Glaucoma is a leading cause of irreversible blindness worldwide, with asymptomatic early stages often delaying diagnosis and treatment. Early and accurate diagnosis requires integrating complementary information from multiple ocular imaging modalities. However, most existing studies rely on single- or dual-modality imaging, such as fundus and optical coherence tomography (OCT), for coarse binary classification, thereby restricting the exploitation of complementary information and hindering both early diagnosis and stage-specific treatment. To address these limitations, we propose glaucoma lesion evaluation and analysis with multimodal imaging (GLEAM), the first publicly available tri-modal glaucoma dataset comprising scanning laser ophthalmoscopy fundus images, circumpapillary OCT images, and visual field pattern deviation maps, annotated with four disease stages, enabling effective exploitation of multimodal complementary information and facilitating accurate diagnosis and treatment across disease stages. To effectively integrate cross-modal information, we propose hierarchical attentive masked modeling (HAMM) for multimodal glaucoma classification. Our framework employs hierarchical attentive encoders and light decoders to focus cross-modal representation learning on the encoder. The attention module, named multimodal-channel graph attention (MCGA), boosts glaucoma classification performance by emulating two key clinical reasoning steps: first, it uses a multi-head modality gating mechanism to replicate ophthalmologists' confidence scoring of fundus, OCT, and VF modalities; then, MCGA leverages a relational graph attention network to cross-examine structural-functional consistencies of weighted modalities. The experiments on GLEAM demonstrate that tri-modal fusion significantly outperforms single-modal and dual-modal configurations. Moreover, our proposed HAMM achieves superior performance compared with state-of-the-art multimodal learning methods. The dataset and code are publicly available via https://github.com/microewing/HAMM.

Recent efforts have promoted programmable wireless environments (PWEs) to enhance the reception quality in high-frequency bands via reconfigurable intelligent surfaces (RISs). However, relevant research efforts are limited to setups with stationary users. This paper shows that crowd mobility in indoor PWEs induces spatio-temporal shadows on the surfaces, resulting in spatio-temporal sparsity in channel gains due to signal blockages. This overlooked aspect impacts the operation strategy of PWEs as the shadowed RIS tiles would contribute to the overheads while offering almost no improvement to the reception quality. Hence, this paper proposes an optimal strategy that excludes the shadowed tiles, which maximizes the utilization efficiency of RISs while minimizing the overheads. Since signal blockage is tied with the details of user mobility, a general model does not exist to identify such shadowed tiles for exclusion. Hence, we follow a data-driven approach that capitalizes on a realistic indoor mobility model and ray-tracing to generate the channel data. However, conventional ray-tracing presents high complexity that hinders data generation. So, we propose an approach to identify the shadow regions with a nine-order of magnitude reduction in complexity to efficiently generate the channel data. Furthermore, we present two exclusion strategies that offer guaranteed and best-effort quality-of-service support, and each can identify the tiles to be excluded via a search method with a complexity of$\mathcal {O}(N)$for$N$tiles. The results indicate that the proposed strategies reduce the overheads by$45-50\%$while maintaining optimal service quality in various environments, operation frequencies, and user and access point density.

Lane change and crash risk prediction are critical technologies for autonomous driving. An attention-based LSTM model is proposed in this paper for lane change behavior prediction in highways, considering both the surrounding vehicles’ information around the target vehicle at the current time and the historical trajectory data of the vehicle, and it shows higher accuracy and better interpretability than other models. There are two sections in this prediction model, including a pre-judgment model based on the C4.5 decision tree and bagging ensemble learning, and a multi-step lane change prediction model of LSTM with attention mechanism. In addition, the model is applied in the NGSIM datasets to test practicability and accuracy, and the precision of left lane change and right lane change is 98% and 94% respectively 1 second before the vehicle changes lane. Finally, to judge the safety of vehicles in the driving process, a crash risk prediction model based on Time-To-Collision is proposed, which further verifies the effectiveness of the method.

Next-generation wireless networks rely on unused high spectrum bands, such as mmWave and visible light, to support high-speed communications. However, such bands suffer from frequent outages of line-of-sight (LoS) signals, which deteriorates the communication's quality. Thus, recent efforts have promoted programmable wireless environments (PWEs), which employs reconfigurable intelligent surfaces (RIS) that can redirect non-LoS (NLoS) signals towards the user to enhance reception quality. Unfortunately, these efforts are limited so far to setups with stationary users. We show in this paper that crowd mobility induces spatio-temporal sparseness across the wireless channels in an indoor layout. As a result, shadowed regions will exist on the walls of the room due to signal blockage. Hence, random allocations of passive RIS tiles across the room will not result in the utmost gains and will unnecessarily complicate the communication system. Therefore, we propose an efficient data driven approach that capitalizes on a realistic indoor mobility model to allocate RIS tiles across the room in a way that maximizes their exploitation efficiency. Our results indicate that efficient design of PWEs can reduce the number of required passive RIS tiles by 70% while maintaining high service quality.

Optical wireless networks, especially those relying on visible light communications, suffer from severe deterioration in signal's quality when the line-of-sight (LOS) link is absent due to user's mobility. In order to enable efficient resource management within such networks, reliable prediction of LOS link outage is essential. Towards this objective, this article proposes a data-driven approach based on deep machine learning techniques to predict the outage events in an LOS link. First, we present a framework to generate sufficient data representing the channel gain in mobile optical wireless networks that consist of visible light communications in the downlink and infrared communications in the uplink. Using the developed dataset, we propose a channel predictor that forecasts the burst outages or signal recoveries in the upcoming frames using a deep recurrent neural network that implements long-short-term-memory (LSTM) units. To achieve this goal, we propose a low-complexity approach to reduce the data sparsity due to the user's mobility by abstracting and densifying the channel state sequence. For a one second prediction interval, the proposed prediction framework achieves an event hit rate of 91.55% for abrupt outages with an average event timing error of 79 ms, and 83.19% for recoveries from outages with 145 ms timing error. This timing error is on the same order of magnitude with the coherence time of the optical wireless channel. Therefore, this predictor is very useful in developing efficient resource management strategies in such optical networks.

This article investigates a heterogeneous network (HetNet) consisting of an overlapped coverage of radio frequency (RF) and optical wireless communication (OWC) to support human connectivity to the Internet of Things (IoT) in the fifth generation and beyond (5G+) mobile networks. Such a HetNet-based IoT benefits from the high throughput of the OWC and the high reliability of the RF communications. To ensure a reliable link with Quality-of-Service (QoS) guarantee in terms of network delay and throughput, vertical handovers are triggered within the HetNet. A cross-layer data-driven approach is adopted to reach optimal handover decisions and tackle the challenges associated with the mobility and reliability of IoT. As mobile time-varying wireless channel gains in optical links are not publicly available, we first present a realistic model for the mobile optical channel that reflects the nature of human mobility, and hence, a useful model that features the spatial-temporal patterns of indoor mobile channels is proposed. Using the created data set, we then present a data-driven algorithm that predicts abrupt outages in Line-of-Sight (LOS) optical links and evaluates the optical channel quality through deep learning. Given the resulting LOS link outage prediction in OWC, a reinforcement-learning-based approach is proposed to implement optimal vertical handover decisions with the QoS guarantee. The proposed handover decision algorithm learns to make a tradeoff between the outage risk and the cost of excessive handovers. The numerical results demonstrate considerable improvement in overall latency and handover rate under indoor mobility for bidirectional links.

Link instability induced by users' mobility is one of the challenges of optical wireless communications (OWC) inherited from the propagation nature of light. Hence, good understanding of the optical channel characteristics in dynamic environments plays a vital role in developing robust resource management strategies in OWC networks. Unfortunately, it is quite difficult to collect accurate indoor optical channel data in dynamic environments. In addition, indoor trajectory dataset is not publicly available. To overcome such limitations, this paper proposes a mobile terminal-centric analytical framework that captures the propagation channel characteristics in a mobile OWC network whose downlink is based on visible light and uplink is based on infrared light. We abstract the nature of human behavior by integrating both macro and micro mobility patterns. These patterns are then used to realize the spatio-temporal characteristics of optical wireless channels under long-term environment-confined mobility. The statistics derived from the developed framework indicate that the mobile line-of-sight (LOS) channel gain follows space-time-dependent multiple-peak Nakagami distributions, whereas the non-line-of-sight (NLOS) channel gain adheres to various space-time-dependent single peak distributions under different indoor layouts. The overall distribution of NLOS bandwidth follows space-time-dependent multiple-peak log-logistic distributions in downlinks and space-time-dependent generalized log-logistic distributions in uplinks. Our investigation demonstrates that the indoor layout and the user's environment-confined mobility pattern significantly impact the LOS dynamics but present limited impact on NLOS components. Motivated by the need for better channel models for mobile OWC, the proposed framework fills up an important gap in literature and help the research community to understand better the indoor optical wireless channel characteristics.

Phantom Go is a derivative of Go with imperfect information. It is challenging in AI field due to its great uncertainty of the hidden information and high game complexity inherited from Go. To deal with this imperfect information game with large game tree complexity, a general search framework named belief-state Monte Carlo tree search (BS-MCTS) is put forward in this paper. BS-MCTS incorporates belief-states into Monte Carlo Tree Search, where belief-state is a notation derived from philosophy to represent the probability that speculation is in accordance with reality. In BS-MCTS, a belief-state tree, in which each node is a belief-state, is constructed and search proceeds in accordance with beliefs. Then, Opponent Guessing and Opponent Predicting are proposed to illuminate the learning mechanism of beliefs with heuristic information. The beliefs are learned by heuristic information during search by specific methods, and we propose Opponent Guessing and Opponent Predicting to illuminate the learning mechanism. Besides, some possible improvements of the framework are investigated, such as incremental updating and all moves as first (AMAF) heuristic. Technical details are demonstrated about applying BS-MCTS to Phantom Go, especially on inference strategy. We examine the playing strength of the BS-MCTS and AMAF-BS-MCTS in Phantom Go by varying search parameters, also testify the proposed improvements.

Automatically acquiring domain knowledge from professional game records, a kind of pattern learning, is an attractive and challenging issue in computer Go. This paper proposes a supervised learning method, by introducing a new generalized Bradley-Terry model, named Only-One-Victor, to learn patterns from game records. Basically, our algorithm applies the same idea with Elo rating algorithm, which considers each move in game records as a group of move patterns, and the selected move as the winner of a kind of competition among all groups on current board. However, being different from the generalized Bradley-Terry model for group competition used in Elo rating algorithm, Only-One-Victor model in our work simulates the process of making selection from a set of possible candidates by considering such process as a group of independent pairwise comparisons. We use a graph theory model to prove the correctness of Only-One-Victor model. In addition, we also apply the Minorization-Maximization (MM) to solve the optimization task. Therefore, our algorithm still enjoys many computational advantages of Elo rating algorithm, such as the scalability with high dimensional feature space. With the training set containing 115,832 moves and the same feature setting, the results of our experiments show that Only-One-Victor outperforms Elo rating, a well-known best supervised pattern learning method.