A novel combined phase perturbation and probabilistic shaping scheme is proposed to mitigate mode crosstalk in MDM-PONs. Receiver sensitivities are improved by 4-dB and 4.2-dB for LP01 and LP11a modes in 20-Gbit/s PAM4 transmission, respectively.

In this work, we propose a Bayesian thinning algorithm for recovering weighted point source functions in the heat equation from boundary flux observations. The major challenge in the classical Bayesian framework lies in constructing suitable priors for such highly structured unknowns. To address this, we introduce a level set representation on a discretized mesh for the unknown, which enables the infinite-dimensional Bayesian framework to the reconstruction. From another perspective, the point source configuration can be modeled as a marked Poisson point process (PPP), then a thinning mechanism is employed to selectively retain points. These two proposals are complementary with the Bayesian level set sampling generating candidate point sources and the thinning process acting as a filter to refine them. This combined framework is validated through numerical experiments, which demonstrate its accuracy in reconstructing point sources.

Data-driven landslide susceptibility assessment (LSA) remains unconvincing owing to the disconnection from the modelling to physical cognition of landslide causation. Most models are mere good fits to certain datasets and can produce unexpected bias in their prediction, misleading high-risk area zoning. To validate data-driven LSA, this study delved into the innate interactions between input landslide features and predictions by model explanation and compared feature permutation results with landslide statistical priors. Furthermore, multi-temporal interferometric synthetic aperture radar (MT-InSAR) derived ground deformation was applied in an alpha pixel fusion and growth method for LSA enhancement. This study took Hong Kong as the research area, employed extreme gradient boosting (XGBoost) for LSA, and utilized Shapley additive explanations (SHAP) method for model explanation. The mean Shapley values - indicating feature importance - for slope, stream power index (SPI) and land use are 1.31, 1.12, and 0.67, respectively. This aligns with the landslide feature permutation derived from prior statistics, verifying the model prediction reliability. The applied InSAR enhancement results in a 13% increase of '(very) high' landslide susceptibility areas. Additionally, LSA results and ground deformation map were cross-validated in the virtual geographic environment. This study improves the reliability of data-driven LSA through model explanation and InSAR enhancement.

Landslide susceptibility assessment (LSA) plays a vital role in disaster prevention and mitigation. Recently, numerous data-driven LSA approaches have emerged. Nonetheless, most of them neglected the rapid oscillations within the landslide-prone environment, primarily due to significant changes in external triggers such as rainfall, which would render landslides susceptible to varying causations over time. Thus, conducting dynamic landslide susceptibility mapping (D-LSM) and revealing the underlying trends in landslide causes, become increasingly important for effective landslide hazard assessment. This study decomposed the entire D-LSM task into yearly LSA subtasks, and innovatively meta-learned intermediate representations that can be well-generalized and finetuned in a fast-adaptation manner. Then, to interpret the model predictions and characterize the variations in landslide causation, Shapley Additive exPlanations (SHAP) was utilized for feature permutation year by year. In addition, MT-InSAR techniques were applied to enhance and validate the D-LSM results. The study area was Lantau Island, Hong Kong, where the yearly LSA was executed from 1992 to 2019. The performance comparison results show that the proposed method outperformed the other approaches with regard to accuracy (3 %-7 %), precision (2 %-9 %), recall (3 %-5 %), and F1-score (2 %-7 %), even when adopting a fast adaptation strategy using only 5 samples and 5 gradient descent updates. This validates the applicability of meta-learning for identifying commonalities across multi-temporal LSA tasks. The overall model interpretation results indicate that slope and extreme rainfall were the primary contributors to landslide occurrences in Hong Kong. The feature permutation results over the 30 years reveal a variation in landslide causation, particularly a dramatic shift in the ranking of some contributing factors under extreme weather conditions. Remarkably, the importance of AERD (Annual Extreme Rainfall Days), a factor indicating extreme rainfall intensity, was deeply affected by global climate change and the government's Landslide Prevention and Mitigation Programme (LPMitP).

Mathematical capabilities were previously believed to emerge in common language models only at a very large scale or require extensive math-related pre-training. This paper shows that the LLaMA-2 7B model with common pre-training already exhibits strong mathematical abilities, as evidenced by its impressive accuracy of 97.7 respectively, when selecting the best response from 256 random generations. The primary issue with the current base model is the difficulty in consistently eliciting its inherent mathematical capabilities. Notably, the accuracy for the first answer drops to 49.5 respectively. We find that simply scaling up the SFT data can significantly enhance the reliability of generating correct answers. However, the potential for extensive scaling is constrained by the scarcity of publicly available math questions. To overcome this limitation, we employ synthetic data, which proves to be nearly as effective as real data and shows no clear saturation when scaled up to approximately one million samples. This straightforward approach achieves an accuracy of 82.6 models, surpassing previous models by 14.2 provide insights into scaling behaviors across different reasoning complexities and error types.

We demonstrate a method for reducing optical multipath interference (MPI) based on probabilistic shaping (PS). The result shows 2.8 dB enhancement in MPI tolerance for 25 GBaud PAM4 transmission over 10 km SSMF.

The detection of ballastless track surface (BTS) defects is a prerequisite for ensuring the safe operation of high-speed railways. Traditional convolutional neural networks fail to fully exploit contextual information and lack global pixel representations. The extensive stacking of convolutions leads deep learning models to play a black-box detection role, lacking interpretability. Due to the current lack of sufficient high-quality surface data for ballastless tracks, it is a severe constraint on the accurate identification of the substructure state in high-speed railways. This paper proposes an intelligent detection method for BTS defects named TrackNet based on self-attention and transfer learning. The method enhances the fusion ability of global features of BTS defects using multihead self-attention. The model's dependence on extensive defect data is reduced by transferring knowledge from large-scale publicly available datasets. Experimental results demonstrate that compared to advanced Swin Transformer model results, the TrackNet model achieves improvements in average accuracy and F1-score by 5.15% and 5.16%, respectively, on limited test data. The TrackNet model visualizes the decision regions of the model in identifying BTS defects, revealing the black-box recognition mechanism of deep learning models. This research performs engineering applications and provides valuable insights for the multiclass recognition of BTS defects in high-speed railways.

Early identification of debris-flow-prone watersheds and determination of the initiation location of debris flows are prerequisites for debris flow monitoring and early warning. The high altitude, steep slope, dense vegetation, and influence of numerous factors triggering debris flows (FTDFs) in Danba, China, pose great difficulties in identifying debris-flow-prone watersheds and locating the possible initiation position of debris flows in this area. We propose a watershed-oriented and multifactor-integrated (WOMI) method for identifying debris-flow-prone watersheds in Danba. This method integrates various FTDFs, uses the watershed as an analysis unit, and detects the debris-flow-prone watersheds with a random forest (RF) algorithm. Moreover, we propose a statistical similarity metric to evaluate the rationality of identification results and introduce the extenics theory to evaluate the debris-flow hazard of the identified debris-flow-prone watersheds and their tributaries, thereby locating the possible initiation positions of debris flow in the identified debris-flow-prone watersheds. Based on 26 FTDFs derived from multi-source data between 2015 and 2019 in Danba, 36 new debris-flow-prone watersheds between 2015 and 2019 were discovered in this study. Nine debris-flow-prone watersheds and their 17 tributaries have very high hazard. “6.17” Meilong Valley debris flow just erupted in one of the identified debris-flow-prone watersheds in 2020. The statistical analysis and occurrence status of debris flows validated the proposed methodology and the identified debris-flow-prone watersheds in Danba. The research results of this article assisted in the investigation, early warning, and prevention of debris flows in Danba.

We propose a novel channel impairment compensation scheme that combines unsupervised denoising and Least Square equalizer for next-generation PONs. The results demonstrate the proposed scheme enhances the BER performance by improving OSNR of received signals.

Landslide susceptibility assessment (LSA) is of paramount importance in mitigating landslide risks. Recently, there has been a surge in the utilization of data-driven methods for predicting landslide susceptibility due to the growing availability of aerial and satellite data. Nonetheless, the rapid oscillations within the landslide-inducing environment (LIE), primarily due to significant changes in external triggers such as rainfall, pose difficulties for contemporary data-driven LSA methodologies to accommodate LIEs over diverse timespans. This study presents dynamic landslide susceptibility mapping that simply employs multiple predictive models for annual LSA. In practice, this will inevitably encounter small sample problems due to the limited number of landslide samples in certain years. Another concern arises owing to the majority of the existing LSA approaches train black-box models to fit distinct datasets, yet often failing in generalization and providing comprehensive explanations concerning the interactions between input features and predictions. Accordingly, we proposed to meta-learn representations with fast adaptation ability using a few samples and gradient updates; and apply SHAP for each model interpretation and landslide feature permutation. Additionally, we applied MT-InSAR for LSA result enhancement and validation. The chosen study area is Lantau Island, Hong Kong, where we conducted a comprehensive dynamic LSA spanning from 1992 to 2019. The model interpretation results demonstrate that the primary factors responsible for triggering landslides in Lantau Island are terrain slope and extreme rainfall. The results also indicate that the variation in landslide causes can be primarily attributed to extreme rainfall events, which result from global climate change, and the implementation of the Landslip Prevention and Mitigation Programme (LPMitP) by the Hong Kong government.

Landslide susceptibility evaluation (LSE) is a critical issue for disaster prevention. Limited by labor cost and observation technology, landslide samples are extremely limited in dense vegetation‐covered and remote areas, making the common supervised learning model underfit with limited samples. Therefore, the reliability of analysis results in mountainous areas is low. Transfer learning can achieve reliable assessment without the need for representative samples. However, transfer learning suffers from environmental heterogeneity in regional LSE and may transfer incorrect classification knowledge of landslide features from dissimilar environments. Aiming at these challenges, we proposed a geo‐environment‐aware LSE method based on unsupervised adversarial transfer learning. The key is to consider the difference in landslide features in different geo‐environments. The study areas were first divided into multiple sub‐environments, and the similarity between the sub‐environments was calculated. Then an environment‐aware adversarial transfer model was built for fine‐grained aligning of the landslide feature with similar sub‐environments and for reducing negative transfer between dissimilar environments. The fitted classification model was employed to predict the target regions and to generate the final LSE. The experimental results indicated that the proposed method achieves reliable LSE for sample‐free regions. The accuracy of the proposed method is 7–12% better than commonly used methods such as support vector machines, random forests, and artificial neural networks. The performance of the proposed method is even close to the results of supervised learning with the presence of representative samples, and it also performs more globally and objectively in susceptibility mapping. These results reveal that the proposed method effectively transfers the knowledge of landslide susceptibility from other regions to the sample‐free region.

Landslide susceptibility assessment (LSA) is of paramount importance in mitigating landslide risks. Recently, there has been a surge in the utilization of data-driven methods for predicting landslide susceptibility due to the growing availability of aerial and satellite data. Nonetheless, the rapid oscillations within the landslide-inducing environment (LIE), primarily due to significant changes in external triggers such as rainfall, pose difficulties for contemporary data-driven LSA methodologies to accommodate LIEs over diverse timespans. This study presents dynamic landslide susceptibility mapping that simply employs multiple predictive models for annual LSA. In practice, this will inevitably encounter small sample problems due to the limited number of landslide samples in certain years. Another concern arises owing to the majority of the existing LSA approaches train black-box models to fit distinct datasets, yet often failing in generalization and providing comprehensive explanations concerning the interactions between input features and predictions. Accordingly, we proposed to meta-learn representations with fast adaptation ability using a few samples and gradient updates; and apply SHAP for each model interpretation and landslide feature permutation. Additionally, we applied MT-InSAR for LSA result enhancement and validation. The chosen study area is Lantau Island, Hong Kong, where we conducted a comprehensive dynamic LSA spanning from 1992 to 2019. The model interpretation results demonstrate that the primary factors responsible for triggering landslides in Lantau Island are terrain slope and extreme rainfall. The results also indicate that the variation in landslide causes can be primarily attributed to extreme rainfall events, which result from global climate change, and the implementation of the Landslip Prevention and Mitigation Programme (LPMitP) by the Hong Kong government.

Landslide susceptibility assessment (LSA) evaluates the likelihood of landslide occurrences and can help mitigate and prevent landslide risks. Recently, there have been vast applications of data-driven LSA methods owing to the increased availability of high-quality satellite data and landslide inventories. However, two issues remain to be addressed, as follows: (a) Items in a landslide inventory are mainly historical landslides from the interpretation of optical images and site investigation, resulting in predictive models trained with these items being insensitive to undetectable slope movements, such as slow-moving landslides that have not yet occurred; (b) Most study areas contain a variety of landslide-prone geographical settings that a single model can not accommodate well. Considering the complex landslide causes in Hong Kong with a land area of approximately 1108 km2, we proposed the utilization of multi-temporal InSAR techniques to generate weak landslide samples from slopes with ground surface movements for landslide inventory augmentation; and meta-learn intermediate representations for the fast adaptation of LSA models corresponding to different landslide-prone geographical settings. Besides, we performed feature permutation to identify dominant landslide-predisposing factors. The LSA results in Hong Kong revealed that slope deformation in several mountainous areas is closely associated with the occurrence of recorded landslides. By augmenting the landslide inventory using InSAR techniques, the proposed method enhanced the LSA models' capacity to identify slow-moving landslides and achieved better statistical performance. The discussion highlights that slope and stream power index (SPI) are the key landslide-predisposing factors in Hong Kong, but the dominant landslide-predisposing factors will vary under different geographical conditions. By comparison with the methods that treat LSA as a binary classification problem, such as support vector machine, multilayer perceptron, deep belief network, and random forest based LSA methods, the proposed method entails a fast-learning strategy and outperforms these methods in data-driven model evaluation indicators, e.g., by 3-6% in accuracy, 2-6% in precision, 1-2% in recall, 3-5% in F1-score, and approximately 10% in Cohen Kappa. The information about the relative importance of landslide predisposing factors, derived through feature permutation, can foster guidance for targeted landslide prevention schemes, such as constructing and maintaining slope consolidation facilities in areas where slope is the dominant landslide-predisposing factor.

A crosstalk mitigation scheme based on convolutional neural networks is proposed for MDM-PON. The simulation results show the proposed scheme can effectively compensate for channel impairment without requiring information from other modes.

Landslide susceptibility assessment (LSA) is of paramount importance in mitigating landslide risks. Recently, there has been a surge in the utilization of data-driven methods for predicting landslide susceptibility due to the growing availability of aerial and satellite data. Nonetheless, the rapid oscillations within the landslide-inducing environment (LIE), primarily due to significant changes in external triggers such as rainfall, pose difficulties for contemporary data-driven LSA methodologies to accommodate LIEs over diverse timespans. This study presents dynamic landslide susceptibility mapping that simply employs multiple predictive models for annual LSA. In practice, this will inevitably encounter small sample problems due to the limited number of landslide samples in certain years. Another concern arises owing to the majority of the existing LSA approaches train black-box models to fit distinct datasets, yet often failing in generalization and providing comprehensive explanations concerning the interactions between input features and predictions. Accordingly, we proposed to meta-learn representations with fast adaptation ability using a few samples and gradient updates; and apply SHAP for each model interpretation and landslide feature permutation. Additionally, we applied MT-InSAR for LSA result enhancement and validation. The chosen study area is Lantau Island, Hong Kong, where we conducted a comprehensive dynamic LSA spanning from 1992 to 2019. The model interpretation results demonstrate that the primary factors responsible for triggering landslides in Lantau Island are terrain slope and extreme rainfall. The results also indicate that the variation in landslide causes can be primarily attributed to extreme rainfall events, which result from global climate change, and the implementation of the Landslip Prevention and Mitigation Programme (LPMitP) by the Hong Kong government.

Predicting a landslide susceptibility map (LSM) is essential for risk recognition and disaster prevention. Despite the successful application of data-driven approaches for LSM prediction, most methods generally apply a single global model to predict the LSM for an entire target region. However, in large-scale areas with significant environmental change, various parts of the region hold different landslide-inducing environments, and therefore, should be predicted with respective models. This study first segmented target scenarios into blocks for individual analysis. Then, the critical problem is that in each block with limited samples, conducting training and testing a model is impossible for a satisfactory LSM prediction, especially in dangerous mountainous areas where landslide surveying is costly. To solve the problem, we trained an intermediate representation by the meta-learning paradigm, which is superior for capturing information valuable for few-shot adaption from LSM tasks. We hypothesized that there are more general and vital concepts concerning landslide causes and are sensitive to variations in input features. Thus, we can quickly adapt the models from the intermediate representation for different blocks or even unseen tasks using very few exemplar samples. Experimental results on the two study areas demonstrated the validity of our block-wise analysis in large scenarios and revealed the top few-shot adaption performances of the proposed methods.

Interleaved frequency division multiple access (IFDMA) is considered as a promising candidate for next-generation optical access networks. In this paper, we proposed a bidirectional long short-term memory (BiLSTM)-based detection method for IFDMA-PON. The performance of the BiLSTM-based method was evaluated with varying modulation formats and laser linewidths. The results showed that BiLSTM could detect system impairments more effectively with higher tolerance for nonlinear distortion of the transmitter than traditional least square-based methods.