Jie Liu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Jie Liu has a total of 68 co-authors. Here are the top co-authors by number of shared publications:
Shaowei Chen
Nankai University
5 shared publicationsZhicheng He
Nankai University
5 shared publicationsYalou Huang
Nankai University
4 shared publications- Shaowei Chen
College of Artificial Intelligence, Nankai University, Tianjin, China
3 shared publications Yuan Wang
Tianjin University of Science and Technology
3 shared publicationsYutai Duan
College of Artificial Intelligence, Nankai University, Tianjin, China
3 shared publications
Latest Publications
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Pattern Analysis and Machine Intelligence
- Vol. 48 (10) , 1-18
Task-KV: Task-aware KV Cache Optimization via Semantic Differentiation of Attention Heads
KV cache is a widely used acceleration technique for large language models (LLMs) inference, but its memory requirement grows rapidly with input length. Previous studies have reduced the size of KV cache by either removing the same number of unimportant tokens for all attention heads or by allocating differentiated KV cache budgets for pre-identified attention heads. However, due to the importance of attention heads varies across different tasks, the pre-identified attention heads fail to adapt effectively to various downstream tasks. To address this issue, we propose Task-KV, a method that leverages the semantic differentiation of attention heads to allocate differentiated KV cache budgets across various tasks. We demonstrate that attention heads far from the semantic center (called heterogeneous heads) make a significant contribution to task outputs and semantic understanding. In contrast, other attention heads play the role of aggregating important information and focusing reasoning. Task-KV allocates full KV cache budget to heterogeneous heads to preserve comprehensive semantic information, while reserving a small number of recent tokens and attention sinks for non-heterogeneous heads. Furthermore, we innovatively introduce middle activations to preserve key contextual information aggregated from non-heterogeneous heads. To dynamically perceive semantic differences among attention heads, we design a semantic separator to distinguish heterogeneous heads from non-heterogeneous ones based on their distances from the semantic center. Experimental results on multiple benchmarks and different model architectures demonstrate that Task-KV significantly outperforms existing baseline methods. Notably, in scenarios requiring full-context processing, Task-KV achieves performance comparable to the full KV cache while utilizing only 30% of the memory.
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Pattern Analysis and Machine Intelligence
- Vol. 48 (10) , 1-18
Task-KV: Task-aware KV Cache Optimization via Semantic Differentiation of Attention Heads
KV cache is a widely used acceleration technique for large language models (LLMs) inference, but its memory requirement grows rapidly with input length. Previous studies have reduced the size of KV cache by either removing the same number of unimportant tokens for all attention heads or by allocating differentiated KV cache budgets for pre-identified attention heads. However, due to the importance of attention heads varies across different tasks, the pre-identified attention heads fail to adapt effectively to various downstream tasks. To address this issue, we propose Task-KV, a method that leverages the semantic differentiation of attention heads to allocate differentiated KV cache budgets across various tasks. We demonstrate that attention heads far from the semantic center (called heterogeneous heads) make a significant contribution to task outputs and semantic understanding. In contrast, other attention heads play the role of aggregating important information and focusing reasoning. Task-KV allocates full KV cache budget to heterogeneous heads to preserve comprehensive semantic information, while reserving a small number of recent tokens and attention sinks for non-heterogeneous heads. Furthermore, we innovatively introduce middle activations to preserve key contextual information aggregated from non-heterogeneous heads. To dynamically perceive semantic differences among attention heads, we design a semantic separator to distinguish heterogeneous heads from non-heterogeneous ones based on their distances from the semantic center. Experimental results on multiple benchmarks and different model architectures demonstrate that Task-KV significantly outperforms existing baseline methods. Notably, in scenarios requiring full-context processing, Task-KV achieves performance comparable to the full KV cache while utilizing only 30% of the memory.
- 4 October 2025
- Published by Springer Nature
- p. 108-124
Community-Aware Graph Transformer: Preserving Community Semantics for Effective Global Aggregation
- 4 October 2025
- Published by Springer Nature
- p. 108-124
Community-Aware Graph Transformer: Preserving Community Semantics for Effective Global Aggregation
- 21 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks and Learning Systems
- Vol. 36 (8) , 15294-15308
2-D Transformer: Extending Large Language Models to Long-Context With Few Memory
The ability of processing long contexts is crucial for large language models (LLMs), but training LLMs with a long-context window requires substantial computational resources. Many sought to mitigate this through the sparse attention mechanism. However, sparse attention faces a noticeable gap compared with full attention in capturing long-distance information, leading to limited long-context processing capabilities. To effectively address this issue, this article proposes a novel sparse transformer architecture called 2-D transformer (2D-former), aimed at extending the context windows of pretrained LLMs while reducing GPU memory requirements. The 2D-former incorporates a 2-D attention mechanism that consists of a long-distance information compressor (LDIC) and a blockwise attention (BA) mechanism. LDIC can self-adaptively extract blockwise representational features by convolution and compress long-distance information into a set of tokens based on the significance of each block. The BA mechanism integrates these features, enabling each token to directly communicate with any of its preceding tokens during the computation of sparse attention. In this way, sparse attention can fully utilize long-distance information to bridge the gap with full attention while greatly reducing computational requirements. The 2D-former only needs to add less than 0.14% of additional trainable parameters to extend the context length of LLaMA2 7B to 32k on 4 A100 GPUs with 40-GB memory. In addition, it is compatible with most current acceleration techniques and parameter-efficient fine-tuning (PEFT) methods. Furthermore, we conduct supervised fine-tuning with 2D-former using our self-collected long-instruction fine-tuning dataset, named LongTuning, which comprises over 11k long-context question-answer (QA) pairs. Experimental results demonstrate that 2D-former achieves efficient long-context extension with minimal GPU memory and computational time consumption, while maintaining superior performance across both downstream long-context and short-context tasks.
- 21 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks and Learning Systems
- Vol. 36 (8) , 15294-15308
2-D Transformer: Extending Large Language Models to Long-Context With Few Memory
The ability of processing long contexts is crucial for large language models (LLMs), but training LLMs with a long-context window requires substantial computational resources. Many sought to mitigate this through the sparse attention mechanism. However, sparse attention faces a noticeable gap compared with full attention in capturing long-distance information, leading to limited long-context processing capabilities. To effectively address this issue, this article proposes a novel sparse transformer architecture called 2-D transformer (2D-former), aimed at extending the context windows of pretrained LLMs while reducing GPU memory requirements. The 2D-former incorporates a 2-D attention mechanism that consists of a long-distance information compressor (LDIC) and a blockwise attention (BA) mechanism. LDIC can self-adaptively extract blockwise representational features by convolution and compress long-distance information into a set of tokens based on the significance of each block. The BA mechanism integrates these features, enabling each token to directly communicate with any of its preceding tokens during the computation of sparse attention. In this way, sparse attention can fully utilize long-distance information to bridge the gap with full attention while greatly reducing computational requirements. The 2D-former only needs to add less than 0.14% of additional trainable parameters to extend the context length of LLaMA2 7B to 32k on 4 A100 GPUs with 40-GB memory. In addition, it is compatible with most current acceleration techniques and parameter-efficient fine-tuning (PEFT) methods. Furthermore, we conduct supervised fine-tuning with 2D-former using our self-collected long-instruction fine-tuning dataset, named LongTuning, which comprises over 11k long-context question-answer (QA) pairs. Experimental results demonstrate that 2D-former achieves efficient long-context extension with minimal GPU memory and computational time consumption, while maintaining superior performance across both downstream long-context and short-context tasks.
- 27 January 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Knowledge and Data Engineering
- Vol. 37 (3) , 1354-1365
Network-to-Network: Self-Supervised Network Representation Learning via Position Prediction
Network Representation Learning (NRL) has achieved remarkable success in learning low-dimensional representations for network nodes. However, most NRL methods, including Graph Neural Networks (GNNs) and their variants, face critical challenges. First, labeled network data, which are required for training most GNNs, are expensive to obtain. Second, existing methods are sub-optimal in preserving comprehensive topological information, including structural and positional information. Finally, most GNN approaches ignore the rich node content information. To address these challenges, we propose a self-supervised Network-to-Network framework (Net2Net) to learn semantically meaningful node representations. Our framework employs a pretext task of node position prediction (PosPredict) to effectively fuse the topological and content knowledge into low-dimensional embeddings for every node in a semi-supervised manner. Specifically, we regard a network as node content and position networks, where Net2Net aims to learn the mapping between them. We utilize a multi-layer recursively composable encoder to integrate the content and topological knowledge into the egocentric network node embeddings. Furthermore, we design a cross-modal decoder to map the egocentric node embeddings into their node position identities (PosIDs) in the node position network. Extensive experiments on eight diverse networks demonstrate the superiority of Net2Net over comparable methods.
- 27 January 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Knowledge and Data Engineering
- Vol. 37 (3) , 1354-1365
Network-to-Network: Self-Supervised Network Representation Learning via Position Prediction
Network Representation Learning (NRL) has achieved remarkable success in learning low-dimensional representations for network nodes. However, most NRL methods, including Graph Neural Networks (GNNs) and their variants, face critical challenges. First, labeled network data, which are required for training most GNNs, are expensive to obtain. Second, existing methods are sub-optimal in preserving comprehensive topological information, including structural and positional information. Finally, most GNN approaches ignore the rich node content information. To address these challenges, we propose a self-supervised Network-to-Network framework (Net2Net) to learn semantically meaningful node representations. Our framework employs a pretext task of node position prediction (PosPredict) to effectively fuse the topological and content knowledge into low-dimensional embeddings for every node in a semi-supervised manner. Specifically, we regard a network as node content and position networks, where Net2Net aims to learn the mapping between them. We utilize a multi-layer recursively composable encoder to integrate the content and topological knowledge into the egocentric network node embeddings. Furthermore, we design a cross-modal decoder to map the egocentric node embeddings into their node position identities (PosIDs) in the node position network. Extensive experiments on eight diverse networks demonstrate the superiority of Net2Net over comparable methods.
- Open Access
Task-KV: Task-aware KV Cache Optimization via Semantic Differentiation of Attention Heads
KV cache is a widely used acceleration technique for large language models (LLMs) inference. However, its memory requirement grows rapidly with input length. Previous studies have reduced the size of KV cache by either removing the same number of unimportant tokens for all attention heads or by allocating differentiated KV cache budgets for pre-identified attention heads. However, due to the importance of attention heads varies across different tasks, the pre-identified attention heads fail to adapt effectively to various downstream tasks. To address this issue, we propose Task-KV, a method that leverages the semantic differentiation of attention heads to allocate differentiated KV cache budgets across various tasks. We demonstrate that attention heads far from the semantic center (called heterogeneous heads) make an significant contribution to task outputs and semantic understanding. In contrast, other attention heads play the role of aggregating important information and focusing reasoning. Task-KV allocates full KV cache budget to heterogeneous heads to preserve comprehensive semantic information, while reserving a small number of recent tokens and attention sinks for non-heterogeneous heads. Furthermore, we innovatively introduce middle activations to preserve key contextual information aggregated from non-heterogeneous heads. To dynamically perceive semantic differences among attention heads, we design a semantic separator to distinguish heterogeneous heads from non-heterogeneous ones based on their distances from the semantic center. Experimental results on multiple benchmarks and different model architectures demonstrate that Task-KV significantly outperforms existing baseline methods.
Open AccessTask-KV: Task-aware KV Cache Optimization via Semantic Differentiation of Attention Heads
KV cache is a widely used acceleration technique for large language models (LLMs) inference. However, its memory requirement grows rapidly with input length. Previous studies have reduced the size of KV cache by either removing the same number of unimportant tokens for all attention heads or by allocating differentiated KV cache budgets for pre-identified attention heads. However, due to the importance of attention heads varies across different tasks, the pre-identified attention heads fail to adapt effectively to various downstream tasks. To address this issue, we propose Task-KV, a method that leverages the semantic differentiation of attention heads to allocate differentiated KV cache budgets across various tasks. We demonstrate that attention heads far from the semantic center (called heterogeneous heads) make an significant contribution to task outputs and semantic understanding. In contrast, other attention heads play the role of aggregating important information and focusing reasoning. Task-KV allocates full KV cache budget to heterogeneous heads to preserve comprehensive semantic information, while reserving a small number of recent tokens and attention sinks for non-heterogeneous heads. Furthermore, we innovatively introduce middle activations to preserve key contextual information aggregated from non-heterogeneous heads. To dynamically perceive semantic differences among attention heads, we design a semantic separator to distinguish heterogeneous heads from non-heterogeneous ones based on their distances from the semantic center. Experimental results on multiple benchmarks and different model architectures demonstrate that Task-KV significantly outperforms existing baseline methods.
- 17 January 2025
- Published by Springer Nature in World Wide Web
- Vol. 28 (1) , 1-20
Sentence-graph-level knowledge injection with multi-task learning
- 17 January 2025
- Published by Springer Nature in World Wide Web
- Vol. 28 (1) , 1-20
Sentence-graph-level knowledge injection with multi-task learning
- 1 November 2024
- Published by Springer Nature
- p. 449-461
Finetuning LLMs for Text-to-SQL with Two-Stage Progressive Learning
- 1 November 2024
- Published by Springer Nature
- p. 449-461
Finetuning LLMs for Text-to-SQL with Two-Stage Progressive Learning
- 1 November 2024
- Published by Springer Nature
- p. 462-473
Retrieval Augmented Tree of Thoughts
- Open Access
Learning Attentional Mixture of LoRAs for Language Model Continual Learning
Fine-tuning large language models (LLMs) with Low-Rank adaption (LoRA) is widely acknowledged as an effective approach for continual learning for new tasks. However, it often suffers from catastrophic forgetting when dealing with multiple tasks sequentially. To this end, we propose Attentional Mixture of LoRAs (AM-LoRA), a continual learning approach tailored for LLMs. Specifically, AM-LoRA learns a sequence of LoRAs for a series of tasks to continually learn knowledge from different tasks. The key of our approach is that we devise an attention mechanism as a knowledge mixture module to adaptively integrate information from each LoRA. With the attention mechanism, AM-LoRA can efficiently leverage the distinctive contributions of each LoRA, while mitigating the risk of mutually negative interactions among them that may lead to catastrophic forgetting. Moreover, we further introduce $L1$ norm in the learning process to make the attention vector more sparse. The sparse constraints can enable the model to lean towards selecting a few highly relevant LoRAs, rather than aggregating and weighting all LoRAs collectively, which can further reduce the impact stemming from mutual interference. Experimental results on continual learning benchmarks indicate the superiority of our proposed method.Open AccessLearning Attentional Mixture of LoRAs for Language Model Continual Learning
Fine-tuning large language models (LLMs) with Low-Rank adaption (LoRA) is widely acknowledged as an effective approach for continual learning for new tasks. However, it often suffers from catastrophic forgetting when dealing with multiple tasks sequentially. To this end, we propose Attentional Mixture of LoRAs (AM-LoRA), a continual learning approach tailored for LLMs. Specifically, AM-LoRA learns a sequence of LoRAs for a series of tasks to continually learn knowledge from different tasks. The key of our approach is that we devise an attention mechanism as a knowledge mixture module to adaptively integrate information from each LoRA. With the attention mechanism, AM-LoRA can efficiently leverage the distinctive contributions of each LoRA, while mitigating the risk of mutually negative interactions among them that may lead to catastrophic forgetting. Moreover, we further introduce $L1$ norm in the learning process to make the attention vector more sparse. The sparse constraints can enable the model to lean towards selecting a few highly relevant LoRAs, rather than aggregating and weighting all LoRAs collectively, which can further reduce the impact stemming from mutual interference. Experimental results on continual learning benchmarks indicate the superiority of our proposed method. - 3 May 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Signal Processing Letters
- Vol. 31 (10709908) , 1525-1529
Type-Specific Modality Alignment for Multi-Modal Information Extraction
Multi-modal information extraction aims to identify structured information, such as entities or relations between entities, from text with the help of visual clues. Although existing studies have achieved great progress, they mainly focused on modality interactions in the global space while neglecting fine-grained modality alignment under the semantic subspace specific to each entity type or relation type. To solve this problem, we propose a multi-space modality alignment method (MSMA) in this letter. The core of our model is a typespecific modality interaction module (TMI), which constructs a unique semantic subspace for each entity/relation type and independently performs type-specific modality alignments under each subspace. To enable mutual promotion between different types, a global modality integration module (GMI) is designed to learn the associations between different subspaces. Furthermore, we execute these two modules iteratively for high-level semantic fusion. Extensive experiments on three benchmark datasets show that our model significantly outperforms advanced methods.
- 3 May 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Signal Processing Letters
- Vol. 31 (10709908) , 1525-1529
Type-Specific Modality Alignment for Multi-Modal Information Extraction
Multi-modal information extraction aims to identify structured information, such as entities or relations between entities, from text with the help of visual clues. Although existing studies have achieved great progress, they mainly focused on modality interactions in the global space while neglecting fine-grained modality alignment under the semantic subspace specific to each entity type or relation type. To solve this problem, we propose a multi-space modality alignment method (MSMA) in this letter. The core of our model is a typespecific modality interaction module (TMI), which constructs a unique semantic subspace for each entity/relation type and independently performs type-specific modality alignments under each subspace. To enable mutual promotion between different types, a global modality integration module (GMI) is designed to learn the associations between different subspaces. Furthermore, we execute these two modules iteratively for high-level semantic fusion. Extensive experiments on three benchmark datasets show that our model significantly outperforms advanced methods.