1 March 2021 in Engineering with Computers
28 November 2023 in IEEE Transactions on Vehicular Technology
In this paper, we employ a reconfigurable intelligent surface (RIS) to spread incident signals and develop RIS-based code index modulation (CIM) for joint active/passive transmission (JAPT). In RIS-CIM, the RIS performs passive beamforming to simultaneously maximize the receive signal-to-noise ratio and spread the incident signals from the active transmission. The indices of spreading codes are used to transmit passive information. We design a low-complexity (LC) yet near-optimal detector to recover the active and passive information in addition to the optimal maximum-likelihood (ML) detector. Upper bounds and approximate expressions are derived in closed-form for the bit error rate of RIS-CIM with the ML and LC detectors, respectively. Subsequently, we extend the idea of RIS-CIM to the scenario of multi-antenna transmission. Simulation results verify the performance analysis and show the superiority of RIS-CIM over the existing RIS-aided JAPT schemes.
19 July 2023 in Frontiers in Computational Neuroscience
An epileptic seizure is the external manifestation of abnormal neuronal discharges, which seriously affecting physical health. The pathogenesis of epilepsy is complex, and the types of epileptic seizures are diverse, resulting in significant variation in epileptic seizure data between subjects. If we feed epilepsy data from multiple patients directly into the model for training, it will lead to underfitting of the model. To overcome this problem, we propose a robust epileptic seizure detection model that effectively learns from multiple patients while eliminating the negative impact of the data distribution shift between patients. The model adopts a multi-level temporal-spectral feature extraction network to achieve feature extraction, a feature separation network to separate features into category-related and patient-related components, and an invariant feature extraction network to extract essential feature information related to categories. The proposed model is evaluated on the TUH dataset using leave-one-out cross-validation and achieves an average accuracy of 85.7%. The experimental results show that the proposed model is superior to the related literature and provides a valuable reference for the clinical application of epilepsy detection.
29 November 2024 in Cognitive Computation
12 December 2025 in IEEE Communications Letters
Affine frequency division multiplexing (AFDM) is a promising modulation scheme for high-mobility wireless communications. However, it faces the issue of high peak-to-average power ratio (PAPR). In this letter, we propose a deep neural network, termed GIS-ACE-Net, to reduce the PAPR of pilot-embedded AFDM signals that have two null guard intervals (GIs) between the pilot and data symbols. By utilizing this signal feature, GIS-ACE-Net replaces the GIs with low-amplitude symbols and performs the active constellation extension on the data symbols simultaneously with the PAPR being the loss function. Simulation results demonstrate that GIS-ACE-Net achieves significant PAPR reduction with a slight impact on the bit error rate performance under pilot-embedded channel estimation.
18 October 2023 in Research Square
In the context of generative multimodal dialogue, the amalgamation of image and text forms the fundamental basis for a wide array of visual and language tasks. Despite the proliferation of large language models (LLMs), research on training LLMs for multimodal reasoning has remained relatively scarce. The progression of existing multimodal models has been propelled by using expansive image-text datasets in broader domains. Nevertheless, these models need refinement when comprehending and engaging in dialogue concerning Chinese forestry biological images. This paper proposes an efficient and adaptive multimodal large language model for forestry (MLLM-For) in pest and disease recognition. The central concept involves harnessing our meticulously constructed, expansive Chinese forestry biological image caption dataset. We utilize open-ended instruction-following data to master the intricacies of open-ended dialogue semantics, facilitating the comprehensive assimilation of knowledge within forestry biology. In particular, the Querying Transformer (Q-Former) module is a conduit linking pre-trained visual and language models. Using image-text pairs, the China Open Source Chat Language Model (ChatGLM) can now align vocabulary specific to forestry pests and diseases. By aligning image features with the pre-trained language model, we undertake concurrent Low-Rank Adaptive (LoRA) fine-tuning of the ChatGLM model. Furthermore, we engage in high-caliber instruction data fine-tuning to 1 tailor it to the proposed multimodal task. The proposed MLLM-For model show-cases impressive proficiency in comprehending and engaging in dialogue regarding Chinese forestry pests and disease images. It also displays exceptional capabilities in multimodal dialogue and is adept at adhering to open-ended directives for retrieving information about forestry pests and diseases. Empirical findings affirm the competitive performance of the MLLM-For model within the realm of Chinese multimodal dialogue, surpassing five leading multimodal large language models.