Previous studies on human activity recognition (HAR) using mmWave radar in IoT scenarios have relied on complex network architectures and heavy preprocessing, making them computationally expensive and difficult to deploy on resource-constrained edge devices (e.g., smart homes, healthcare). To address this, we propose LWDSC-Net, a lightweight wavelet dual-branch coupling network that embodies a novel, physically-interpretable design paradigm. Inspired by classic signal processing, LWDSC-Net embeds a 1D Haar wavelet transform (WT) as a front-end layer, enabling an end-to-end, structured decoupling of raw signals to construct a dual-branch structure with low and high frequencies (LF & HF) without external manual preprocessing. This signal-architecture co-design dedicates each branch to its respective frequency components, with the multi-level WT concept further extended to the HF branch. For the LF component branch, we propose an improved lightweight temporal convolutional network (LightTCN), which builds the feed-forward network using depthwise convolution. LightTCN reduces the depth of module stacking and efficiently models long-term global dependencies across the entire action sequence. For the HF components branch, we propose the wavelet-domain depthwise separable convolution (WD-DSC) module. It enhances micro-motion features by combining secondary wavelet reconstruction with wavelet-domain convolution, and further refines local detail information through deep convolutional layers. Experimental results show that LWDSC-Net achieves an accuracy of 98.76% on the public dataset while significantly reducing the number of model parameters and computational complexity. More importantly, deployment on an NVIDIA Jetson Nano confirms its excellent real-world performance, featuring ultra-low latency and power consumption, which highlights its outstanding potential for edge deployment.

Recent research into solving vehicle routing problems (VRPs) has gained significant traction, particularly through the application of deep (reinforcement) learning for end-to-end solution construction. However, many current construction-based neural solvers predominantly utilize Transformer architectures, which can face scalability challenges and struggle to produce diverse solutions. To address these limitations, we introduce a novel framework beyond Transformer-based approaches, i.e., Adversarial Generative Flow Networks (AGFN). This framework integrates the generative flow network (GFlowNet)—a probabilistic model inherently adept at generating diverse solutions (routes)—with a complementary model for discriminating (or evaluating) the solutions. These models are trained alternately in an adversarial manner to improve the overall solution quality, followed by a proposed hybrid decoding method to construct the solution. We apply the AGFN framework to solve the capacitated vehicle routing problem (CVRP) and travelling salesman problem (TSP), and our experimental results demonstrate that AGFN surpasses the popular construction-based neural solvers, showcasing strong generalization capabilities on synthetic and real-world benchmark instances.

Self-supervised monocular methods can efficiently learn depth information of weakly textured surfaces or reflective objects. However, the depth accuracy is limited due to the inherent ambiguity in monocular geometric modeling. In contrast, multi-frame depth estimation methods improve depth accuracy thanks to the success of Multi-View Stereo (MVS), which directly makes use of geometric constraints. Unfortunately, MVS often suffers from texture-less regions, non-Lambertian surfaces, and moving objects, especially in real-world video sequences without known camera motion and depth supervision. Therefore, we propose MOVEDepth, which exploits the MOnocular cues and VElocity guidance to improve multi-frame Depth learning. Unlike existing methods that enforce consistency between MVS depth and monocular depth, MOVEDepth boosts multi-frame depth learning by directly addressing the inherent problems of MVS. The key of our approach is to utilize monocular depth as a geometric priority to construct MVS cost volume, and adjust depth candidates of cost volume under the guidance of predicted camera velocity. We further fuse monocular depth and MVS depth by learning uncertainty in the cost volume, which results in a robust depth estimation against ambiguity in multi-view geometry. Extensive experiments show MOVEDepth achieves state-of-the-art performance: Compared with Monodepth2 and PackNet, our method relatively improves the depth accuracy by 20% and 19.8% on the KITTI benchmark. MOVEDepth also generalizes to the more challenging DDAD benchmark, relatively outperforming ManyDepth by 7.2%. The code is available at https://github.com/JeffWang987/MOVEDepth.

Semantic occupancy perception is essential for autonomous driving, as automated vehicles require a fine-grained perception of the 3D urban structures. However, existing relevant benchmarks lack diversity in urban scenes, and they only evaluate front-view predictions. Towards a comprehensive benchmarking of surrounding perception algorithms, we propose OpenOccupancy, which is the first surrounding semantic occupancy perception benchmark. In the OpenOccupancy benchmark, we extend the large-scale nuScenes dataset with dense semantic occupancy annotations. Previous annotations rely on LiDAR points superimposition, where some occupancy labels are missed due to sparse LiDAR channels. To mitigate the problem, we introduce the Augmenting And Purifying (AAP) pipeline to ~ 2× densify the annotations, where ∼4000 human hours are involved in the labeling process. Besides, camera-based, LiDAR-based and multi-modal baselines are established for the OpenOccupancy benchmark. Furthermore, considering the complexity of surrounding occupancy perception lies in the computational burden of high-resolution 3D predictions, we propose the Cascade Occupancy Network (CONet) to refine the coarse prediction, which relatively enhances the performance by ∼30% than the baseline. We hope the OpenOccupancy benchmark‡will boost the development of surrounding occupancy perception algorithms.

We present GradMask, a simple adversarial example detection scheme for natural language processing (NLP) models. It uses gradient signals to detect adversarially perturbed tokens in an input sequence and occludes such tokens by a masking process. GradMask provides several advantages over existing methods including improved detection performance and an interpretation of its decision with a only moderate computational cost. Its approximated inference cost is no more than a single forward- and back-propagation through the target model without requiring any additional detection module. Extensive evaluation on widely adopted NLP benchmark datasets demonstrates the efficiency and effectiveness of GradMask. Code and models are available at https://github.com/Han8931/grad_mask_detection

Identifying new gene functions and pathways underlying diseases and biological processes are major challenges in genomics research. Particularly, most methods for interpreting the pathways characteristic of an experimental gene list defined by genomic data are limited by their dependence on assessing the overlapping genes or their interactome topology, which cannot account for the variety of functional relations. This is particularly problematic for pathway discovery from single-cell genomics with low gene coverage or interpreting complex pathway changes such as during change of cell states. Here, we exploited the comprehensive sets of molecular concepts that combine ontologies, pathways, interactions and domains to help inform the functional relations. We first developed a universal concept signature (uniConSig) analysis for genome-wide quantification of new gene functions underlying biological or pathological processes based on the signature molecular concepts computed from known functional gene lists. We then further developed a novel concept signature enrichment analysis (CSEA) for deep functional assessment of the pathways enriched in an experimental gene list. This method is grounded on the framework of shared concept signatures between gene sets at multiple functional levels, thus overcoming the limitations of the current methods. Through meta-analysis of transcriptomic data sets of cancer cell line models and single hematopoietic stem cells, we demonstrate the broad applications of CSEA on pathway discovery from gene expression and single-cell transcriptomic data sets for genetic perturbations and change of cell states, which complements the current modalities. The R modules for uniConSig analysis and CSEA are available through https://github.com/wangxlab/uniConSig.

Sample-based evolutionary algorithms (EAs) are widely used for optimizing problems with multi (greater than one but less than four) or even many (greater than or equal to four) objectives of interest. In general, the difficulty of a problem exponentially increases with the number of objectives, serving as a clear example of the curse of dimensionality. The exploratory approach an EA takes in these cases has led to it being thought of as a big data generator, progressively sampling and evaluating solutions in high performing regions of a decision space to guide the search towards optimal solutions. Notably, in both multi- and many-objective EAs, the sampled data can be further utilized for building inverse generative models, mapping points in objective space back to solutions in the decision space. Such models offer immense flexibility to a decision maker in generating new target solutions on the fly, thereby facilitating real-time a posteriori preference incorporation into the search. In this paper, we show that the data distribution resulting from a many-objective formulation is in fact more conducive to building accurate inverse models than its multiobjective counterpart. Given the potential utility of these models, we in turn shed light on a rare blessing of dimensionality that is yet to be explored in the context of optimization. We first present simple theoretical arguments supporting our claim. Thereafter, experimental studies of Gaussian process-based inverse modeling for a synthetic and a real-world example are carried out to further confirm the theory.

Transfer optimization is an emerging concept that promises to enhance productivity of planning and decision-making processes by allowing for the adaptive reuse of knowledge (data) drawn from various “source” problems in a related ongoing “target” task of interest. Despite the recent advances in transfer optimization, however, a continuing challenge is the scalability of associated algorithms given big data of source problem instances. This paper tackles the scaling problem of an online adaptive knowledge transfer framework under big source data. We propose an efficient source selection algorithm based on the theory of multi-armed bandits such that the most related source task to the target is chosen for knowledge transfer, as opposed to extracting knowledge from all sources simultaneously. For this purpose, we introduce a novel and principled reward measure to reflect the source-target similarities. The efficacy of our proposed approach is assessed on the well-known knapsack problem that has practical implications in optimization of supply chain and manufacturing processes. Extensive experiments are conducted under big data of source problem instances. The numerical results clearly reveal that the incorporation of the proposed source selection mechanism in the existing adaptive knowledge transfer framework makes it successfully feasible for fast/real-time decision-making in the big data source setting.

CIA (Confidentiality, Integrity and Availability) is getting more and more important in the cloud based manufacturing and IIoT (Industrial Internet-of-Things)/Indutry 4.0 domain. Recently, blockchain, a decentralized ledger technique, has attracted considerable attention to realize secure and robust data management systems for IIoT. Although several blockchain-based data management platforms for IIoT have been proposed, sensor data are stored outside of the blockchain, meaning that the data integrity and accessibility are not improved. In this paper, we take a different blockchain-based approach for sensor data management; any important sensor data are stored in the blockchain. For realizing this, a series of sensor data is compressed-then-stored in the blockchain by leveraging a fact that many sensor data often have a certain level of periodicity and stability. We have tested our idea against the real sensor data measured at our model factory, and evaluated several performance metrics such as compression ratio, compression/decompression time, process time required to store and retrieve sensor data from our Proof-of-Concept system with Ethereum and the expected blockchain size. From the results, it can be concluded that the proposed platform is viable for small factory cases but needs more technological advancement is required for large scale cases.

Han Cheol Moon, Tasnim Mohiuddin, Shafiq Joty, Chi Xu. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019.

With the increasing adoption of innovative business practices the product lifecycle management is becoming incredibly volatile and challenging. In order to reduce the transportation time many supply chain managers are relying on express air freight based transport services. With the increasing average demand for capacity provided by the air freight service providers, various business entities are appearing in the value chain, such as service providers, integrators. Typically, an integrator owns airbuses to provide service to their clients as well as they can buy extra capacities from other commercial airlines based on need. As per the state of the art practices the capacity for the long term clients are provided through a deterministic model and priced through a bid based mechanism which maximizes the revenue. The demand that realizes close to the time of actual shipment is fulfilled through allocating the capacity through spot market. The spot market capacity allocation is performed based on a popular newsvendor framework. This paper proposes a holistic framework to develop a stochastic model to allocate the capacity for different types of capacity. The objective is to maximize the expected revenue with uncertainties in the system. This problem is fundamentally a dynamic programming problem. We implement an affine controller to develop a computationally tractable formulation of this dynamic programming problem. We conduct experiments from real life data obtained from a global player in this industry to demonstrate the superiority of the proposed model, over the state of the art practices.

As the global energy demand is increasing, the share of renewable energy and specifically wind energy in the supply is growing. While vast literature exists on the design and operation of wind turbines, there exists a gap in the literature with regards to the investigation and analysis of wind turbine accidents. This paper describes the application of text mining and machine learning techniques for discovering actionable insights and knowledge from news articles on wind turbine accidents. The applied analysis methods are text processing, clustering, and multidimensional scaling (MDS). These methods have been combined under a single analysis framework, and new insights have been discovered for the domain. The results of our research can be used by wind turbine manufacturers, engineering companies, insurance companies, and government institutions to address problem areas and enhance systems and processes throughout the wind energy value chain.

With more proliferation of services and higher degree of personalization, higher accurate approaches to service recommendation are becoming more and more pivotal. Performance of existing service recommendation approaches is not satisfactory due to the sparseness of available data set or the incomplete information of the global service market, which make it difficult to identify a customer's potential preferences on available services. In this paper, we extract finegrained value features from customer reviews, and identify the personalized distribution of each value features to demonstrate the value preference of a specific customer. Then, a novel recommendation algorithm (VFDSR) is proposed. An algorithm VFMine based on text mining is presented to effectively extract value features from customer reviews. A VFDAnalysis algorithm based on sentiment analysis is employed to identify the value feature distributions. Based on it, VFDSR recommends top-satisfying services to customers. In addition, the value feature distributions are visualized in the form of "heatmaps". Comprehensive experiments are conducted on a Yelp dataset and the experimental results show the superiority of our approach.

A schedule-based system is a system that operates on or contains within a schedule of events and breaks at particular time intervals. Entities within the system show presence or absence in these events by entering or exiting the locations of the events. Given radio frequency identification (RFID) data from a schedule-based system, what can we learn about the system (the events and entities) through data mining? Which data mining methods can be applied so that one can obtain rich actionable insights regarding the system and the domain? The research goal of this paper is to answer these posed research questions, through the development of a framework that systematically produces actionable insights for a given schedule-based system. We show that through integrating appropriate data mining methodologies as a unified framework, one can obtain many insights from even a very simple RFID dataset, which contains only very few fields. The developed framework is general, and is applicable to any schedule-based system, as long as it operates under certain basic assumptions. The types of insights are also general, and are formulated in this paper in the most abstract way. The applicability of the developed framework is illustrated through a case study, where real world data from a schedule-based system is analyzed using the introduced framework. Insights obtained include the profiling of entities and events, the interactions between entity and events, and the relations between events.

It is a challenge to visualize high dimensional data such as project data to yield new and interesting types of insights. To address this, we augment the traditional PERT network diagram with additional nodes that represent resources, and with arcs from the resource nodes to the activities that use those resources. Subsequently, we apply various graph layout algorithms that can reveal the hidden patterns in the graph data. Finally, we also map various attributes of the activities to the features of activity nodes. We illustrate the applicability and usefulness of our methodology through two case studies, where we visualize data from a benchmark data library and from the real world.

It is a challenge to estimate expected benefits from recommender systems based on association rule mining. This paper aims to address this challenge and presents a study of buying preferences of a sample of retail customers. It reveals a monotonic, non-linear relationship between the expected profits (as a function of information loss) and minimum support threshold levels, when considering transactions for a recommender system based on association rules. This finding is significant for recommender systems that utilize potential profits as a decision-making criterion.