This technical report briefly describes our JDExplore d-team's Vega v2 submission on the SuperGLUE leaderboard. SuperGLUE is more challenging than the widely used general language understanding evaluation (GLUE) benchmark, containing eight difficult language understanding tasks, including question answering, natural language inference, word sense disambiguation, coreference resolution, and reasoning. [Method] Instead of arbitrarily increasing the size of a pretrained language model (PLM), our aim is to 1) fully extract knowledge from the input pretraining data given a certain parameter budget, e.g., 6B, and 2) effectively transfer this knowledge to downstream tasks. To achieve goal 1), we propose self-evolution learning for PLMs to wisely predict the informative tokens that should be masked, and supervise the masked language modeling (MLM) process with rectified smooth labels. For goal 2), we leverage the prompt transfer technique to improve the low-resource tasks by transferring the knowledge from the foundation model and related downstream tasks to the target task. [Results] According to our submission record (Oct. 2022), with our optimized pretraining and fine-tuning strategies, our 6B Vega method achieved new state-of-the-art performance on 4/8 tasks, sitting atop the SuperGLUE leaderboard on Oct. 8, 2022, with an average score of 91.3.
Cross-lingual named entity recognition (NER) suffers from data scarcity in the target languages, especially under zero-shot settings. Existing translate-train or knowledge distillation methods attempt to bridge the language gap, but often introduce a high level of noise. To solve this problem, consistency training methods regularize the model to be robust towards perturbations on data or hidden states. However, such methods are likely to violate the consistency hypothesis, or mainly focus on coarse-grain consistency. We propose ConNER as a novel consistency training framework for cross-lingual NER, which comprises of: (1) translation-based consistency training on unlabeled target-language data, and (2) dropoutbased consistency training on labeled source-language data. ConNER effectively leverages unlabeled target-language data and alleviates overfitting on the source language to enhance the cross-lingual adaptability. Experimental results show our ConNER achieves consistent improvement over various baseline methods.
When users move in a physical space (e.g., an urban space), they would have some records called mobility records (e.g., trajectories) generated by devices such as mobile phones and GPS devices. Naturally, mobility records capture essential information of how users work, live and entertain in their daily lives, and therefore, they have been used in a wide range of tasks such as user profile inference, mobility prediction and traffic management. In this paper, we expand this line of research by investigating the problem of inferring user socioeconomic statuses (such as prices of users' living houses as a proxy of users' socioeconomic statuses) based on their mobility records, which can potentially be used in real-life applications such as the car loan business. For this task, we propose a socioeconomic-aware deep model called DeepSEI. The DeepSEI model incorporates two networks called deep network and recurrent network, which extract the features of the mobility records from three aspects, namely spatiality, temporality and activity, one at a coarse level and the other at a detailed level. We conduct extensive experiments on real mobility records data, POI data and house prices data. The results verify that the DeepSEI model achieves superior performance than existing studies. All datasets used in this paper will be made publicly available.
In this paper we present a novel multi-attribute face manipulation method based on textual descriptions. Previous text-based image editing methods either require test-time optimization for each individual image or are restricted to single attribute editing. Extending these methods to multi-attribute face image editing scenarios will introduce undesired excessive attribute change, e.g., text-relevant attributes are overly manipulated and text-irrelevant attributes are also changed. In order to address these challenges and achieve natural editing over multiple face attributes, we propose a new decoupling training scheme where we use group sampling to get text segments from same attribute categories, instead of whole complex sentences. Further, to preserve other existing face attributes, we encourage the model to edit the latent code of each attribute separately via a entropy constraint. During the inference phase, our model is able to edit new face images without any test-time optimization, even from complex textual prompts. We show extensive experiments and analysis to demonstrate the efficacy of our method, which generates natural manipulated faces with minimal text-irrelevant attribute editing. Code and pre-trained model will be released.
Person re-identification (re-ID) via 3D skeletons is an important emerging topic with many merits. Existing solutions rarely explore valuable body-component relations in skeletal structure or motion, and they typically lack the ability to learn general representations with unlabeled skeleton data for person re-ID. This paper proposes a generic unsupervised Skeleton Prototype Contrastive learning paradigm with Multi-level Graph Relation learning (SPC-MGR) to learn effective representations from unlabeled skeletons to perform person re-ID. Specifically, we first construct unified multi-level skeleton graphs to fully model body structure within skeletons. Then we propose a multi-head structural relation layer to comprehensively capture relations of physically-connected body-component nodes in graphs. A full-level collaborative relation layer is exploited to infer collaboration between motion-related body parts at various levels, so as to capture rich body features and recognizable walking patterns. Lastly, we propose a skeleton prototype contrastive learning scheme that clusters feature-correlative instances of unlabeled graph representations and contrasts their inherent similarity with representative skeleton features ("skeleton prototypes") to learn discriminative skeleton representations for person re-ID. Empirical evaluations show that SPC-MGR significantly outperforms several state-of-the-art skeleton-based methods, and it also achieves highly competitive person re-ID performance for more general scenarios.
Semantic communication technologies enable wireless edge devices to communicate effectively by transmitting semantic meaning of data. Edge components, such as vehicles in next-generation intelligent transport systems, use well-trained semantic models to encode and decode semantic information extracted from raw and sensor data. However, the limitation in computing resources makes it difficult to support the training process of accurate semantic models on edge devices. As such, edge devices can buy the pretrained semantic models from semantic model providers, which is called "semantic model trading". Upon collecting semantic information with the semantic models, the edge devices can then sell the extracted semantic information, e.g., information about urban road conditions or traffic signs, to the interested buyers for profit, which is called "semantic information trading". To facilitate both types of the trades, effective incentive mechanisms should be designed. Thus, in this paper, we propose a hierarchical trading system to support both semantic model trading and semantic information trading jointly. The proposed incentive mechanism helps to maximize the revenue of semantic model providers in the semantic model trading, and effectively incentivizes model providers to participate in the development of semantic communication systems. For semantic information trading, our designed auction approach can support the trading between multiple semantic information sellers and buyers, while ensuring individual rationality, incentive compatibility, and budget balance, and moreover, allowing them achieve higher utilities than the baseline method.
This paper investigates an open research problem of generating text-image pairs to improve the training of fine-grained image-to-text cross-modal retrieval task, and proposes a novel framework for paired data augmentation by uncovering the hidden semantic information of StyleGAN2 model. Specifically, we first train a StyleGAN2 model on the given dataset. We then project the real images back to the latent space of StyleGAN2 to obtain the latent codes. To make the generated images manipulatable, we further introduce a latent space alignment module to learn the alignment between StyleGAN2 latent codes and the corresponding textual caption features. When we do online paired data augmentation, we first generate augmented text through random token replacement, then pass the augmented text into the latent space alignment module to output the latent codes, which are finally fed to StyleGAN2 to generate the augmented images. We evaluate the efficacy of our augmented data approach on two public cross-modal retrieval datasets, in which the promising experimental results demonstrate the augmented text-image pair data can be trained together with the original data to boost the image-to-text cross-modal retrieval performance.
In this paper, we investigate an open research task of generating 3D cartoon face shapes from single 2D GAN generated human faces and without 3D supervision, where we can also manipulate the facial expressions of the 3D shapes. To this end, we discover the semantic meanings of StyleGAN latent space, such that we are able to produce face images of various expressions, poses, and lighting by controlling the latent codes. Specifically, we first finetune the pretrained StyleGAN face model on the cartoon datasets. By feeding the same latent codes to face and cartoon generation models, we aim to realize the translation from 2D human face images to cartoon styled avatars. We then discover semantic directions of the GAN latent space, in an attempt to change the facial expressions while preserving the original identity. As we do not have any 3D annotations for cartoon faces, we manipulate the latent codes to generate images with different poses and lighting, such that we can reconstruct the 3D cartoon face shapes. We validate the efficacy of our method on three cartoon datasets qualitatively and quantitatively.
Recommendation models utilizing Graph Convolutional Networks (GCNs) have achieved state-of-the-art performance, as they can integrate both the node information and the topological structure of the user-item interaction graph. However, these GCN-based recommendation models not only suffer from over-smoothing when stacking too many layers but also bear performance degeneration resulting from the existence of noise in user-item interactions. In this paper, we first identify a recommendation dilemma of over-smoothing and solution collapsing in current GCN-based models. Specifically, these models usually aggregate all layer embeddings for node updating and achieve their best recommendation performance within a few layers because of over-smoothing. Conversely, if we place learnable weights on layer embeddings for node updating, the weight space will always collapse to a fixed point, at which the weighting of the ego layer almost holds all. We propose a layer-refined GCN model, dubbed LayerGCN, that refines layer representations during information propagation and node updating of GCN. Moreover, previous GCN-based recommendation models aggregate all incoming information from neighbors without distinguishing the noise nodes, which deteriorates the recommendation performance. Our model further prunes the edges of the user-item interaction graph following a degree-sensitive probability instead of the uniform distribution. Experimental results show that the proposed model outperforms the state-of-the-art models significantly on four public datasets with fast training convergence. The implementation code of the proposed method is available at https://github.com/enoche/ImRec.
This paper studies the multi-modal recommendation problem, where the item multi-modality information (eg. images and textual descriptions) is exploited to improve the recommendation accuracy. Besides the user-item interaction graph, existing state-of-the-art methods usually use auxiliary graphs (eg. user-user or item-item relation graph) to augment the learned representations of users and/or items. These representations are often propagated and aggregated on auxiliary graphs using graph convolutional networks, which can be prohibitively expensive in computation and memory, especially for large graphs. Moreover, existing multi-modal recommendation methods usually leverage randomly sampled negative examples in Bayesian Personalized Ranking (BPR) loss to guide the learning of user/item representations, which increases the computational cost on large graphs and may also bring noisy supervision signals into the training process. To tackle the above issues, we propose a novel self-supervised multi-modal recommendation model, dubbed BM3, which requires neither augmentations from auxiliary graphs nor negative samples. Specifically, BM3 first bootstraps latent contrastive views from the representations of users and items with a simple dropout augmentation. It then jointly optimizes three multi-modal objectives to learn the representations of users and items by reconstructing the user-item interaction graph and aligning modality features under both inter- and intra-modality perspectives. BM3 alleviates both the need for contrasting with negative examples and the complex graph augmentation from an additional target network for contrastive view generation. We show BM3 outperforms prior recommendation models on three datasets with number of nodes ranging from 20K to 200K, while achieving a 2-9X reduction in training time. Our code is available at https://github.com/enoche/BM3.