Changjiang Zhou

Changjiang Zhou

Auto-generated profile, based on publicly available information

Changjiang Zhou

State Key Laboratory of AI Safety; Institute of Computing Technology, Chinese Academy of Sciences; University of Chinese Academy of Sciences, China

Research Focus & Metrics

Academic Career Length

3Years (2024-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications

25 August 2025 in ACM Transactions on Information Systems
Verified
Journal Article
Open access
JGJiafeng Guo
Jiafeng Guo
CZChangjiang Zhou
Changjiang Zhou
RZRuqing Zhang
Ruqing Zhang
JCJiangui Chen
Jiangui Chen
MRMaarten de Rijke
Maarten de Rijke
XCXueqi Cheng
Xueqi Cheng
Knowledge-intensive language tasks (KILTs) typically require retrieving relevant documents from trustworthy corpora, e.g., Wikipedia, to produce specific answers. Very recently, a pre-trained generative retrieval model for KILTs, named CorpusBrain, was proposed and reached new state-of-the-art retrieval performance. However, most research on KILTs, including CorpusBrain, has predominantly focused on a static document collection, overlooking the dynamic nature of real-world scenarios, where new documents are continuously being incorporated into the source corpus. To address this gap, it is crucial to explore the capability of retrieval models to effectively handle the dynamic retrieval scenario inherent in KILTs. In this work, we first introduce the continual document learning (CDL) task for KILTs and build a novel benchmark dataset named KILT++ based on the original KILT dataset for evaluation. Then, we conduct a comprehensive study of the use of pre-trained CorpusBrain on KILT++. Unlike the promising results in the stationary scenario, CorpusBrain is prone to catastrophic forgetting in the dynamic scenario, hence hampering retrieval performance. To alleviate this issue, we propose CorpusBrain++, a continual generative pre-training framework that enhances the original model along two key dimensions: (i) We employ a backbone-adapter architecture: the dynamic adapter is learned for each downstream KILT task via task-specific pre-training objectives; the backbone parameters that are task-shared are kept unchanged to offer foundational retrieval capacity. (ii) We use an experience replay strategy based on exemplar documents that are similar to new documents, to prevent catastrophic forgetting of old documents. Empirical results demonstrate the effectiveness and efficiency of CorpusBrain++ in comparison to both traditional and generative information retrieval methods.

12 April 2026
Verified
Conference Paper
Open access
CZChangjiang Zhou
Changjiang Zhou
RZRuqing Zhang
Ruqing Zhang
JGJiafeng Guo
Jiafeng Guo
MRMaarten de Rijke
Maarten de Rijke
YFYixing Fan
Yixing Fan
XCXueqi Cheng
Xueqi Cheng

20 September 2026 in Preprints.org
Verified
Preprint
Open access
PDF file
YZYingchen Zhang
Yingchen Zhang
CZChangjiang Zhou
Changjiang Zhou
RZRuqing Zhang
Ruqing Zhang
JGJiafeng Guo
Jiafeng Guo
XCXueqi Cheng
Xueqi Cheng
Large language model (LLM) agents can learn through interaction with external environments, obtain training signals and reusable experience from task execution. Scaling this form of learning requires a sufficient supply of interactive environments and tasks grounded in those environments, with explicit objectives and completion criteria. We refer to these complementary resources as interactive learning resources. This survey reviews their automated construction at scale and continuous adaptation through three interconnected themes: environment synthesis, task synthesis, and learning-driven evolution of both. For environment synthesis, we examine functional components, construction paradigms, and intrinsic quality assurance. For task synthesis, we review intent formation, grounding, completion verification, and task quality assurance. We then examine how a target agent's performance and learning progress inform assessments of resource suitability and guide adaptation to its evolving capabilities and learning needs across successive learning rounds. Finally, we discuss open challenges concerning the order of environment and task synthesis, the trade-off between intrinsic quality and learning utility, long-term resource evolution, and quality assurance at scale.


26 February 2024 in arXiv
Verified
Preprint
Open access
PDF file
JGJiafeng Guo
Jiafeng Guo
CZChangjiang Zhou
Changjiang Zhou
RZRuqing Zhang
Ruqing Zhang
JCJiangui Chen
Jiangui Chen
MRMaarten de Rijke
Maarten de Rijke
XCXueqi Cheng
Xueqi Cheng
Knowledge-intensive language tasks (KILTs) typically require retrieving relevant documents from trustworthy corpora, e.g., Wikipedia, to produce specific answers. Very recently, a pre-trained generative retrieval model for KILTs, named CorpusBrain, was proposed and reached new state-of-the-art retrieval performance. However, most existing research on KILTs, including CorpusBrain, has predominantly focused on a static document collection, overlooking the dynamic nature of real-world scenarios, where new documents are continuously being incorporated into the source corpus. To address this gap, it is crucial to explore the capability of retrieval models to effectively handle the dynamic retrieval scenario inherent in KILTs. In this work, we first introduce the continual document learning (CDL) task for KILTs and build a novel benchmark dataset named KILT++ based on the original KILT dataset for evaluation. Then, we conduct a comprehensive study over the use of pre-trained CorpusBrain on KILT++. Unlike the promising results in the stationary scenario, CorpusBrain is prone to catastrophic forgetting in the dynamic scenario, hence hampering the retrieval performance. To alleviate this issue, we propose CorpusBrain++, a continual generative pre-training framework. Empirical results demonstrate the significant effectiveness and remarkable efficiency of CorpusBrain++ in comparison to both traditional and generative IR methods.