentangle them, enabling the model to capture diverse user interests among domains. Empirical analysis demonstrates that our method successfully addresses both challenges and outperforms all state-of-the-art methods on public datasets. During online A/B testing in Tencent's advertising platform, Crocodile achieves 0.72% CTR lift and 0.73% GMV lift on a primary advertising scenario. The code is openly accessible at: https://github.com/SkylerLinn/Crocodile.

Lifelong user behavior sequences are crucial for capturing user interests and predicting user responses in modern recommendation systems. A two-stage paradigm is typically adopted to handle these long sequences: a subset of relevant behaviors is first searched from the original long sequences via an attention mechanism in the first stage and then aggregated with the target item to construct a discriminative representation for prediction in the second stage. In this work, we identify and characterize, for the first time, a neglected deficiency in existing long-sequence recommendation models: a single set of embeddings struggles with learning both attention and representation, leading to interference between these two processes. Initial attempts to address this issue with some common methods (e.g., linear projections---a technique borrowed from language processing) proved ineffective, shedding light on the unique challenges of recommendation models. To overcome this, we propose the Decoupled Attention and Representation Embeddings (DARE) model, where two distinct embedding tables are initialized and learned separately to fully decouple attention and representation. Extensive experiments and analysis demonstrate that DARE provides more accurate searches of correlated behaviors and outperforms baselines with AUC gains up to 9‰ on public datasets and notable improvements on Tencent's advertising platform. Furthermore, decoupling embedding spaces allows us to reduce the attention embedding dimension and accelerate the search procedure by 50\% without significant performance impact, enabling more efficient, high-performance online serving. Code in PyTorch for experiments, including model analysis, is available at https://github.com/thuml/DARE.

Click-Through Rate (CTR) prediction, a core task in recommendation systems, aims to estimate the probability of users clicking on items. Existing models predominantly follow a discriminative paradigm, which relies heavily on explicit interactions between raw ID embeddings. However, this paradigm inherently renders them susceptible to two critical issues: embedding dimensional collapse and information redundancy, stemming from the over-reliance on feature interactions over raw ID embeddings. To address these limitations, we propose a novel Supervised Feature Generation (SFG) framework, shifting the paradigm from discriminative "feature interaction" to generative "feature generation". Specifically, SFG comprises two key components: an Encoder that constructs hidden embeddings for each feature, and a Decoder tasked with regenerating the feature embeddings of all features from these hidden representations. Unlike existing generative approaches that adopt self-supervised losses, we introduce a supervised loss to utilize the supervised signal, i.e., click or not, in the CTR prediction task. This framework exhibits strong generalizability: it can be seamlessly integrated with most existing CTR models, reformulating them under the generative paradigm. Extensive experiments demonstrate that SFG consistently mitigates embedding collapse and reduces information redundancy, while yielding substantial performance gains across various datasets and base models. The code is available at https://github.com/USTC-StarTeam/GE4Rec.

Multi-task learning (MTL) has gained significant popularity in recommender systems as it enables simultaneous optimization of multiple objectives. A key challenge in MTL is negative transfer, but existing studies explored negative transfer on all samples, overlooking the inherent complexities within them. We split the samples according to the relative amount of positive feedback among tasks. Surprisingly, negative transfer still occurs in existing MTL methods on samples that receive comparable feedback across tasks. Existing work commonly employs a shared-embedding paradigm, limiting the ability of modeling diverse user preferences on different tasks. In this paper, we introduce a novel Shared and Task-specific EMbeddings (STEM) paradigm that aims to incorporate both shared and task-specific embeddings to effectively capture task-specific user preferences. Under this paradigm, we propose a simple model STEM-Net, which is equipped with an All Forward Task-specific Backward gating network to facilitate the learning of task-specific embeddings and direct knowledge transfer across tasks. Remarkably, STEM-Net demonstrates exceptional performance on comparable samples, achieving positive transfer. Comprehensive evaluation on three public MTL recommendation datasets demonstrates that STEM-Net outperforms state-of-the-art models by a substantial margin. Our code is released at https://github.com/LiangcaiSu/STEM.

Multi-domain learning (MDL) aims to train a model with minimal average risk across multiple overlapping but non-identical domains. To tackle the challenges of dataset bias and domain domination, numerous MDL approaches have been proposed from the perspectives of seeking commonalities by aligning distributions to reduce domain gap or reserving differences by implementing domain-specific towers, gates, and even experts. MDL models are becoming more and more complex with sophisticated network architectures or loss functions, introducing extra parameters and enlarging computation costs. In this paper, we propose a frustratingly easy and hyperparameter-free multi-domain learning method named Decoupled Training (D-Train). D-Train is a tri-phase general-to-specific training strategy that first pre-trains on all domains to warm up a root model, then post-trains on each domain by splitting into multi-heads, and finally fine-tunes the heads by fixing the backbone, enabling decouple training to achieve domain independence. Despite its extraordinary simplicity and efficiency, D-Train performs remarkably well in extensive evaluations of various datasets from standard benchmarks to applications of satellite imagery and recommender systems.

Recent advances in foundation models have led to a promising trend of developing large recommendation models to leverage vast amounts of available data. Still, mainstream models remain embarrassingly small in size and naive enlarging does not lead to sufficient performance gain, suggesting a deficiency in the model scalability. In this paper, we identify the embedding collapse phenomenon as the inhibition of scalability, wherein the embedding matrix tends to occupy a low-dimensional subspace. Through empirical and theoretical analysis, we demonstrate a two-sided effect of feature interaction specific to recommendation models. On the one hand, interacting with collapsed embeddings restricts embedding learning and exacerbates the collapse issue. On the other hand, interaction is crucial in mitigating the fitting of spurious features as a scalability guarantee. Based on our analysis, we propose a simple yet effective multi-embedding design incorporating embedding-set-specific interaction modules to learn embedding sets with large diversity and thus reduce collapse. Extensive experiments demonstrate that this proposed design provides consistent scalability and effective collapse mitigation for various recommendation models. Code is available at this repository: https://github.com/thuml/Multi-Embedding.

Click-through rate (CTR) prediction is a crucial area of research in online advertising. While binary cross entropy (BCE) has been widely used as the optimization objective for treating CTR prediction as a binary classification problem, recent advancements have shown that combining BCE loss with an auxiliary ranking loss can significantly improve performance. However, the full effectiveness of this combination loss is not yet fully understood. In this paper, we uncover a new challenge associated with the BCE loss in scenarios where positive feedback is sparse: the issue of gradient vanishing for negative samples. We introduce a novel perspective on the effectiveness of the auxiliary ranking loss in CTR prediction: it generates larger gradients on negative samples, thereby mitigating the optimization difficulties when using the BCE loss only and resulting in improved classification ability. To validate our perspective, we conduct theoretical analysis and extensive empirical evaluations on public datasets. Additionally, we successfully integrate the ranking loss into Tencent's online advertising system, achieving notable lifts of 0.70% and 1.26% in Gross Merchandise Value (GMV) for two main scenarios. The code is openly accessible at: https://github.com/SkylerLinn/Understanding-the-Ranking-Loss.

We present Tencent's ads recommendation system and examine the challenges and practices of learning appropriate recommendation representations. Our study begins by showcasing our approaches to preserving prior knowledge when encoding features of diverse types into embedding representations. We specifically address sequence features, numeric features, and pre-trained embedding features. Subsequently, we delve into two crucial challenges related to feature representation: the dimensional collapse of embeddings and the interest entanglement across different tasks or scenarios. We propose several practical approaches to address these challenges that result in robust and disentangled recommendation representations. We then explore several training techniques to facilitate model optimization, reduce bias, and enhance exploration. Additionally, we introduce three analysis tools that enable us to study feature correlation, dimensional collapse, and interest entanglement. This work builds upon the continuous efforts of Tencent's ads recommendation team over the past decade. It summarizes general design principles and presents a series of readily applicable solutions and analysis tools. The reported performance is based on our online advertising platform, which handles hundreds of billions of requests daily and serves millions of ads to billions of users.

In the open world, various label sets and domain configurations give rise to a variety of Domain Adaptation (DA) setups, including closed-set, partial-set, open-set, and universal DA, as well as multi-source and multi-target DA. It is notable that existing DA methods are generally designed only for a specific setup, and may under-perform in setups they are not tailored to. This paper shifts the common paradigm of DA to Versatile Domain Adaptation (VDA), where one method can handle several different DA setups without any modification. Towards this goal, we first delve into a general inductive bias: class confusion, and then uncover that reducing such pairwise class confusion leads to significant transfer gains. With this insight, we propose one general class confusion loss (CC-Loss) to learn many setups. We estimate class confusion based only on classifier predictions and minimize the class confusion to enable accurate target predictions. Further, we improve the loss by enforcing the consistency of confusion matrices under different data augmentations to encourage its invariance to distribution perturbations. Experiments on 2D vision and 3D vision benchmarks show that the CC-Loss performs competitively in different mainstream DA setups.

Multi-task learning (MTL) models have demonstrated impressive results in computer vision, natural language processing, and recommender systems. Even though many approaches have been proposed, how well these approaches balance different tasks on each parameter still remains unclear. In this paper, we propose to measure the task dominance degree of a parameter by the total updates of each task on this parameter. Specifically, we compute the total updates by the exponentially decaying Average of the squared Updates (AU) on a parameter from the corresponding task. Based on this novel metric, we observe that many parameters in existing MTL methods, especially those in the higher shared layers, are still dominated by one or several tasks. The dominance of AU is mainly due to the dominance of accumulative gradients from one or several tasks. Motivated by this, we propose a Task-wise Adaptive learning rate approach, AdaTask in short, to separate the accumulative gradients and hence the learning rate of each task for each parameter in adaptive learning rate approaches (e.g., AdaGrad, RMSProp, and Adam). Comprehensive experiments on computer vision and recommender system MTL datasets demonstrate that AdaTask significantly improves the performance of dominated tasks, resulting SOTA average task-wise performance. Analysis on both synthetic and real-world datasets shows AdaTask balance parameters in every shared layer well.

Out-of-distribution (OOD) generalization, where the model needs to handle distribution shifts from training, is a major challenge of machine learning. Contrastive language-image pre-training (CLIP) models have shown impressive zero-shot ability, but the further adaptation of CLIP on downstream tasks undesirably degrades OOD performances. This paper aims at generalizing CLIP to out-of-distribution test data on downstream tasks. We propose CLIPood, a fine-tuning method that can adapt CLIP models to OOD situations where both domain shifts and open classes may occur on the unseen test data. To exploit the semantic relations between classes from the text modality, CLIPood introduces a new training objective, margin metric softmax (MMS), with class adaptive margins for fine-tuning. To incorporate both pre-trained zero-shot model and fine-tuned task-adaptive model, CLIPood leverages a new optimization strategy, Beta moving average (BMA), to maintain a temporal ensemble weighted by Beta distribution. Experiments on diverse datasets with different OOD scenarios show that CLIPood consistently outperforms existing generalization techniques.

Auxiliary-Task Learning (ATL) aims to improve the performance of the target task by leveraging the knowledge obtained from related tasks. Occasionally, learning multiple tasks simultaneously results in lower accuracy than learning only the target task, which is known as negative transfer. This problem is often attributed to the gradient conflicts among tasks, and is frequently tackled by coordinating the task gradients in previous works. However, these optimization-based methods largely overlook the auxiliary-target generalization capability. To better understand the root cause of negative transfer, we experimentally investigate it from both optimization and generalization perspectives. Based on our findings, we introduce ForkMerge, a novel approach that periodically forks the model into multiple branches, automatically searches the varying task weights by minimizing target validation errors, and dynamically merges all branches to filter out detrimental task-parameter updates. On a series of auxiliary-task learning benchmarks, ForkMerge outperforms existing methods and effectively mitigates negative transfer.

Deep neural networks achieve remarkable performances on a wide range of tasks with the aid of large-scale labeled datasets. Yet these datasets are time-consuming and labor-exhaustive to obtain on realistic tasks. To mitigate the requirement for labeled data, self-training is widely used in semi-supervised learning by iteratively assigning pseudo labels to unlabeled samples. Despite its popularity, self-training is well-believed to be unreliable and often leads to training instability. Our experimental studies further reveal that the bias in semi-supervised learning arises from both the problem itself and the inappropriate training with potentially incorrect pseudo labels, which accumulates the error in the iterative self-training process. To reduce the above bias, we propose Debiased Self-Training (DST). First, the generation and utilization of pseudo labels are decoupled by two parameter-independent classifier heads to avoid direct error accumulation. Second, we estimate the worst case of self-training bias, where the pseudo labeling function is accurate on labeled samples, yet makes as many mistakes as possible on unlabeled samples. We then adversarially optimize the representations to improve the quality of pseudo labels by avoiding the worst case. Extensive experiments justify that DST achieves an average improvement of 6.3% against state-of-the-art methods on standard semi-supervised learning benchmark datasets and 18.9% against FixMatch on 13 diverse tasks. Furthermore, DST can be seamlessly adapted to other self-training methods and help stabilize their training and balance performance across classes in both cases of training from scratch and finetuning from pre-trained models.

Domain adaptation (DA) aims at transferring knowledge from a labeled source domain to an unlabeled target domain. Though many DA theories and algorithms have been proposed, most of them are tailored into classification settings and may fail in regression tasks, especially in the practical keypoint detection task. To tackle this difficult but significant task, we present a method of regressive domain adaptation (RegDA) for unsupervised keypoint detection. Inspired by the latest theoretical work, we first utilize an adversarial regressor to maximize the disparity on the target domain and train a feature generator to minimize this disparity. However, due to the high dimension of the output space, this regressor fails to detect samples that deviate from the support of the source. To overcome this problem, we propose two important ideas. First, based on our observation that the probability density of the output space is sparse, we introduce a spatial probability distribution to describe this sparsity and then use it to guide the learning of the adversarial regressor. Second, to alleviate the optimization difficulty in the high-dimensional space, we innovatively convert the minimax game in the adversarial training to the minimization of two opposite goals. Extensive experiments show that our method brings large improvement by 8% to 11% in terms of PCK on different datasets.

Deep learning has made revolutionary advances to diverse applications in the presence of large-scale labeled datasets. However, it is prohibitively time-costly and labor-expensive to collect sufficient labeled data in most realistic scenarios. To mitigate the requirement for labeled data, semi-supervised learning (SSL) focuses on simultaneously exploring both labeled and unlabeled data, while transfer learning (TL) popularizes a favorable practice of fine-tuning a pre-trained model to the target data. A dilemma is thus encountered: Without a decent pre-trained model to provide an implicit regularization, SSL through self-training from scratch will be easily misled by inaccurate pseudo-labels, especially in large-sized label space; Without exploring the intrinsic structure of unlabeled data, TL through fine-tuning from limited labeled data is at risk of under-transfer caused by model shift. To escape from this dilemma, we present Self-Tuning to enable data-efficient deep learning by unifying the exploration of labeled and unlabeled data and the transfer of a pre-trained model, as well as a Pseudo Group Contrast (PGC) mechanism to mitigate the reliance on pseudo-labels and boost the tolerance to false labels. Self-Tuning outperforms its SSL and TL counterparts on five tasks by sharp margins, e.g. it doubles the accuracy of fine-tuning on Cars with $15%$ labels.

Domain Adaptation (DA) aims at transferring knowledge from a labeled source domain to an unlabeled target domain. While re- markable advances have been witnessed recently, the power of DA methods still heavily depends on the network depth, especially when the domain discrepancy is large, posing an unprecedented challenge to DA in low-resource scenarios where fast and adaptive inference is required. How to bridge transferability and resource- efficient inference in DA becomes an important problem. In this paper, we propose Resource Efficient Domain Adaptation (REDA), a general framework that can adaptively adjust computation re- sources across 'easier' and 'harder' inputs. Based on existing multi- exit architectures, REDA has two novel designs: 1) Transferable distillation to distill the transferability of top classifier into the early exits; 2) Consistency weighting to control the distillation degree via prediction consistency. As a general method, REDA can be easily applied with a variety of DA methods. Empirical results and analy- ses justify that REDA can substantially improve the accuracy and accelerate the inference under domain shift and low resource.

Domain Adaptation (DA) enables transferring a learning machine from a labeled source domain to an unlabeled target one. While remarkable advances have been made, most of the existing DA methods focus on improving the target accuracy at inference. How to estimate the predictive uncertainty of DA models is vital for decision-making in safety-critical scenarios but remains the boundary to explore. In this paper, we delve into the open problem of Calibration in DA, which is extremely challenging due to the coexistence of domain shift and the lack of target labels. We first reveal the dilemma that DA models learn higher accuracy at the expense of well-calibrated probabilities. Driven by this finding, we propose Transferable Calibration (TransCal) to achieve more accurate calibration with lower bias and variance in a unified hyperparameter-free optimization framework. As a general post-hoc calibration method, TransCal can be easily applied to recalibrate existing DA methods. Its efficacy has been justified both theoretically and empirically.

Recent work in domain adaptation bridges different domains by adversarially learning a domain-invariant representation that cannot be distinguished by a domain discriminator. Existing methods of adversarial domain adaptation mainly align the global images across the source and target domains. However, it is obvious that not all regions of an image are transferable, while forcefully aligning the untransferable regions may lead to negative transfer. Furthermore, some of the images are significantly dissimilar across domains, resulting in weak image-level transferability. To this end, we present Transferable Attention for Domain Adaptation (TADA), focusing our adaptation model on transferable regions or images. We implement two types of complementary transferable attention: transferable local attention generated by multiple region-level domain discriminators to highlight transferable regions, and transferable global attention generated by single image-level domain discriminator to highlight transferable images. Extensive experiments validate that our proposed models exceed state of the art results on standard domain adaptation datasets.

Deep unsupervised domain adaptation (Deep UDA) methods successfully leverage rich labeled data in a source domain to boost the performance on related but unlabeled data in a target domain. However, algorithm comparison is cumbersome in Deep UDA due to the absence of accurate and standardized model selection method, posing an obstacle to further advances in the field. Existing model selection methods for Deep UDA are either highly biased, restricted, unstable, or even controversial (requiring labeled target data). To this end, we propose Deep Embedded Validation (DEV), which embeds adapted feature representation into the validation procedure to obtain unbiased estimation of the target risk with bounded variance. The variance is further reduced by the technique of control variate. The efficacy of the method has been justified both theoretically and empirically.

Deep neural networks (DNNs) excel at learning representations when trained on large-scale datasets. Pre-trained DNNs also show strong transferability when fine-tuned to other labeled datasets. However, such transferability becomes weak when the target dataset is fully unlabeled as in Unsupervised Domain Adaptation (UDA). We envision that the loss of transferability may stem from the intrinsic limitation of the architecture design of DNNs. In this paper, we delve into the components of DNN architectures and propose Transferable Normalization (TransNorm) in place of existing normalization techniques. TransNorm is an end-to-end trainable layer to make DNNs more transferable across domains. As a general method, TransNorm can be easily applied to various deep neural networks and domain adaption methods, without introducing any extra hyper-parameters or learnable parameters. Empirical results justify that TransNorm not only improves classification accuracies but also accelerates convergence for mainstream DNN-based domain adaptation methods.