Yilong Chen, Junyuan Shang, Zhenyu Zhang, Yanxi Xie, Jiawei Sheng, Tingwen Liu, Shuohuan Wang, Yu Sun, Hua Wu, Haifeng Wang. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Large language models (LLMs) with one or more fine-tuning phases have become necessary to unlock various capabilities, enabling LLMs to follow natural language instructions and align with human preferences. However, it carries the risk of catastrophic forgetting during sequential training, the parametric knowledge or the ability learned in previous stages may be overwhelmed by incoming training data. This paper finds that LLMs can restore some original knowledge by regularly resetting partial parameters. Inspired by this, we introduce Half Fine-Tuning (HFT) for LLMs, as a substitute for full fine-tuning (FFT), to mitigate the forgetting issues, where half of the parameters are selected to learn new tasks. In contrast, the other half are frozen to retain previous knowledge. We provide a feasibility analysis from the optimization perspective and interpret the parameter selection operation as a regularization term. HFT could be seamlessly integrated into existing fine-tuning frameworks without changing the model architecture. Extensive experiments and analysis on supervised fine-tuning, direct preference optimization, and continual learning consistently demonstrate the effectiveness, robustness, and efficiency of HFT. Compared with FFT, HFT not only significantly alleviates the forgetting problem, but also achieves the best performance in a series of downstream benchmarks, with an approximately 30% reduction in training time.

Mixture-of-Experts (MoE) shines brightly in large language models (LLMs) and demonstrates outstanding performance in plentiful natural language processing tasks.However, existing methods transforming LLMs from dense to MoE face significant data requirements and typically rely on large-scale post-training.In this paper, we propose Upcycling Instruction Tuning (UpIT), a data-efficient approach for tuning a dense pre-trained model into a MoE instruction model.Specifically, we first point out that intermediate checkpoints during instruction tuning of the dense model are naturally suitable for specialized experts, and then propose an expert expansion stage to flexibly achieve models with flexible numbers of experts, where genetic algorithm and parameter merging are introduced to ensure sufficient diversity of new extended experts.To ensure that each specialized expert in the MoE model works as expected, we select a small amount of seed data that each expert excels to preoptimize the router.Extensive experiments with various data scales and upcycling settings demonstrate the outstanding performance and data efficiency of UpIT, as well as stable improvement in expert or data scaling.Further analysis reveals the importance of ensuring expert diversity in upcycling.

Transformer models encounter inefficiency when scaling hidden dimensions due to the uniform expansion of parameters. When delving into the sparsity of hidden dimensions, we observe that only a small subset of dimensions are highly activated, where some dimensions are commonly activated across tokens, and some others uniquely activated for individual tokens. To leverage this, we propose MoHD (Mixture of Hidden Dimensions), a sparse architecture that combines shared sub-dimensions for common features and dynamically routes specialized sub-dimensions per token. To address the potential information loss from sparsity, we introduce activation scaling and group fusion mechanisms. MoHD efficiently expands hidden dimensions with minimal computational increases, outperforming vanilla Transformers in both parameter efficiency and task performance across 10 NLP tasks. MoHD achieves 1.7% higher performance with 50% fewer activatied parameters and 3.7% higher performance with 3$\times$ total parameters expansion at constant activated parameters cost. MoHD offers a new perspective for scaling the model, showcasing the potential of hidden dimension sparsity.

Yilong Chen, Junyuan Shang, Zhenyu Zhang, Shiyao Cui, Tingwen Liu, Shuohuan Wang, Yu Sun, Hua Wu. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Heterogeneous Graph Neural Networks (HGNNs) have gained significant popularity in various heterogeneous graph learning tasks. However, most existing HGNNs rely on spatial domain-based methods to aggregate information, i.e., manually selected meta-paths or some heuristic modules, lacking theoretical guarantees. Furthermore, these methods cannot learn arbitrary valid heterogeneous graph filters within the spectral domain, which have limited expressiveness. To tackle these issues, we present a positive spectral heterogeneous graph convolution via positive noncommutative polynomials. Then, using this convolution, we propose PSHGCN, a novel Positive Spectral Heterogeneous Graph Convolutional Network. PSHGCN offers a simple yet effective method for learning valid heterogeneous graph filters. Moreover, we demonstrate the rationale of PSHGCN in the graph optimization framework. We conducted an extensive experimental study to show that PSHGCN can learn diverse heterogeneous graph filters and outperform all baselines on open benchmarks. Notably, PSHGCN exhibits remarkable scalability, efficiently handling large real-world graphs comprising millions of nodes and edges. Our codes are available at https://github.com/ivam-he/PSHGCN.

Node classification is a substantial problem in graph-based fraud detection. Many existing works adopt Graph Neural Networks (GNNs) to enhance fraud detectors. While promising, currently most GNN-based fraud detectors fail to generalize to the low homophily setting. Besides, label utilization has been proved to be significant factor for node classification problem. But we find they are less effective in fraud detection tasks due to the low homophily in graphs. In this work, we propose GAGA, a novel Group AGgregation enhanced TrAnsformer, to tackle the above challenges. Specifically, the group aggregation provides a portable method to cope with the low homophily issue. Such an aggregation explicitly integrates the label information to generate distinguishable neighborhood information. Along with group aggregation, an attempt towards end-to-end trainable group encoding is proposed which augments the original feature space with the class labels. Meanwhile, we devise two additional learnable encodings to recognize the structural and relational context. Then, we combine the group aggregation and the learnable encodings into a Transformer encoder to capture the semantic information. Experimental results clearly show that GAGA outperforms other competitive graph-based fraud detectors by up to 24.39% on two trending public datasets and a real-world industrial dataset from Baidu. Even more, the group aggregation is demonstrated to outperform other label utilization methods (e.g., C&S, BoT/UniMP) in the low homophily setting.

Pre-trained language representation models (PLMs) such as BERT and Enhanced Representation through kNowledge IntEgration (ERNIE) have been integral to achieving recent improvements on various downstream tasks, including information retrieval. However, it is nontrivial to directly utilize these models for the large-scale web search due to the following challenging issues: (1) the prohibitively expensive computations of massive neural PLMs, especially for long texts in the web document, prohibit their deployments in the web search system that demands extremely low latency; (2) the discrepancy between existing task-agnostic pre-training objectives and the ad hoc retrieval scenarios that demand comprehensive relevance modeling is another main barrier for improving the online retrieval and ranking effectiveness; and (3) to create a significant impact on real-world applications, it also calls for practical solutions to seamlessly interweave the resultant PLM and other components into a cooperative system to serve web-scale data. Accordingly, we contribute a series of successfully applied techniques in tackling these exposed issues in this work when deploying the state-of-the-art Chinese pre-trained language model, i.e., ERNIE, in the online search engine system. We first present novel practices to perform expressive PLM-based semantic retrieval with a flexible poly-interaction scheme and cost-efficiently contextualize and rank web documents with a cheap yet powerful Pyramid-ERNIE architecture. We then endow innovative pre-training and fine-tuning paradigms to explicitly incentivize the query-document relevance modeling in PLM-based retrieval and ranking with the large-scale noisy and biased post-click behavioral data. We also introduce a series of effective strategies to seamlessly interwoven the designed PLM-based models with other conventional components into a cooperative system. Extensive offline and online experimental results show that our proposed techniques are crucial to achieving more effective search performance. We also provide a thorough analysis of our methodology and experimental results.

Relational graph neural networks have garnered particular attention to encode graph context in knowledge graphs (KGs). Although they achieved competitive performance on small KGs, how to efficiently and effectively utilize graph context for large KGs remains an open problem. To this end, we propose the Relation-based Embedding Propagation (REP) method. It is a post-processing technique to adapt pre-trained KG embeddings with graph context. As relations in KGs are directional, we model the incoming head context and the outgoing tail context separately. Accordingly, we design relational context functions with no external parameters. Besides, we use averaging to aggregate context information, making REP more computation-efficient. We theoretically prove that such designs can avoid information distortion during propagation. Extensive experiments also demonstrate that REP has significant scalability while improving or maintaining prediction quality. Particularly, it averagely brings about 10% relative improvement to triplet-based embedding methods on OGBL-WikiKG2 and takes 5%-83% time to achieve comparable results as the state-of-the-art GC-OTE.

Recent efforts of multimodal Transformers have improved Visually Rich Document Understanding (VrDU) tasks via incorporating visual and textual information. However, existing approaches mainly focus on fine-grained elements such as words and document image patches, making it hard for them to learn from coarse-grained elements, including natural lexical units like phrases and salient visual regions like prominent image regions. In this paper, we attach more importance to coarse-grained elements containing high-density information and consistent semantics, which are valuable for document understanding. At first, a document graph is proposed to model complex relationships among multi-grained multimodal elements, in which salient visual regions are detected by a cluster-based method. Then, a multi-grained multimodal Transformer called mmLayout is proposed to incorporate coarse-grained information into existing pre-trained fine-grained multimodal Transformers based on the graph. In mmLayout, coarse-grained information is aggregated from fine-grained, and then, after further processing, is fused back into fine-grained for final prediction. Furthermore, common sense enhancement is introduced to exploit the semantic information of natural lexical units. Experimental results on four tasks, including information extraction and document question answering, show that our method can improve the performance of multimodal Transformers based on fine-grained elements and achieve better performance with fewer parameters. Qualitative analyses show that our method can capture consistent semantics in coarse-grained elements.

We propose a knowledge-enhanced approach, ERNIE-ViL, which incorporates structured knowledge obtained from scene graphs to learn joint representations of vision-language. ERNIE-ViL tries to build the detailed semantic connections (objects, attributes of objects and relationships between objects) across vision and language, which are essential to vision-language cross-modal tasks. Utilizing scene graphs of visual scenes, ERNIE-ViL constructs Scene Graph Prediction tasks, i.e., Object Prediction, Attribute Prediction and Relationship Prediction tasks in the pre-training phase. Specifically, these prediction tasks are implemented by predicting nodes of different types in the scene graph parsed from the sentence. Thus, ERNIE-ViL can learn the joint representations characterizing the alignments of the detailed semantics across vision and language. After pre-training on large scale image-text aligned datasets, we validate the effectiveness of ERNIE-ViL on 5 cross-modal downstream tasks. ERNIE-ViL achieves state-of-the-art performances on all these tasks and ranks the first place on the VCR leaderboard with an absolute improvement of 3.7%.

Recent studies have demonstrated that pre-trained cross-lingual models achieve impressive performance in downstream cross-lingual tasks. This improvement benefits from learning a large amount of monolingual and parallel corpora. Although it is generally acknowledged that parallel corpora are critical for improving the model performance, existing methods are often constrained by the size of parallel corpora, especially for low-resource languages. In this paper, we propose Ernie-M, a new training method that encourages the model to align the representation of multiple languages with monolingual corpora, to overcome the constraint that the parallel corpus size places on the model performance. Our key insight is to integrate back-translation into the pre-training process. We generate pseudo-parallel sentence pairs on a monolingual corpus to enable the learning of semantic alignments between different languages, thereby enhancing the semantic modeling of cross-lingual models. Experimental results show that Ernie-M outperforms existing cross-lingual models and delivers new state-of-the-art results in various cross-lingual downstream tasks. The codes and pre-trained models will be made publicly available.

Graph neural network (GNN) and label propagation algorithm (LPA) are both message passing algorithms, which have achieved superior performance in semi-supervised classification. GNN performs feature propagation by a neural network to make predictions, while LPA uses label propagation across graph adjacency matrix to get results. However, there is still no effective way to directly combine these two kinds of algorithms. To address this issue, we propose a novel Unified Message Passaging Model (UniMP) that can incorporate feature and label propagation at both training and inference time. First, UniMP adopts a Graph Transformer network, taking feature embedding and label embedding as input information for propagation. Second, to train the network without overfitting in self-loop input label information, UniMP introduces a masked label prediction strategy, in which some percentage of input label information are masked at random, and then predicted. UniMP conceptually unifies feature propagation and label propagation and is empirically powerful. It obtains new state-of-the-art semi-supervised classification results in Open Graph Benchmark (OGB).