Large Language Models (LLMs) perform well across many tasks but degrade when processing large collections of repetitive or highly similar inputs, a common scenario in applications such as near-duplicate search results and large e-commerce catalogs. In these settings, concatenation-based approaches—long-context prompting and supervised fine-tuning—suffer from attention saturation and diminished signal-to-noise ratio, causing models to miss subtle but important distinctions as input size grows. We introduce SEER (Set Encoding for Efficient Representation), a framework that enables LLMs to handle massive sets of near-duplicate items through a single learned token. SEER first encodes individual items with a pretrained embedding model, then aggregates them using an adapter that captures inter-item relationships and preserves fine-grained differences while mitigating redundancy. To ensure both discriminative and generative capabilities, we propose a multi-task alignment strategy that supervises set-level descriptions across multiple semantic dimensions. Experiments on a large-scale e-commerce dataset demonstrate that SEER substantially outperforms in-context and fine-tuned LLM baselines, maintaining stable performance even when processing thousands of highly similar items. These results establish SEER as an effective and scalable approach for LLM processing of dense, redundant input sets.

Recent research on Generalized Zero-Shot Learning (GZSL) has focused primarily on generation-based methods. However, current literature has overlooked the fundamental principles of these methods and has made limited progress in a complex manner. In this paper, we aim to deconstruct the generator-classifier framework and provide guidance for its improvement and extension. We begin by breaking down the generator-learned unseen class distribution into class-level and instance-level distributions. Through our analysis of the role of these two types of distributions in solving the GZSL problem, we generalize the focus of the generation-based approach, emphasizing the importance of (i) attribute generalization in generator learning and (ii) independent classifier learning with partially biased data. We present a simple method based on this analysis that outperforms SotAs on four public GZSL datasets, demonstrating the validity of our deconstruction. Furthermore, our proposed method remains effective even without a generative model, representing a step towards simplifying the generator-classifier structure. Our code is available at https://github.com/cdb342/DGZ.

The recent trend in unsupervised hashing requires not only a discrete representation space but also the ability to mine the similarities between data points. Determining and maintaining the relations between each datum and all the others maximally utilize the semantic diversity of the training set, but can we explore this on the holistic dataset using a single network end-to-end? In this paper, we take a step towards this vision by proposing Overview Hashing (OH). OH unifies the two ultimate goals of unsupervised hashing, i.e., (1) encoding compact features and (2) learning data similarities on a large scale end-to-end, into one model. In particular, we split the top of an encoder into a binary hash head and a continuous one. For an arbitrary datum, its similarities to all the others in the dataset are reflected in the Hamming distances of their hash heads. The distances then act as the weights to aggregate the continuous heads, shaping the final representation of this datum for loss computation. Hence, training with this representation simultaneously tunes the similarities of this datum to the whole dataset. In the context of a contrastive learning framework, we theoretically endorse our design by linking it to knowledge distillation and the attention mechanisms. Our experiments on the benchmarked datasets show the superiority of OH over the state-of-the-art hashing methods. Code is available at \hrefhttps://github.com/RosieYuu/OH \textcolorred https://github.com/RosieYuu/OH.

User post-click conversion prediction is of high interest to researchers and developers. Recent studies employ multi-task learning to tackle the selection bias and data sparsity problem, two severe challenges in post-click behavior prediction, by incorporating click data. However, prior works mainly foucsed on pointwise learning and the orders of labels (i.e., click and post-click) are not well explored, which naturally poses a listwise learning problem. Inspired by recent advances on differentiable sorting, in this paper, we propose a novel multi-task framework that leverages orders of user behaviors to predict user post-click conversion in an end-to-end approach. Specifically, we define an aggregation operator to combine predicted outputs of different tasks to a unified score, then we use the computed scores to model the label relations via differentiable sorting. Extensive experiments on public and industrial datasets show the superiority of our proposed model against competitive baselines.

Abstract Remote photoplethysmography (rPPG), which aims at measuring heart activities and physiological signals from facial video without any contact, has great potential in many applications (e.g., remote healthcare and affective computing). Recent deep learning approaches focus on mining subtle rPPG clues using convolutional neural networks with limited spatio-temporal receptive fields, which neglect the long-range spatio-temporal perception and interaction for rPPG modeling. In this paper, we propose two end-to-end video transformer based architectures, namely PhysFormer and PhysFormer++, to adaptively aggregate both local and global spatio-temporal features for rPPG representation enhancement. As key modules in PhysFormer, the temporal difference transformers first enhance the quasi-periodic rPPG features with temporal difference guided global attention, and then refine the local spatio-temporal representation against interference. To better exploit the temporal contextual and periodic rPPG clues, we also extend the PhysFormer to the two-pathway SlowFast based PhysFormer++ with temporal difference periodic and cross-attention transformers. Furthermore, we propose the label distribution learning and a curriculum learning inspired dynamic constraint in frequency domain, which provide elaborate supervisions for PhysFormer and PhysFormer++ and alleviate overfitting. Comprehensive experiments are performed on four benchmark datasets to show our superior performance on both intra- and cross-dataset testings. Unlike most transformer networks needed pretraining from large-scale datasets, the proposed PhysFormer family can be easily trained from scratch on rPPG datasets, which makes it promising as a novel transformer baseline for the rPPG community.

Zero-shot learning (ZSL) aims to recognize unknown categories that are unavailable during training. Recently, generative models have shown the potential to address this challenging problem by synthesizing unseen features conditioned on semantic embeddings such as attributes. However, unidirectional generative models cannot guarantee the effective coupling between visual and semantic spaces. To this end, we propose a visual-semantic aligned bidirectional network with cycle consistency to alleviate the gap between these two spaces, generating unseen features of high quality. More importantly, we incorporate two carefully designed strategies into our bidirectional framework to improve the overall ZSL performance. Specifically, we enhance the intra-domain class divergence in both visual and semantic spaces, and in the meantime, mitigate the inter-domain shift to preserve seen-unseen domain discrimination. Experimental results on four standard benchmarks show the superiority of our framework over existing state-of-the-art methods under both conventional and generalized ZSL settings.

The recent advance in deep generative models outlines a promising perspective in the realm of Zero-Shot Learning (ZSL). Most generative ZSL methods use category semantic attributes plus a Gaussian noise to generate visual features. After generating unseen samples, this family of approaches effectively transforms the ZSL problem into a supervised classification scheme. However, the existing models use a single semantic attribute, which contains the complete attribute information of the category. The generated data also carry the complete attribute information, but in reality, visual samples usually have limited attributes. Therefore, the generated data from attribute could have incomplete semantics. Based on this fact, we propose a novel framework to boost ZSL by synthesizing diverse features. This method uses augmented semantic attributes to train the generative model, so as to simulate the real distribution of visual features. We evaluate the proposed model on four benchmark datasets, observing significant performance improvement against the state-of-the-art.

Remote photoplethysmography (rPPG), which aims at measuring heart activities and physiological signals from facial video without any contact, has great potential in many applications. Recent deep learning approaches focus on mining subtle rPPG clues using convolutional neural networks with limited spatio-temporal receptive fields, which neglect the long-range spatio-temporal perception and interaction for rPPG modeling. In this paper, we propose the PhysFormer, an end-to-end video transformer based architecture, to adaptively aggregate both local and global spatio-temporal features for rPPG representation enhancement. As key modules in PhysFormer, the temporal difference transformers first enhance the quasi-periodic rPPG features with temporal difference guided global attention, and then refine the local spatio-temporal representation against interference. Furthermore, we also propose the label distribution learning and a curriculum learning inspired dynamic constraint in frequency domain, which provide elaborate supervisions for PhysFormer and alleviate overfitting. Comprehensive experiments are performed on four benchmark datasets to show our superior performance on both intra- and cross-dataset testings. One highlight is that, unlike most transformer networks needed pretraining from large-scale datasets, the proposed PhysFormer can be easily trained from scratch on rPPG datasets, which makes it promising as a novel transformer baseline for the rPPG community. The codes are available at https://github.com/ZitongYu/PhysFormer.

Semantic-descriptor-based Generalized Zero-Shot Learning (GZSL) poses challenges in recognizing novel classes in the test phase. The development of generative models enables current GZSL techniques to probe further into the semantic-visual link, culminating in a two-stage form that includes a generator and a classifier. However, existing generation-based methods focus on enhancing the generator's effect while neglecting the improvement of the classifier. In this paper, we first analyze of two properties of the generated pseudo unseen samples: bias and homogeneity. Then, we perform variational Bayesian inference to back-derive the evaluation metrics, which reflects the balance of the seen and unseen classes. As a consequence of our derivation, the aforementioned two properties are incorporated into the classifier training as seen-unseen priors via logit adjustment. The Zero-Shot Logit Adjustment further puts semantic-based classifiers into effect in generation-based GZSL. Our experiments demonstrate that the proposed technique achieves state-of-the-art when combined with the basic generator, and it can improve various generative Zero-Shot Learning frameworks. Our codes are available on https://github.com/cdb342/IJCAI-2022-ZLA.

Learning to hash pictures a list-wise sorting problem. Its testing metrics, e.g., mean-average precision, count on a sorted candidate list ordered by pair-wise code similarity. However, scarcely does one train a deep hashing model with the sorted results end-to-end because of the non-differentiable nature of the sorting operation. This inconsistency in the objectives of training and test may lead to sub-optimal performance since the training loss often fails to reflect the actual retrieval metric. In this paper, we tackle this problem by introducing Naturally-Sorted Hashing (NSH). We sort the Hamming distances of samples' hash codes and accordingly gather their latent representations for self-supervised training. Thanks to the recent advances in differentiable sorting approximations, the hash head receives gradients from the sorter so that the hash encoder can be optimized along with the training procedure. Additionally, we describe a novel Sorted Noise-Contrastive Estimation (SortedNCE) loss that selectively picks positive and negative samples for contrastive learning, which allows NSH to mine data semantic relations during training in an unsupervised manner. Our extensive experiments show the proposed NSH model significantly outperforms the existing unsupervised hashing methods on three benchmarked datasets.

Binary pointwise labels (aka implicit feedback) are heavily leveraged by deep learning based recommendation algorithms nowadays. In this paper we discuss the limited expressiveness of these labels may fail to accommodate varying degrees of user preference, and thus lead to conflicts during model training, which we call annotation bias. To solve this issue, we find the soft-labeling property of pairwise labels could be utilized to alleviate the bias of pointwise labels. To this end, we propose a momentum contrast framework (\method ) that combines pointwise and pairwise learning for recommendation. \method has a three-tower network structure: one user network and two item networks. The two item networks are used for computing pointwise and pairwise loss respectively. To alleviate the influence of the annotation bias, we perform a momentum update to ensure a consistent item representation. Extensive experiments on real-world datasets demonstrate the superiority of our method against state-of-the-art recommendation algorithms.

With the rapid development of convolutional neural networks (CNNs), significant progress has been achieved in semantic segmentation. Despite the great success, such deep learning approaches require large scale real-world datasets with pixel-level annotations. However, considering that pixel-level labeling of semantics is extremely laborious, many researchers turn to utilize synthetic data with free annotations. But due to the clear domain gap, the segmentation model trained with the synthetic images tends to perform poorly on the real-world datasets. Unsupervised domain adaptation (UDA) for semantic segmentation recently gains an increasing research attention, which aims at alleviating the domain discrepancy. Existing methods in this scope either simply align features or the outputs across the source and target domains or have to deal with the complex image processing and post-processing problems. In this work, we propose a novel multi-level UDA model named Confidence-and-Refinement Adaptation Model (CRAM), which contains a confidence-aware entropy alignment (CEA) module and a style feature alignment (SFA) module. Through CEA, the adaptation is done locally via adversarial learning in the output space, making the segmentation model pay attention to the high-confident predictions. Furthermore, to enhance the model transfer in the shallow feature space, the SFA module is applied to minimize the appearance gap across domains. Experiments on two challenging UDA benchmarks “GTA5-to-Cityscapes” and “SYNTHIA-to-Cityscapes” demonstrate the effectiveness of CRAM. We achieve comparable performance with the existing state-of-the-art works with advantages in simplicity and convergence speed.

Xinying Qiu, Yuan Chen, Hanwu Chen, Jian-Yun Nie, Yuming Shen, Dawei Lu. Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.

Recent advances in self-supervised learning with instance-level contrastive objectives facilitate unsupervised clustering. However, a standalone datum is not perceiving the context of the holistic cluster, and may undergo sub-optimal assignment. In this paper, we extend the mainstream contrastive learning paradigm to a cluster-level scheme, where all the data subjected to the same cluster contribute to a unified representation that encodes the context of each data group. Contrastive learning with this representation then rewards the assignment of each datum. To implement this vision, we propose twin-contrast clustering (TCC). We define a set of categorical variables as clustering assignment confidence, which links the instance-level learning track with the cluster-level one. On one hand, with the corresponding assignment variables being the weight, a weighted aggregation along the data points implements the set representation of a cluster. We further propose heuristic cluster augmentation equivalents to enable cluster-level contrastive learning. On the other hand, we derive the evidence lower-bound of the instance-level contrastive objective with the assignments. By reparametrizing the assignment variables, TCC is trained end-to-end, requiring no alternating steps. Extensive experiments show that TCC outperforms the state-of-the-art on benchmarked datasets.

Conventional unsupervised hashing methods usually take advantage of similarity graphs, which are either pre-computed in the high-dimensional space or obtained from random anchor points. On the one hand, existing methods uncouple the procedures of hash function learning and graph construction. On the other hand, graphs empirically built upon original data could introduce biased prior knowledge of data relevance, leading to sub-optimal retrieval performance. In this paper, we tackle the above problems by proposing an efficient and adaptive code-driven graph, which is updated by decoding in the context of an auto-encoder. Specifically, we introduce into our framework twin bottlenecks (i.e., latent variables) that exchange crucial information collaboratively. One bottleneck (i.e., binary codes) conveys the high-level intrinsic data structure captured by the code-driven graph to the other (i.e., continuous variables for low-level detail information), which in turn propagates the updated network feedback for the encoder to learn more discriminative binary codes. The auto-encoding learning objective literally rewards the code-driven graph to learn an optimal encoder. Moreover, the proposed model can be simply optimized by gradient descent without violating the binary constraints. Experiments on benchmarked datasets clearly show the superiority of our framework over the state-of-the-art hashing methods. Our source code can be found at https://github.com/ymcidence/TBH.

The label-free nature of unsupervised cross-modal hashing hinders models from exploiting the exact semantic data similarity. Existing research typically simulates the semantics by a heuristic geometric prior in the original feature space. However, this introduces heavy bias into the model as the original features are not fully representing the underlying multi-view data relations. To address the problem above, in this paper, we propose a novel unsupervised hashing method called Semantic-Rebased Cross-modal Hashing (SRCH). A novel ‘Set-and-Rebase’ process is defined to initialize and update the cross-modal similarity graph of training data. In particular, we set the graph according to the intra-modal feature geometric basis and then alternately rebase it to update the edges within according to the hashing results. We develop an alternating optimization routine to rebase the graph and train the hashing auto-encoders with closed-form solutions so that the overall framework is efficiently trained. Our experimental results on benchmarked datasets demonstrate the superiority of our model against state-of-the-art algorithms.

This paper proposes a validation mechanism for newly added triples in a growing knowledge graph. Given a logical theory, a knowledge graph, a text corpus, and a new triple to be validated, this mechanism computes a sorted list of explanations for the new triple to facilitate the validation of it, where an explanation, called an abductive text evidence, is a set of pairs of the form (triple, window) where appending the set of triples on the left to the knowledge graph enforces entailment of the new triple under the logical theory, while every sentence window on the right which is contained in the text corpus explains to some degree why the triple on the left is true. From the angle of practice, a special class of abductive text evidences called TEP-based abductive text evidence is proposed, which is constructed from explanation patterns seen before in the knowledge graph. Accordingly, a method for computing the complete set of TEP-based abductive text evidences is proposed. Moreover, a method for sorting abductive text evidences based on distantly supervised learning is proposed. To evaluate the proposed validation mechanism, four knowledge graphs with logical theories are constructed from the four great classical masterpieces of Chinese literature. Experimental results on these datasets demonstrate the efficiency and effectiveness of the proposed mechanism.

Instance-level retrieval has become an essential paradigm to index and retrieves images from large-scale databases. Conventional instance search requires at least an example of the query image to retrieve images that contain the same object instance. Existing semantic retrieval can only search semantically-related images, such as those sharing the same category or a set of tags, not the exact instances. Meanwhile, the unrealistic assumption is that all categories or tags are known beforehand. Training models for these semantic concepts highly rely on instance-level attributes or human captions which are expensive to acquire. Given the above challenges, this paper studies the Zero-shot Retrieval problem that aims for instance-level image search using only a few dominant attributes. The contributions are: 1) we utilise automatic word embedding to infer class-level attributes to circumvent expensive human labelling; 2) the inferred class-attributes can be extended into discriminative instance attributes through our proposed Latent Instance Attributes Discovery (LIAD) algorithm; 3) our method is not restricted to complete attribute signatures, query of dominant attributes can also be dealt with. On two benchmarks, CUB and SUN, extensive experiments demonstrate that our method can achieve promising performance for the problem. Moreover, our approach can also benefit conventional ZSL tasks.

Recent studies show that large-scale sketch-based image retrieval (SBIR) can be efficiently tackled by cross-modal binary representation learning methods, where Hamming distance matching significantly speeds up the process of similarity search. Providing training and test data subjected to a fixed set of pre-defined categories, the cutting-edge SBIR and cross-modal hashing works obtain acceptable retrieval performance. However, most of the existing methods fail when the categories of query sketches have never been seen during training. In this paper, the above problem is briefed as a novel but realistic zero-shot SBIR hashing task. We elaborate the challenges of this special task and accordingly propose a zero-shot sketch-image hashing (ZSIH) model. An end-to-end three-network architecture is built, two of which are treated as the binary encoders. The third network mitigates the sketch-image heterogeneity and enhances the semantic relations among data by utilizing the Kronecker fusion layer and graph convolution, respectively. As an important part of ZSIH, we formulate a generative hashing scheme in reconstructing semantic knowledge representations for zero-shot retrieval. To the best of our knowledge, ZSIH is the first zero-shot hashing work suitable for SBIR and cross-modal search. Comprehensive experiments are conducted on two extended datasets, i.e., Sketchy and TU-Berlin with a novel zero-shot train-test split. The proposed model remarkably outperforms related works.

Free-hand sketch-based image retrieval (SBIR) is a specific cross-view retrieval task, in which queries are abstract and ambiguous sketches while the retrieval database is formed with natural images. Work in this area mainly focuses on extracting representative and shared features for sketches and natural images. However, these can neither cope well with the geometric distortion between sketches and images nor be feasible for large-scale SBIR due to the heavy continuous-valued distance computation. In this paper, we speed up SBIR by introducing a novel binary coding method, named Deep Sketch Hashing (DSH), where a semi-heterogeneous deep architecture is proposed and incorporated into an end-to-end binary coding framework. Specifically, three convolutional neural networks are utilized to encode free-hand sketches, natural images and, especially, the auxiliary sketch-tokens which are adopted as bridges to mitigate the sketch-image geometric distortion. The learned DSH codes can effectively capture the cross-view similarities as well as the intrinsic semantic correlations between different categories. To the best of our knowledge, DSH is the first hashing work specifically designed for category-level SBIR with an end-to-end deep architecture. The proposed DSH is comprehensively evaluated on two large-scale datasets of TU-Berlin Extension and Sketchy, and the experiments consistently show DSHs superior SBIR accuracies over several state-of-the-art methods, while achieving significantly reduced retrieval time and memory footprint.

Understanding the semantic relations between vision and language data has become a research trend in artificial intelligence and robotic systems. The lack of training data is an essential issue for vision-language understanding. We address the problem of image and sentence cross-modal retrieval when paired training samples are not sufficient. Inspired by recent works in variational inference, in this paper, the autoencoding variational Bayes framework is novelly extended to a semi-supervised model for image-sentence mapping task. Our method does not require all training images and sentences to be paired. The proposed model is an end-to-end system, and consists of a two-level variational embedding structure where unpaired data are involved in the first level embedding to give support to intra-modality statistics so that the lower bound of the joint marginal likelihood of paired data embeddings can be better approximated. The proposed retrieval model is evaluated on two popular datasets, i.e. Flickr30K and Flickr8K, producing superior performances compared with related state-of-the-art methods.