Ziqiang Cao
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Ziqiang Cao has a total of 111 co-authors. Here are the top co-authors by number of shared publications:
- Wenjie Li
The Hong Kong Polytechnic University, Hong Kong, China
7 shared publications - Min Cao
Soochow University, Suzhou, China
6 shared publications - Guohong Fu
Institute of Artificial Intelligence, School of Computer Science and Technology, Soochow University, Suzhou, China
5 shared publications Xu Yan
Institute of Artificial Intelligence, Soochow University
5 shared publicationsJuntao Li
Institute of Artificial Intelligence, Soochow University
4 shared publicationsLei Geng
School of Computer Science and Technology School of Computer Science and Technology School of Computer Science and Technology School of Computer Science and Technology Institution of Artificial Intelligence Soochow University Harbin Institute of Technology (Shenzhen Soochow University Soochow University Kong Polytechnic University
4 shared publications
Latest Publications
- 19 September 2026
- Published by American Chemical Society (ACS) in The Journal of Physical Chemistry C
- Vol. 130 (39) , 13591-13603
Data-Driven and Quantum-Chemical Insights into Emergence of Multiresonance Effect in Thermally Activated Delayed Fluorescence Emitters
Multiresonance thermally activated delayed fluorescence (MR-TADF) emitters enable narrow-bandwidth emission and high efficiency in organic light-emitting diodes; however, the fundamental origin of the multiresonance effect remains incompletely understood. Herein, a fine-tuned Uni-Mol model trained on reported MR-TADF emitters identifies BN-embedded aromatic units as the primary structural motif governing small singlet–triplet energy gaps. Wave function-based quantum chemical calculations subsequently validate this prediction and provide a mechanistic understanding of the underlying electronic structure. While the minimal BN-containing unit dictates the singlet–triplet splitting, π-extension primarily controls vibronic stabilization and emission bandwidth. The multiresonance effect is shown to arise from a cooperative interplay of phase-separated orbital topology, suppressed exchange interactions and enhanced excited-state correlation, further reinforced by heteroatom-induced electronic polarization that yields nearly identical S1 and T1 charge distributions. Combined with the data-driven identification of boron positioning as the dominant modulator of singlet–triplet splitting and nitrogen−π coupling as a key determinant of emission color, this work provides actionable design principles for next-generation MR-TADF emitters with controlled spectral purity and near-zero singlet–triplet gaps.
- 19 September 2026
- Published by American Chemical Society (ACS) in The Journal of Physical Chemistry C
- Vol. 130 (39) , 13591-13603
Data-Driven and Quantum-Chemical Insights into Emergence of Multiresonance Effect in Thermally Activated Delayed Fluorescence Emitters
Multiresonance thermally activated delayed fluorescence (MR-TADF) emitters enable narrow-bandwidth emission and high efficiency in organic light-emitting diodes; however, the fundamental origin of the multiresonance effect remains incompletely understood. Herein, a fine-tuned Uni-Mol model trained on reported MR-TADF emitters identifies BN-embedded aromatic units as the primary structural motif governing small singlet–triplet energy gaps. Wave function-based quantum chemical calculations subsequently validate this prediction and provide a mechanistic understanding of the underlying electronic structure. While the minimal BN-containing unit dictates the singlet–triplet splitting, π-extension primarily controls vibronic stabilization and emission bandwidth. The multiresonance effect is shown to arise from a cooperative interplay of phase-separated orbital topology, suppressed exchange interactions and enhanced excited-state correlation, further reinforced by heteroatom-induced electronic polarization that yields nearly identical S1 and T1 charge distributions. Combined with the data-driven identification of boron positioning as the dominant modulator of singlet–triplet splitting and nitrogen−π coupling as a key determinant of emission color, this work provides actionable design principles for next-generation MR-TADF emitters with controlled spectral purity and near-zero singlet–triplet gaps.
- 19 June 2026
- Published by Elsevier in Information Processing & Management
- Vol. 64 (1)
VotingBench : Evaluating generative capabilities of Large Language Models via voting games
- 19 June 2026
- Published by Elsevier in Information Processing & Management
- Vol. 64 (1)
VotingBench : Evaluating generative capabilities of Large Language Models via voting games
- 25 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Industrial Informatics
- Vol. 22 (6) , 4765-4776
Optimizing Prompts With MLLM for Multimodal Product Style Recognition in Industry 5.0
The transition to Industry 5.0, supported by the high-bandwidth and low-latency capabilities of 6G networks, accelerates the demand for personalized products, making efficient and accurate product style recognition a critical task. However, traditional recognition models, which rely heavily on large-labeled datasets, struggle with generalization and efficiency in such data-rich environments. In contrast, multimodal large language models (MLLMs), such as Qwen-VL, offer new possibilities by integrating visual and textual inputs, requiring less annotated data and demonstrating strong generalization. Nevertheless, most existing MLLMs are general-purpose and lack optimized prompts tailored for domain-specific tasks, limiting their recognition performance. To address this, we propose a prompt optimization framework that combines genetic algorithms (GA) and chain of thought (CoT) reasoning. GA is used to iteratively refine the instruction component of prompts, while CoT encourages step-by-step reasoning, improving the model’s interpretability and recognition accuracy. Validated on a custom-built apparel dataset and the public FashionStyle14 dataset, our method enables Qwen-VL to achieve significant accuracy improvements of 8.7% and 19.6%, respectively, over the baseline. This study makes three contributions: a framework to adapt general MLLMs for fine-grained style recognition; a demonstration of 6G-enabled real-time artificial intelligence reasoning loops; and a high-quality, multidimensional style dataset for future research. This demonstrates the effectiveness and stability of the proposed framework in handling fine-grained style recognition tasks, contributing to the development of intelligent design systems for Industry 5.0.
- 25 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Industrial Informatics
- Vol. 22 (6) , 4765-4776
Optimizing Prompts With MLLM for Multimodal Product Style Recognition in Industry 5.0
The transition to Industry 5.0, supported by the high-bandwidth and low-latency capabilities of 6G networks, accelerates the demand for personalized products, making efficient and accurate product style recognition a critical task. However, traditional recognition models, which rely heavily on large-labeled datasets, struggle with generalization and efficiency in such data-rich environments. In contrast, multimodal large language models (MLLMs), such as Qwen-VL, offer new possibilities by integrating visual and textual inputs, requiring less annotated data and demonstrating strong generalization. Nevertheless, most existing MLLMs are general-purpose and lack optimized prompts tailored for domain-specific tasks, limiting their recognition performance. To address this, we propose a prompt optimization framework that combines genetic algorithms (GA) and chain of thought (CoT) reasoning. GA is used to iteratively refine the instruction component of prompts, while CoT encourages step-by-step reasoning, improving the model’s interpretability and recognition accuracy. Validated on a custom-built apparel dataset and the public FashionStyle14 dataset, our method enables Qwen-VL to achieve significant accuracy improvements of 8.7% and 19.6%, respectively, over the baseline. This study makes three contributions: a framework to adapt general MLLMs for fine-grained style recognition; a demonstration of 6G-enabled real-time artificial intelligence reasoning loops; and a high-quality, multidimensional style dataset for future research. This demonstrates the effectiveness and stability of the proposed framework in handling fine-grained style recognition tasks, contributing to the development of intelligent design systems for Industry 5.0.
- 6 February 2026
- Published by Springer Nature
- p. 457-470
M$^3$ITR: Modeling Many-to-Many Relationships for Robust Image-Text Retrieval- 6 February 2026
- Published by Springer Nature
- p. 457-470
M$^3$ITR: Modeling Many-to-Many Relationships for Robust Image-Text Retrieval- 7 November 2025
- Published by Elsevier in Information Processing & Management
- Vol. 63 (2)
Bidirectional GPT
- 7 November 2025
- Published by Elsevier in Information Processing & Management
- Vol. 63 (2)
Bidirectional GPT
- 6 September 2025
- Published by Elsevier in Journal of Chromatography A
- Vol. 1762, 466362
Open-source liquid chromatography-mass spectrometry data processing tools for RNA modification analysis
- 6 September 2025
- Published by Elsevier in Journal of Chromatography A
- Vol. 1762, 466362
Open-source liquid chromatography-mass spectrometry data processing tools for RNA modification analysis
- Open Access
Adaptive Sparse Softmax: An Effective and Efficient Softmax Variant
Softmax with the cross entropy loss is the standard configuration for current neural classification models. The gold score for a target class is supposed to be 1, but it is never reachable under the softmax schema. Such a problem makes the training process continue forever and leads to overfitting. Moreover, the "target-approach-1" training goal forces the model to continuously learn all samples, leading to a waste of time in handling some samples which have already been classified correctly with high confidence, while the test goal simply requires the target class of each sample to hold the maximum score. To solve the above weaknesses, we propose the Adaptive Sparse softmax (AS-Softmax) which designs a reasonable and test-matching transformation on top of softmax. For more purposeful learning, we discard the classes with far smaller scores compared with the actual class during training. Then the model could focus on learning to distinguish the target class from its strong opponents, which is also the great challenge in test. In addition, since the training losses of easy samples will gradually drop to 0 in AS-Softmax, we develop an adaptive gradient accumulation strategy based on the masked sample ratio to speed up training. We verify the proposed AS-Softmax on a variety of text multi-class, text multi-label, text token classification, image classification and audio classification tasks with class sizes ranging from 5 to 5000+. The results show that AS-Softmax consistently outperforms softmax and its variants, and the loss of AS-Softmax is remarkably correlated with classification performance in validation. Furthermore, adaptive gradient accumulation strategy can bring about 1.2x training speedup comparing with the standard softmax while maintaining classification effectiveness.
Open AccessAdaptive Sparse Softmax: An Effective and Efficient Softmax Variant
Softmax with the cross entropy loss is the standard configuration for current neural classification models. The gold score for a target class is supposed to be 1, but it is never reachable under the softmax schema. Such a problem makes the training process continue forever and leads to overfitting. Moreover, the "target-approach-1" training goal forces the model to continuously learn all samples, leading to a waste of time in handling some samples which have already been classified correctly with high confidence, while the test goal simply requires the target class of each sample to hold the maximum score. To solve the above weaknesses, we propose the Adaptive Sparse softmax (AS-Softmax) which designs a reasonable and test-matching transformation on top of softmax. For more purposeful learning, we discard the classes with far smaller scores compared with the actual class during training. Then the model could focus on learning to distinguish the target class from its strong opponents, which is also the great challenge in test. In addition, since the training losses of easy samples will gradually drop to 0 in AS-Softmax, we develop an adaptive gradient accumulation strategy based on the masked sample ratio to speed up training. We verify the proposed AS-Softmax on a variety of text multi-class, text multi-label, text token classification, image classification and audio classification tasks with class sizes ranging from 5 to 5000+. The results show that AS-Softmax consistently outperforms softmax and its variants, and the loss of AS-Softmax is remarkably correlated with classification performance in validation. Furthermore, adaptive gradient accumulation strategy can bring about 1.2x training speedup comparing with the standard softmax while maintaining classification effectiveness.
- 29 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Audio, Speech and Language Processing
- Vol. 33, 3148-3159
Adaptive Sparse Softmax: An Effective and Efficient Softmax Variant
Softmax with the cross entropy loss is the standard configuration for current neural classification models. The gold score for a target class is supposed to be 1, but it is never reachable under the softmax schema. Such a problem makes the training process continue forever and leads to overfitting. Moreover, the “target-approach-1” training goal forces the model to continuously learn all samples, leading to a waste of time in handling some samples which have already been classified correctly with high confidence, while the test goal simply requires the target class of each sample to hold the maximum score. To solve the above weaknesses, we propose the Adaptive Sparse softmax (AS-Softmax) which designs a reasonable and test-matching transformation on top of softmax. For more purposeful learning, we discard the classes with far smaller scores compared with the actual class during training. Then the model could focus on learning to distinguish the target class from its strong opponents, which is also the great challenge in test. In addition, since the training losses of easy samples will gradually drop to 0 in AS-Softmax, we develop an adaptive gradient accumulation strategy based on the masked sample ratio to speed up training. We verify the proposed AS-Softmax on a variety of text multi-class, text multi-label, text token classification, image classification and audio classification tasks with class sizes ranging from 5 to 5000+. The results show that AS-Softmax consistently outperforms softmax and its variants, and the loss of AS-Softmax is remarkably correlated with classification performance in validation. Furthermore, adaptive gradient accumulation strategy can bring about 1.2× training speedup comparing with the standard softmax while maintaining classification effectiveness.
- 29 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Audio, Speech and Language Processing
- Vol. 33, 3148-3159
Adaptive Sparse Softmax: An Effective and Efficient Softmax Variant
Softmax with the cross entropy loss is the standard configuration for current neural classification models. The gold score for a target class is supposed to be 1, but it is never reachable under the softmax schema. Such a problem makes the training process continue forever and leads to overfitting. Moreover, the “target-approach-1” training goal forces the model to continuously learn all samples, leading to a waste of time in handling some samples which have already been classified correctly with high confidence, while the test goal simply requires the target class of each sample to hold the maximum score. To solve the above weaknesses, we propose the Adaptive Sparse softmax (AS-Softmax) which designs a reasonable and test-matching transformation on top of softmax. For more purposeful learning, we discard the classes with far smaller scores compared with the actual class during training. Then the model could focus on learning to distinguish the target class from its strong opponents, which is also the great challenge in test. In addition, since the training losses of easy samples will gradually drop to 0 in AS-Softmax, we develop an adaptive gradient accumulation strategy based on the masked sample ratio to speed up training. We verify the proposed AS-Softmax on a variety of text multi-class, text multi-label, text token classification, image classification and audio classification tasks with class sizes ranging from 5 to 5000+. The results show that AS-Softmax consistently outperforms softmax and its variants, and the loss of AS-Softmax is remarkably correlated with classification performance in validation. Furthermore, adaptive gradient accumulation strategy can bring about 1.2× training speedup comparing with the standard softmax while maintaining classification effectiveness.
- Open Access
- 28 July 2025
- Published by Oxford University Press (OUP) in Bioinformatics
- Vol. 41 (8)
Prot2Chat: Protein LLM with Early-Fusion of Text, Sequence and Structure
Motivation: Proteins are of great significance in living organisms. However, understanding their functions encounters numerous challenges, such as insufficient integration of multimodal information, a large number of training parameters, limited flexibility of classification-based methods, and the lack of systematic evaluation metrics for protein Q&A systems. To tackle these issues, we propose the Prot2Chat framework. Results: We modified ProteinMPNN to encode protein sequence and structural information in a unified way. We used a large language model (LLM) to encode questions into vectors and developed a protein-text adapter to compress protein information into virtual tokens based on these vectors, achieving the early fusion of text and protein information. Finally, the same LLM reads the virtual tokens and the questions to generate answers. To optimize training efficiency, we froze the encoder and employed Low-Rank Adaptation (LoRA) techniques for the LLM. Experiments on two datasets show that both automated metrics and expert evaluations demonstrate the superior performance of our model, and zero-shot prediction results highlight its generalization ability. We have developed an easy-to-use web interactive platform and a rapid installation option, allowing users to swiftly engage with Prot2Chat. Availability and Implementation: The models and codes are available at https://github.com/wangzc1233/Prot2Chat. Supplementary information: Supplementary data are available at Bioinformatics.
- 28 July 2025
- Published by Oxford University Press (OUP) in Bioinformatics
- Vol. 41 (8)
Open AccessProt2Chat: Protein LLM with Early-Fusion of Text, Sequence and Structure
Motivation: Proteins are of great significance in living organisms. However, understanding their functions encounters numerous challenges, such as insufficient integration of multimodal information, a large number of training parameters, limited flexibility of classification-based methods, and the lack of systematic evaluation metrics for protein Q&A systems. To tackle these issues, we propose the Prot2Chat framework. Results: We modified ProteinMPNN to encode protein sequence and structural information in a unified way. We used a large language model (LLM) to encode questions into vectors and developed a protein-text adapter to compress protein information into virtual tokens based on these vectors, achieving the early fusion of text and protein information. Finally, the same LLM reads the virtual tokens and the questions to generate answers. To optimize training efficiency, we froze the encoder and employed Low-Rank Adaptation (LoRA) techniques for the LLM. Experiments on two datasets show that both automated metrics and expert evaluations demonstrate the superior performance of our model, and zero-shot prediction results highlight its generalization ability. We have developed an easy-to-use web interactive platform and a rapid installation option, allowing users to swiftly engage with Prot2Chat. Availability and Implementation: The models and codes are available at https://github.com/wangzc1233/Prot2Chat. Supplementary information: Supplementary data are available at Bioinformatics.
- Open Access
- 11 April 2025
- Published by Association for the Advancement of Artificial Intelligence (AAAI) in Proceedings of the AAAI Conference on Artificial Intelligence
- Vol. 39 (3) , 3077-3085
AIM: Let Any Multimodal Large Language Models Embrace Efficient In-Context Learning
In-context learning (ICL) advances Large Language Models (LLMs) exhibiting emergent ability on downstream tasks without updating billions of parameters. However, in the area of multimodal Large Language Models (MLLMs), two problems hinder the application of multimodal ICL: (1) Most primary MLLMs are only trained on single-image datasets, making them unable to read extra multimodal demonstrations. (2) With the demonstrations increasing, thousands of visual tokens highly challenge hardware and degrade ICL performance. During preliminary explorations, we discovered that the inner LLM focuses more on the linguistic modality within multimodal demonstrations during generation. Therefore, we propose a general and lightweight framework AIM to tackle the mentioned problems through Aggregating Image information of Multimodal demonstrations to the latent space of the corresponding textual labels. After aggregation, AIM substitutes each demonstration with generated fused virtual tokens whose length is reduced to the same as its texts. Except for shortening input length, AIM further upgrades MLLMs pre-trained on image-text pairs to support multimodal ICL, as images from demonstrations are disregarded. Furthermore, benefiting from aggregating different demonstrations independently, AIM configures Demonstration Bank (DB) to avoid repeated aggregation, which significantly boosts model efficiency. We build AIM upon QWen-VL and LLaVA-Next, and AIM is comprehensively evaluated on image caption, VQA, and hateful speech detection. Outstanding results reveal that AIM provides an efficient and effective solution in upgrading MLLMs for multimodal ICL.
- 11 April 2025
- Published by Association for the Advancement of Artificial Intelligence (AAAI) in Proceedings of the AAAI Conference on Artificial Intelligence
- Vol. 39 (3) , 3077-3085
Open AccessAIM: Let Any Multimodal Large Language Models Embrace Efficient In-Context Learning
In-context learning (ICL) advances Large Language Models (LLMs) exhibiting emergent ability on downstream tasks without updating billions of parameters. However, in the area of multimodal Large Language Models (MLLMs), two problems hinder the application of multimodal ICL: (1) Most primary MLLMs are only trained on single-image datasets, making them unable to read extra multimodal demonstrations. (2) With the demonstrations increasing, thousands of visual tokens highly challenge hardware and degrade ICL performance. During preliminary explorations, we discovered that the inner LLM focuses more on the linguistic modality within multimodal demonstrations during generation. Therefore, we propose a general and lightweight framework AIM to tackle the mentioned problems through Aggregating Image information of Multimodal demonstrations to the latent space of the corresponding textual labels. After aggregation, AIM substitutes each demonstration with generated fused virtual tokens whose length is reduced to the same as its texts. Except for shortening input length, AIM further upgrades MLLMs pre-trained on image-text pairs to support multimodal ICL, as images from demonstrations are disregarded. Furthermore, benefiting from aggregating different demonstrations independently, AIM configures Demonstration Bank (DB) to avoid repeated aggregation, which significantly boosts model efficiency. We build AIM upon QWen-VL and LLaVA-Next, and AIM is comprehensively evaluated on image caption, VQA, and hateful speech detection. Outstanding results reveal that AIM provides an efficient and effective solution in upgrading MLLMs for multimodal ICL.