Dongqi Cai
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Dongqi Cai has a total of 57 co-authors. Here are the top co-authors by number of shared publications:
- Mengwei Xu
Beijing University of Posts and Telecommunications
11 shared publications - Shangguang Wang
Beijing University of Posts and Telecommunications
11 shared publications - Yaozong Wu
Beiyou Shenzhen Institute, Shenzhen, China
4 shared publications - Felix Xiaozhu Lin
University of Virginia, Charlottesville, VA, USA
3 shared publications - Nicholas Donald Lane
University of Cambridge
3 shared publications - Zeling Zhang
State Key Laboratory of Networking and Switching Technology, Beijing University of Posts and Telecommunications, Beijing, China
3 shared publications
Latest Publications
- Open Access
- 11 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 7-13
RTSG-SLAM: Towards Real-Time Semantic Gaussian SLAM for Vision-Language Navigation
- 11 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 7-13
Open AccessRTSG-SLAM: Towards Real-Time Semantic Gaussian SLAM for Vision-Language Navigation
- 10 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Mobile Computing
- Vol. 25 (8) , 12735-12749
FwdLLM+: Accelerating Forward-Only FedLLM With Low-Rank Perturbations
Federated Learning (FL) facilitates privacy-preserving fine-tuning of Large Language Models (LLMs) for mobile applications, termed FedLLM. A vital challenge of FedLLM is the tension between LLM complexity and resource constraint of mobile devices. In response to this challenge, we first introduce FwdLLM (our conference version), an innovative FL framework designed to enhance the FedLLM efficiency. The key idea of FwdLLM is to employ backpropagation (BP)-free training methods, requiring devices only to execute memory-efficient “perturbed inference”. Enabled by mobile NPU acceleration and an expanded array of participating devices, FwdLLM delivers substantially better wall-clock efficiency than BP-based FedLLM. However, FwdLLM based on vanilla BP-free optimization theoretically requires more optimization steps to converge. In this work, we further enhance FwdLLM to FwdLLM+, which incorporates advanced zeroth-order optimization techniques and low-rank perturbation decomposition to reduce convergence steps. Finally, we conduct extensive experiments on 4 models (ranging from 110 M to 7B) and 8 more datasets, demonstrating that FwdLLM+ achieves up to 151× faster training, a 93$\%$ memory reduction compared to vanilla BP-based FedLLM, and superior performance compared to FwdLLM, enabling efficient federated fine-tuning of billion-parameter LLMs on commodity mobile devices.
- 10 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Mobile Computing
- Vol. 25 (8) , 12735-12749
FwdLLM+: Accelerating Forward-Only FedLLM With Low-Rank Perturbations
Federated Learning (FL) facilitates privacy-preserving fine-tuning of Large Language Models (LLMs) for mobile applications, termed FedLLM. A vital challenge of FedLLM is the tension between LLM complexity and resource constraint of mobile devices. In response to this challenge, we first introduce FwdLLM (our conference version), an innovative FL framework designed to enhance the FedLLM efficiency. The key idea of FwdLLM is to employ backpropagation (BP)-free training methods, requiring devices only to execute memory-efficient “perturbed inference”. Enabled by mobile NPU acceleration and an expanded array of participating devices, FwdLLM delivers substantially better wall-clock efficiency than BP-based FedLLM. However, FwdLLM based on vanilla BP-free optimization theoretically requires more optimization steps to converge. In this work, we further enhance FwdLLM to FwdLLM+, which incorporates advanced zeroth-order optimization techniques and low-rank perturbation decomposition to reduce convergence steps. Finally, we conduct extensive experiments on 4 models (ranging from 110 M to 7B) and 8 more datasets, demonstrating that FwdLLM+ achieves up to 151× faster training, a 93$\%$ memory reduction compared to vanilla BP-based FedLLM, and superior performance compared to FwdLLM, enabling efficient federated fine-tuning of billion-parameter LLMs on commodity mobile devices.
- Open Access
- 21 February 2026
- Published by Springer Nature in Nature Communications
- Vol. 17 (1)
Auditing unauthorized training data from AI generated content using information isotopes
The rapid growth of AI systems has been fueled by large-scale human data, intensifying concerns over the unauthorized use of intellectual property and privacy-sensitive content during model training. Auditing such misuse is particularly challenging since mainstream AI services operate as black boxes, exposing only generated outputs while concealing their training and inference processes. In this work, inspired by chemical isotope tracing, we introduce the concept of information isotopes to trace training data within opaque AI systems. We propose an information-isotope tracing framework that selectively marks target data elements and detects their propagation in model outputs, providing concrete evidence of data utilization under black-box access. Experiments on thirteen AI models across six datasets demonstrate that our method distinguishes training from non-training data with up to 99% accuracy and strong statistical significance (p < 0.01) using approximately 4,000 words of evidence. An open-source tool is released to support practical data rights protection.
- 21 February 2026
- Published by Springer Nature in Nature Communications
- Vol. 17 (1)
Open AccessAuditing unauthorized training data from AI generated content using information isotopes
The rapid growth of AI systems has been fueled by large-scale human data, intensifying concerns over the unauthorized use of intellectual property and privacy-sensitive content during model training. Auditing such misuse is particularly challenging since mainstream AI services operate as black boxes, exposing only generated outputs while concealing their training and inference processes. In this work, inspired by chemical isotope tracing, we introduce the concept of information isotopes to trace training data within opaque AI systems. We propose an information-isotope tracing framework that selectively marks target data elements and detects their propagation in model outputs, providing concrete evidence of data utilization under black-box access. Experiments on thirteen AI models across six datasets demonstrate that our method distinguishes training from non-training data with up to 99% accuracy and strong statistical significance (p < 0.01) using approximately 4,000 words of evidence. An open-source tool is released to support practical data rights protection.
- Open Access
- 23 June 2025
- Published by Association for Computing Machinery (ACM)
- p. 657-658
Federated LLM Pre-Training on Mobile Phones
- 23 June 2025
- Published by Association for Computing Machinery (ACM)
- p. 657-658
Open AccessFederated LLM Pre-Training on Mobile Phones
- Open Access
- 19 June 2025
- Published by Springer Nature in Nature Communications
- Vol. 16 (1) , 1-12
Ubiquitous memory augmentation via mobile multimodal embedding system
Forgetting is inevitable in human memory. Recently, multimodal embedding models have been proposed to vectorize multimodal reality into a unified embedding space. Once generated, these embeddings allow mobile users to quickly retrieve relevant information, effectively augmenting their memory. However, as the model’s capacity increases, its resource consumption also rises. The resulting slow throughput and significant computational resource requirements hinder its deployment on mobile devices. In this paper, we present Reminisce, an efficient on-device multimodal embedding system that enables high-throughput embedding and precise retrieval on resource-constrained mobile devices. The core design draws inspiration from the memory functions of the human brain, utilizing coarse-grained embeddings to identify likely candidates, which are then refined through query-driven fine-grained retrieval. A series of algorithm-hardware orchestrated optimizations automatically navigates this process and strengthen the embedding quality. Experiments show that Reminisce provides high-quality embedding representation with high throughput while operating silently in the background with negligible memory usage and reduced energy consumption.
- 19 June 2025
- Published by Springer Nature in Nature Communications
- Vol. 16 (1) , 1-12
Open AccessUbiquitous memory augmentation via mobile multimodal embedding system
Forgetting is inevitable in human memory. Recently, multimodal embedding models have been proposed to vectorize multimodal reality into a unified embedding space. Once generated, these embeddings allow mobile users to quickly retrieve relevant information, effectively augmenting their memory. However, as the model’s capacity increases, its resource consumption also rises. The resulting slow throughput and significant computational resource requirements hinder its deployment on mobile devices. In this paper, we present Reminisce, an efficient on-device multimodal embedding system that enables high-throughput embedding and precise retrieval on resource-constrained mobile devices. The core design draws inspiration from the memory functions of the human brain, utilizing coarse-grained embeddings to identify likely candidates, which are then refined through query-driven fine-grained retrieval. A series of algorithm-hardware orchestrated optimizations automatically navigates this process and strengthen the embedding quality. Experiments show that Reminisce provides high-quality embedding representation with high throughput while operating silently in the background with negligible memory usage and reduced energy consumption.
- Open Access
- 16 January 2025
- Published by Springer Nature in Research Square
Ubiquitous Memory Augmentation via Mobile Multimodal Embedding System
Forgetting is inevitable in human memory. Recently, multimodal embedding models have been proposed to vectorize multimodal reality into a unified embedding space. The generated embeddings can be easily retrieved to help mobile users remember and recall information when needed. However, as the model’s capacity increases, its resource consumption also rises. The resulting slow throughput and significant computational resource requirements hinder its deployment on mobile devices. In this paper, we present Reminisce, the first efficient on-device multimodal embedding system that enables high-throughput and precise retrieval on resource-constrained mobile devices. The core design draws inspiration from the memory functions of the human brain, utilizing coarse-grained embeddings to identify likely candidates, which are then refined through query-driven fine-grained retrieval. A series of algorithm-hardware orchestrated optimizations automatically navigates this process and strengthens the embedding quality. Experiments show that Reminisce provides high-quality embedding representation with high throughput while operating silently in the background with negligible memory usage and reduced energy consumption.
- 16 January 2025
- Published by Springer Nature in Research Square
Open AccessUbiquitous Memory Augmentation via Mobile Multimodal Embedding System
Forgetting is inevitable in human memory. Recently, multimodal embedding models have been proposed to vectorize multimodal reality into a unified embedding space. The generated embeddings can be easily retrieved to help mobile users remember and recall information when needed. However, as the model’s capacity increases, its resource consumption also rises. The resulting slow throughput and significant computational resource requirements hinder its deployment on mobile devices. In this paper, we present Reminisce, the first efficient on-device multimodal embedding system that enables high-throughput and precise retrieval on resource-constrained mobile devices. The core design draws inspiration from the memory functions of the human brain, utilizing coarse-grained embeddings to identify likely candidates, which are then refined through query-driven fine-grained retrieval. A series of algorithm-hardware orchestrated optimizations automatically navigates this process and strengthens the embedding quality. Experiments show that Reminisce provides high-quality embedding representation with high throughput while operating silently in the background with negligible memory usage and reduced energy consumption.
- Top Cited PapersOpen Access
- 9 January 2025
- Published by Association for Computing Machinery (ACM) in ACM Computing Surveys
- Vol. 57 (5) , 1-39
Resource-efficient Algorithms and Systems of Foundation Models: A Survey
Large foundation models, including large language models, vision transformers, diffusion, and large language model based multimodal models, are revolutionizing the entire machine learning lifecycle, from training to deployment. However, the substantial advancements in versatility and performance these models offer come at a significant cost in terms of hardware resources. To support the growth of these large models in a scalable and environmentally sustainable way, there has been a considerable focus on developing resource-efficient strategies. This survey delves into the critical importance of such research, examining both algorithmic and systemic aspects. It offers a comprehensive analysis and valuable insights gleaned from existing literature, encompassing a broad array of topics from cutting-edge model architectures and training/serving algorithms to practical system designs and implementations. The goal of this survey is to provide an overarching understanding of how current approaches are tackling the resource challenges posed by large foundation models and to potentially inspire future breakthroughs in this field.
- 9 January 2025
- Published by Association for Computing Machinery (ACM) in ACM Computing Surveys
- Vol. 57 (5) , 1-39
Top Cited PapersOpen AccessResource-efficient Algorithms and Systems of Foundation Models: A Survey
Large foundation models, including large language models, vision transformers, diffusion, and large language model based multimodal models, are revolutionizing the entire machine learning lifecycle, from training to deployment. However, the substantial advancements in versatility and performance these models offer come at a significant cost in terms of hardware resources. To support the growth of these large models in a scalable and environmentally sustainable way, there has been a considerable focus on developing resource-efficient strategies. This survey delves into the critical importance of such research, examining both algorithmic and systemic aspects. It offers a comprehensive analysis and valuable insights gleaned from existing literature, encompassing a broad array of topics from cutting-edge model architectures and training/serving algorithms to practical system designs and implementations. The goal of this survey is to provide an overarching understanding of how current approaches are tackling the resource challenges posed by large foundation models and to potentially inspire future breakthroughs in this field.
- 20 September 2024
- Published by Springer Nature
- p. 132-141
GPT4D: Automatic Cross-Version Linux Driver Upgrade Toolkit
- 20 September 2024
- Published by Springer Nature
- p. 132-141
GPT4D: Automatic Cross-Version Linux Driver Upgrade Toolkit
- Open Access
- 3 June 2024
- Published by Association for Computing Machinery (ACM)
- p. 1-6
Large Language Models on Mobile Devices: Measurements, Analysis, and Insights
Deploying large language models (LLMs) inference into mobile devices is cost-efficient for companies, and well addresses the privacy concern of users. However, the limited computation capacity and memory constraints of mobile devices hinder their practical deployment. Prior work strives to expand model size for better accuracy performance, while there is a lack of systematic understanding of "small" sub-10 billion LLMs that are already feasible for current commodity devices. To better reveal the current landscape of LLMs on mobile devices, we conducted a comprehensive measurement study, deploying 22 models across 4 mobile devices. Our measurements focus on accuracy, inference latency, and memory footprint across various input lengths, devices, and execution engines. The observations from the measurements point us toward promising directions for efficient LLM deployment on mobile devices.
- 3 June 2024
- Published by Association for Computing Machinery (ACM)
- p. 1-6
Open AccessLarge Language Models on Mobile Devices: Measurements, Analysis, and Insights
Deploying large language models (LLMs) inference into mobile devices is cost-efficient for companies, and well addresses the privacy concern of users. However, the limited computation capacity and memory constraints of mobile devices hinder their practical deployment. Prior work strives to expand model size for better accuracy performance, while there is a lack of systematic understanding of "small" sub-10 billion LLMs that are already feasible for current commodity devices. To better reveal the current landscape of LLMs on mobile devices, we conducted a comprehensive measurement study, deploying 22 models across 4 mobile devices. Our measurements focus on accuracy, inference latency, and memory footprint across various input lengths, devices, and execution engines. The observations from the measurements point us toward promising directions for efficient LLM deployment on mobile devices.
- Open Access
- 29 May 2024
- Published by Association for Computing Machinery (ACM)
- p. 279-295
Mobile Foundation Model as Firmware
- 29 May 2024
- Published by Association for Computing Machinery (ACM)
- p. 279-295
Open AccessMobile Foundation Model as Firmware