Search

Flash floods, as a type of devastating natural disasters, can cause significant damage to infrastructure, agriculture, and people’s livelihoods. Mapping flash flood susceptibility has long been an effective measure to help with the development of flash flood risk reduction and management strategies. Recent studies have shown that machine learning (ML) techniques perform better than traditional statistical and process-based models in estimating flash flood susceptibility. However, a major limitation of standard ML models is that they ignore the local geographic context where flash floods occur. To address this limitation, we developed a local Geographically Weighted Random Forest (GWRF) model and compared its performance against other global and local statistical and ML alternatives using an empirical flash floods model of Jiangxi Province, China.

Existing reference-free turn-level evaluation metrics for chatbots inadequately capture the interaction between the user and the system. Consequently, they often correlate poorly with human evaluations. To address this issue, we propose a novel model-agnostic approach that leverages Conditional Pointwise Mutual Information (C-PMI) to measure the turn-level interaction between the system and the user based on a given evaluation dimension. Experimental results on the widely used FED dialogue evaluation dataset demonstrate that our approach significantly improves the correlation with human judgment compared with existing evaluation systems. By replacing the negative log-likelihood-based scorer with our proposed C-PMI scorer, we achieve a relative 62.6% higher Spearman correlation on average for the FED evaluation metric. Our code is publicly available at https://github.com/renll/C-PMI., Comment: Published at ACL2023 DialDoc Workshop; Updated Results

Recent hybrid models combining Linear State Space Models (SSMs) with self-attention mechanisms have demonstrated impressive results across a range of sequence modeling tasks. However, current approaches apply attention modules statically and uniformly to all elements in the input sequences, leading to sub-optimal quality-efficiency trade-offs. To address this limitation, we introduce Sparse Modular Activation (SMA), a general mechanism enabling neural networks to sparsely and dynamically activate sub-modules for sequence elements in a differentiable manner. Through allowing each element to skip non-activated sub-modules, SMA reduces computation and memory consumption of neural networks at both training and inference stages. To validate the effectiveness of SMA on sequence modeling, we design a novel neural architecture, SeqBoat, which employs SMA to sparsely activate a Gated Attention Unit (GAU) based on the state representations learned from an SSM. By constraining the GAU to only conduct local attention on the activated inputs, SeqBoat can achieve linear inference complexity with theoretically infinite attention span, and provide substantially better quality-efficiency trade-off than the chunking-based models. With experiments on a wide range of tasks, including long sequence modeling, speech classification and language modeling, SeqBoat brings new state-of-the-art results among hybrid models with linear complexity, and reveals the amount of attention needed for each task through the learned sparse activation patterns. Our code is publicly available at https://github.com/renll/SeqBoat., Comment: Accepted by NeurIPS 2023. Camera-ready Version

Existing approaches to dialogue state tracking rely on pre-defined ontologies consisting of a set of all possible slot types and values. Though such approaches exhibit promising performance on single-domain benchmarks, they suffer from computational complexity that increases proportionally to the number of pre-defined slots that need tracking. This issue becomes more severe when it comes to multi-domain dialogues which include larger numbers of slots. In this paper, we investigate how to approach DST using a generation framework without the pre-defined ontology list. Given each turn of user utterance and system response, we directly generate a sequence of belief states by applying a hierarchical encoder-decoder structure. In this way, the computational complexity of our model will be a constant regardless of the number of pre-defined slots. Experiments on both the multi-domain and the single domain dialogue state tracking dataset show that our model not only scales easily with the increasing number of pre-defined domains and slots but also reaches the state-of-the-art performance.

Existing approaches to dialogue state tracking rely on pre-defined ontologies consisting of a set of all possible slot types and values. Though such approaches exhibit promising performance on single-domain benchmarks, they suffer from computational complexity that increases proportionally to the number of pre-defined slots that need tracking. This issue becomes more severe when it comes to multi-domain dialogues which include larger numbers of slots. In this paper, we investigate how to approach DST using a generation framework without the pre-defined ontology list. Given each turn of user utterance and system response, we directly generate a sequence of belief states by applying a hierarchical encoder-decoder structure. In this way, the computational complexity of our model will be a constant regardless of the number of pre-defined slots. Experiments on both the multi-domain and the single domain dialogue state tracking dataset show that our model not only scales easily with the increasing number of pre-defined domains and slots but also reaches the state-of-the-art performance.

Modern large-scale Pre-trained Language Models (PLMs) have achieved tremendous success on a wide range of downstream tasks. However, most of the LM pre-training objectives only focus on text reconstruction, but have not sought to learn latent-level interpretable representations of sentences. In this paper, we manage to push the language models to obtain a deeper understanding of sentences by proposing a new pre-training objective, Sparse Latent Typing, which enables the model to sparsely extract sentence-level keywords with diverse latent types. Experimental results show that our model is able to learn interpretable latent type categories in a self-supervised manner without using any external knowledge. Besides, the language model pre-trained with such an objective also significantly improves Information Extraction related downstream tasks in both supervised and few-shot settings. Our code is publicly available at: https://github.com/renll/SparseLT., Comment: EMNLP 2022 (Oral)

The term "flash drought"describes a type of drought with rapid onset and strong intensity, which is co-affected by both water-limited and energy-limited conditions. It has aroused widespread attention in related research communities due to its devastating impacts on agricultural production and natural systems. Based on a global reanalysis dataset, we identify flash droughts across China during 1979-2016 by focusing on the depletion rate of weekly soil moisture percentile. The relationship between the rate of intensification (RI) and nine related climate variables is constructed using three machine learning (ML) technologies, namely, multiple linear regression (MLR), long short-term memory (LSTM), and random forest (RF) models. On this basis, the capabilities of these algorithms in estimating RI and detecting droughts (flash droughts and traditional slowly evolving droughts) were analyzed. Results showed that the RF model achieved the highest skill in terms of RI estimation and flash drought identification among the three approaches. Spatially, the RF-based RI performed best in southeastern China, with an average CC of 0.90 and average RMSE of the 2.6 percentile per week, while poor performances were found in the Xinjiang region. For drought detection, all three ML technologies presented a better performance in monitoring flash droughts than in conventional slowly evolving droughts. Particularly, the probability of detection (POD), false alarm ratio (FAR), and critical success index (CSI) of flash drought derived from RF were 0.93, 0.15, and 0.80, respectively, indicating that RF technology is preferable in estimating the RI and monitoring flash droughts by considering multiple meteorological variable anomalies in adjacent weeks to drought onset. In terms of the meteorological driving mechanism of flash drought, the negative precipitation (P) anomalies and positive potential evapotranspiration (PET) anomalies exhibited a stronger synergistic effect on flash droughts

Soil moisture (SM) is a vital variable for controlling water and heat exchange between the atmosphere and land surface. Spatiotemporally continuous satellite-derived SM information is urgently needed for large-scale meteorological and hydrological applications. This study proposes the application of a spatial gap-filling method using an artificial neural network (ANN) to reconstruct missing records of daily surface SM from the Climate Change Initiative program of the European Space Agency (ESA CCI) in the growing seasons of 1982–2015 across China. Ten environmental variables were taken into consideration, including meteorological forcing, geographic and topographic features, vegetation conditions, and soil texture. The ANN-reconstructed SM and RF-reconstructed SM were validated using the ground-based observations and three global reanalysis SM datasets. The gap-filling results by the ANN model were compared to that by the original kriging (OK) model under three simulation scenarios: SM images with removed swaths, varying vegetation densities, and different percentages of data gaps. The results showed that the ANN-reconstructed SM was in good agreement with in-situ SM observations, with the CC more than 0.607 and the RMSE less than 0.074 m3/m3 over the northwestern and eastern parts of China. Compared with the original ESA CCI SM, the ANN model performed better than the OK model in reconstructing SM with absent swaths and in densely and sparsely vegetated regions. Specifically, the ANN-reconstructed SM provided a higher estimation accuracy in regions with low-density vegetation than in those with high-density vegetation. The weaker performance may due to the complex interactions between surface SM and various environmental variables underneath the dense vegetation. In addition, the ANN model significantly outperformed the OK model by accurately estimating SM when the percentage of data gaps succeeded 40%. Our study is valuable for providing an alternative reference t

Text-to-Graph extraction aims to automatically extract information graphs consisting of mentions and types from natural language texts. Existing approaches, such as table filling and pairwise scoring, have shown impressive performance on various information extraction tasks, but they are difficult to scale to datasets with longer input texts because of their second-order space/time complexities with respect to the input length. In this work, we propose a Hybrid Span Generator (HySPA) that invertibly maps the information graph to an alternating sequence of nodes and edge types, and directly generates such sequences via a hybrid span decoder which can decode both the spans and the types recurrently in linear time and space complexities. Extensive experiments on the ACE05 dataset show that our approach also significantly outperforms state-of-the-art on the joint entity and relation extraction task., Comment: Accepted by ACL 2021 Findings

Convolutional Neural Networks (CNN) has been widely applied in the realm of computer vision. However, given the fact that CNN models are translation invariant, they are not aware of the coordinate information of each pixel. Thus the generalization ability of CNN will be limited since the coordinate information is crucial for a model to learn affine transformations which directly operate on the coordinate of each pixel. In this project, we proposed a simple approach to incorporate the coordinate information to the CNN model through coordinate embedding. Our approach does not change the downstream model architecture and can be easily applied to the pre-trained models for the task like object detection. Our experiments on the German Traffic Sign Detection Benchmark show that our approach not only significantly improve the model performance but also have better robustness with respect to the affine transformation., Comment: 6 pages, 8 figures, Course Project for ECE271B

Natural gradient has been recently introduced to the field of boosting to enable the generic probabilistic predication capability. Natural gradient boosting shows promising performance improvements on small datasets due to better training dynamics, but it suffers from slow training speed overhead especially for large datasets. We present a replication study of NGBoost(Duan et al., 2019) training that carefully examines the impacts of key hyper-parameters under the circumstance of best-first decision tree learning. We find that with the regularization of leaf number clipping, the performance of NGBoost can be largely improved via a better choice of hyperparameters. Experiments show that our approach significantly beats the state-of-the-art performance on various kinds of datasets from the UCI Machine Learning Repository while still has up to 4.85x speed up compared with the original approach of NGBoost.

Existing approaches to dialogue state tracking rely on pre-defined ontologies consisting of a set of all possible slot types and values. Though such approaches exhibit promising performance on single-domain benchmarks, they suffer from computational complexity that increases proportionally to the number of pre-defined slots that need tracking. This issue becomes more severe when it comes to multi-domain dialogues which include larger numbers of slots. In this paper, we investigate how to approach DST using a generation framework without the pre-defined ontology list. Given each turn of user utterance and system response, we directly generate a sequence of belief states by applying a hierarchical encoder-decoder structure. In this way, the computational complexity of our model will be a constant regardless of the number of pre-defined slots. Experiments on both the multi-domain and the single domain dialogue state tracking dataset show that our model not only scales easily with the increasing number of pre-defined domains and slots but also reaches the state-of-the-art performance., Comment: The 2019 Conference on Empirical Methods in Natural Language Processing (EMNLP 2019); Updated empirical results

Dialogue state tracking is the core part of a spoken dialogue system. It estimates the beliefs of possible user's goals at every dialogue turn. However, for most current approaches, it's difficult to scale to large dialogue domains. They have one or more of following limitations: (a) Some models don't work in the situation where slot values in ontology changes dynamically; (b) The number of model parameters is proportional to the number of slots; (c) Some models extract features based on hand-crafted lexicons. To tackle these challenges, we propose StateNet, a universal dialogue state tracker. It is independent of the number of values, shares parameters across all slots, and uses pre-trained word vectors instead of explicit semantic dictionaries. Our experiments on two datasets show that our approach not only overcomes the limitations, but also significantly outperforms the performance of state-of-the-art approaches., Comment: EMNLP 2018 Oral

The capability of acquiring accurate and dense three-dimensional geospatial information that covers large survey areas rapidly enables airborne light detection and ranging (LiDAR) has become a powerful technology in numerous fields of geospatial applications and analysis. LiDAR data filtering is the first and essential step for digital elevation model generation, land cover classification, and object reconstruction. The morphological filtering approaches have the advantages of simple concepts and easy implementation, which are able to filter non-ground points effectively. However, the filtering quality of morphological approaches is sensitive to the structuring elements that are the key factors for the filtering success of mathematical operations. Aiming to deal with the dependence on the selection of structuring elements, this paper proposes a novel filter of LiDAR point clouds based on geodesic transformations of mathematical morphology. In comparison to traditional morphological transformations, the geodesic transformations only use the elementary structuring element and converge after a finite number of iterations. Therefore, this algorithm makes it unnecessary to select different window sizes or determine the maximum window size, which can enhance the robustness and automation for unknown environments. Experimental results indicate that the new filtering method has promising and competitive performance for diverse landscapes, which can effectively preserve terrain details and filter non-ground points in various complicated environments, OLD Department of GIS Technology

The capability of acquiring accurate and dense three-dimensional geospatial information that covers large survey areas rapidly enables airborne light detection and ranging (LiDAR) has become a powerful technology in numerous fields of geospatial applications and analysis. LiDAR data filtering is the first and essential step for digital elevation model generation, land cover classification, and object reconstruction. The morphological filtering approaches have the advantages of simple concepts and easy implementation, which are able to filter non-ground points effectively. However, the filtering quality of morphological approaches is sensitive to the structuring elements that are the key factors for the filtering success of mathematical operations. Aiming to deal with the dependence on the selection of structuring elements, this paper proposes a novel filter of LiDAR point clouds based on geodesic transformations of mathematical morphology. In comparison to traditional morphological transformations, the geodesic transformations only use the elementary structuring element and converge after a finite number of iterations. Therefore, this algorithm makes it unnecessary to select different window sizes or determine the maximum window size, which can enhance the robustness and automation for unknown environments. Experimental results indicate that the new filtering method has promising and competitive performance for diverse landscapes, which can effectively preserve terrain details and filter non-ground points in various complicated environments, OLD Department of GIS Technology

We propose the Recurrent Soft Attention Model, which integrates the visual attention from the original image to a LSTM memory cell through a down-sample network. The model recurrently transmits visual attention to the memory cells for glimpse mask generation, which is a more natural way for attention integration and exploitation in general object detection and recognition problem. We test our model under the metric of the top-1 accuracy on the CIFAR-10 dataset. The experiment shows that our down-sample network and feedback mechanism plays an effective role among the whole network structure., Comment: 5 pages, 4 figures

Considerable biases in precipitation simulations in climate models have required the adoption of delta-change approaches to construct future precipitation scenarios for hydrological climate change impact studies. However, different delta-change methods yield different future precipitation scenarios that might significantly affect the projected future streamflow. To assess these effects, two delta-change methods were compared: The simple delta-change (SDC) method with a constant scaling factor and the quantile-quantile delta-change (QQDC) method with a quantile mapping-based non-uniform delta factor. The Xinanjiang (XAJ) hydrological model was applied using historical climatic data and two future precipitation scenarios for streamflow simulations in the Pearl River basin, China. The results show that the two delta-change methods have significant influences on future precipitation and streamflow projections, and these impacts become more distinct at finer and extreme event time scales. For instance, the QQDC method projects the 20-year daily maximum precipitation to be 8.1-98.6% higher than the SDC method. Consequently, the XAJ model with the QQDC future precipitation produces the 20-year daily maximum streamflow to be approximately 7.0-65.0% higher than that using the SDC precipitation. It implies that future precipitation transformation methods are a source of uncertainty, affecting future discharge projections. Such uncertainty should be considered in water resources management and flood control strategies for future climate change adaptations. © IWA Publishing 2016.

Considerable biases in precipitation simulations in climate models have required the adoption of delta-change approaches to construct future precipitation scenarios for hydrological climate change impact studies. However, different delta-change methods yield different future precipitation scenarios that might significantly affect the projected future streamflow. To assess these effects, two delta-change methods were compared: The simple delta-change (SDC) method with a constant scaling factor and the quantile-quantile delta-change (QQDC) method with a quantile mapping-based non-uniform delta factor. The Xinanjiang (XAJ) hydrological model was applied using historical climatic data and two future precipitation scenarios for streamflow simulations in the Pearl River basin, China. The results show that the two delta-change methods have significant influences on future precipitation and streamflow projections, and these impacts become more distinct at finer and extreme event time scales. For instance, the QQDC method projects the 20-year daily maximum precipitation to be 8.1-98.6% higher than the SDC method. Consequently, the XAJ model with the QQDC future precipitation produces the 20-year daily maximum streamflow to be approximately 7.0-65.0% higher than that using the SDC precipitation. It implies that future precipitation transformation methods are a source of uncertainty, affecting future discharge projections. Such uncertainty should be considered in water resources management and flood control strategies for future climate change adaptations. © IWA Publishing 2016.

Considerable biases in precipitation simulations in climate models have required the adoption of delta-change approaches to construct future precipitation scenarios for hydrological climate change impact studies. However, different delta-change methods yield different future precipitation scenarios that might significantly affect the projected future streamflow. To assess these effects, two delta-change methods were compared: The simple delta-change (SDC) method with a constant scaling factor and the quantile-quantile delta-change (QQDC) method with a quantile mapping-based non-uniform delta factor. The Xinanjiang (XAJ) hydrological model was applied using historical climatic data and two future precipitation scenarios for streamflow simulations in the Pearl River basin, China. The results show that the two delta-change methods have significant influences on future precipitation and streamflow projections, and these impacts become more distinct at finer and extreme event time scales. For instance, the QQDC method projects the 20-year daily maximum precipitation to be 8.1-98.6% higher than the SDC method. Consequently, the XAJ model with the QQDC future precipitation produces the 20-year daily maximum streamflow to be approximately 7.0-65.0% higher than that using the SDC precipitation. It implies that future precipitation transformation methods are a source of uncertainty, affecting future discharge projections. Such uncertainty should be considered in water resources management and flood control strategies for future climate change adaptations.

In this study, a hydrological modelling framework was introduced to assess the climate change impacts on future river flow in the West River basin, China, especially on streamflow variability and extremes. The modelling framework includes a delta-change method with the quantile-mapping technique to construct future climate forcings on the basis of observed meteorological data and the downscaled climate model outputs. This method is able to retain the signals of extreme weather events, as projected by climate models, in the constructed future forcing scenarios. Fed with the historical and future forcing data, a large-scale hydrologic model (the Variable Infiltration Capacity model, VIC) was executed for streamflow simulations and projections at daily time scales. A bootstrapping resample approach was used as an indirect alternative to test the equality of means, standard deviations and the coefficients of variation for the baseline and future streamflow time series, and to assess the future changes in flood return levels. The West River basin case study confirms that the introduced modelling framework is an efficient effective tool to quantify streamflow variability and extremes in response to future climate change. Copyright © 2014 IAHS Press.