Eccentric decoupled charge(EDC)blasting is a widely used technique for rock fragmentation and tunnel excavation,yet the underlying rock damage mechanisms,particularly in relation to in-situ stresses and multi-borehole...Eccentric decoupled charge(EDC)blasting is a widely used technique for rock fragmentation and tunnel excavation,yet the underlying rock damage mechanisms,particularly in relation to in-situ stresses and multi-borehole combinations,remain underexplored.First,we developed an analytical model for single-borehole EDC blasting,providing insights into the theoretical relationship between the formation of different rock damage zones around the borehole and various influencing factors,including decoupling coefficient,in-situ stress,rock and explosive properties,and peak blast pressure.Using afinite elementfluid-solid coupling algorithm,we performed numerical simulations for a simple case of single-borehole EDC blasting,verifying the effectiveness of the adopted numerical approach.We then performed numerical simulations for dual-borehole EDC blasting with varying in-situ stress conditions and borehole combinations.The results indicate that:(1)rock damage is primarily concentrated on the eccentric side of the borehole due to its smaller decoupling coefficients and the resulting larger peak blast pressure;(2)the formation of through cracks between two boreholes is highly dependent on the relative angleφbetween them,while the extent and direction of the cracks are largely controlled by the application of in-situ stress.This work provides a theoretical basis and reference for optimizing the design of multi-borehole contour blasting in deep rock excavation under significant in-situ stresses,facilitating desired crack propagation while minimizing damage to the surrounding rock.展开更多
Large-scale pre-training has shown remarkable performance in building open-domain dialogue systems.However,previous works mainly focus on showing and evaluating the conversational performance of the released dialogue ...Large-scale pre-training has shown remarkable performance in building open-domain dialogue systems.However,previous works mainly focus on showing and evaluating the conversational performance of the released dialogue model,ignoring the discussion of some key factors towards a powerful human-like chatbot,especially in Chinese scenarios.In this paper,we conduct extensive experiments to investigate these under-explored factors,including data quality control,model architecture designs,training approaches,and decoding strategies.We propose EVA2.0,a large-scale pre-trained open-domain Chinese dialogue model with 2.8 billion parameters,and will make our models and codes publicly available.Automatic and human evaluations show that EVA2.0 significantly outperforms other open-source counterparts.We also discuss the limitations of this work by presenting some failure cases and pose some future research directions on large-scale Chinese open-domain dialogue systems.展开更多
Containerized microservices have been widely deployed in the industry.Meanwhile,security issues also arise.Many security enhancement mechanisms for containerized microservices require predefined rules and policies.How...Containerized microservices have been widely deployed in the industry.Meanwhile,security issues also arise.Many security enhancement mechanisms for containerized microservices require predefined rules and policies.However,it is challenging when it comes to thousands of microservices and a massive amount of real-time unstructured data.Hence,automatic policy generation becomes indispensable.In this paper,we focus on the automatic solution for the security problem:irregular traffic detection for RPCs.We propose Informer,a two-phase machine learning framework to track the traffic of each RPC and automatically report anomalous points.We first identify RPC chain patterns using density-based clustering techniques and build a graph for each critical pattern.Next,we solve the irregular RPC traffic detection problem as a prediction problem for attributed graphs with time series by leveraging spatial-temporal graph convolution networks.Since the framework builds multiple models and makes individual predictions for each RPC chain pattern,it can be efficiently updated upon legitimate changes in any graphs.In evaluations,we applied Informer to a dataset containing more than 7 billion lines of raw RPC logs sampled from a large Kubernetes system for two weeks.We provide two case studies of detected real-world threats.As a result,our framework found fine-grained RPC chain patterns and accurately captured the anomalies in a dynamic and complicated microservice production scenario,which demonstrates the effectiveness of Informer.Furthermore,we extensively evaluated the risk of adversarial attacks for our prediction model under different reality constraints and showed that the model is robust to such attacks in most real-world scenarios.展开更多
The springing up of large language models(LLMs)has shifted the community from single-task-orientated natural language processing(NLP)research to a holistic end-to-end multi-task learning paradigm.Along this line of re...The springing up of large language models(LLMs)has shifted the community from single-task-orientated natural language processing(NLP)research to a holistic end-to-end multi-task learning paradigm.Along this line of research endeavors in the area,LLM-based prompting methods have attracted much attention,partially due to the technological advantages brought by prompt engineering(PE)as well as the underlying NLP principles disclosed by various prompting methods.Traditional supervised learning usually requires training a model based on labeled data and then making predictions.In contrast,PE methods directly use the powerful capabilities of existing LLMs(e.g.,GPT-3 and GPT-4)via composing appropriate prompts,especially under few-shot or zero-shot scenarios.Facing the abundance of studies related to the prompting and the ever-evolving nature of this field,this article aims to 1)illustrate a novel perspective to review existing PE methods within the well-established communication theory framework,2)facilitate a better/deeper understanding of developing trends of existing PE methods used in three typical tasks,and 3)shed light on promising research directions for future PE methods.展开更多
基金supported by the National Natural Science Foundation of China(Grant No.42207177)the Natural Science Foundation of Shaanxi Province,China(Grant No.2022JQ-383)the Innovation Capacity Support Plan Project of Shaanxi Province,China(Grant No.2024RS-CXTD-49).
文摘Eccentric decoupled charge(EDC)blasting is a widely used technique for rock fragmentation and tunnel excavation,yet the underlying rock damage mechanisms,particularly in relation to in-situ stresses and multi-borehole combinations,remain underexplored.First,we developed an analytical model for single-borehole EDC blasting,providing insights into the theoretical relationship between the formation of different rock damage zones around the borehole and various influencing factors,including decoupling coefficient,in-situ stress,rock and explosive properties,and peak blast pressure.Using afinite elementfluid-solid coupling algorithm,we performed numerical simulations for a simple case of single-borehole EDC blasting,verifying the effectiveness of the adopted numerical approach.We then performed numerical simulations for dual-borehole EDC blasting with varying in-situ stress conditions and borehole combinations.The results indicate that:(1)rock damage is primarily concentrated on the eccentric side of the borehole due to its smaller decoupling coefficients and the resulting larger peak blast pressure;(2)the formation of through cracks between two boreholes is highly dependent on the relative angleφbetween them,while the extent and direction of the cracks are largely controlled by the application of in-situ stress.This work provides a theoretical basis and reference for optimizing the design of multi-borehole contour blasting in deep rock excavation under significant in-situ stresses,facilitating desired crack propagation while minimizing damage to the surrounding rock.
基金supported by the 2030 National Key AI Program of China(No.2021ZD0113304)the National Science Foundation for Distinguished Young Scholars(No.62125604)+2 种基金the NSFC projects(Key project with No.61936010 and regular project with No.61876096)the Guoqiang Institute of Tsinghua University,China(Nos.2019GQG1 and 2020GQG0005)Tsinghua-Toyota Joint Research Fund.
文摘Large-scale pre-training has shown remarkable performance in building open-domain dialogue systems.However,previous works mainly focus on showing and evaluating the conversational performance of the released dialogue model,ignoring the discussion of some key factors towards a powerful human-like chatbot,especially in Chinese scenarios.In this paper,we conduct extensive experiments to investigate these under-explored factors,including data quality control,model architecture designs,training approaches,and decoding strategies.We propose EVA2.0,a large-scale pre-trained open-domain Chinese dialogue model with 2.8 billion parameters,and will make our models and codes publicly available.Automatic and human evaluations show that EVA2.0 significantly outperforms other open-source counterparts.We also discuss the limitations of this work by presenting some failure cases and pose some future research directions on large-scale Chinese open-domain dialogue systems.
基金supported by the National Science Foundation of the United States under Grant Nos.1801751 and 1956364.
文摘Containerized microservices have been widely deployed in the industry.Meanwhile,security issues also arise.Many security enhancement mechanisms for containerized microservices require predefined rules and policies.However,it is challenging when it comes to thousands of microservices and a massive amount of real-time unstructured data.Hence,automatic policy generation becomes indispensable.In this paper,we focus on the automatic solution for the security problem:irregular traffic detection for RPCs.We propose Informer,a two-phase machine learning framework to track the traffic of each RPC and automatically report anomalous points.We first identify RPC chain patterns using density-based clustering techniques and build a graph for each critical pattern.Next,we solve the irregular RPC traffic detection problem as a prediction problem for attributed graphs with time series by leveraging spatial-temporal graph convolution networks.Since the framework builds multiple models and makes individual predictions for each RPC chain pattern,it can be efficiently updated upon legitimate changes in any graphs.In evaluations,we applied Informer to a dataset containing more than 7 billion lines of raw RPC logs sampled from a large Kubernetes system for two weeks.We provide two case studies of detected real-world threats.As a result,our framework found fine-grained RPC chain patterns and accurately captured the anomalies in a dynamic and complicated microservice production scenario,which demonstrates the effectiveness of Informer.Furthermore,we extensively evaluated the risk of adversarial attacks for our prediction model under different reality constraints and showed that the model is robust to such attacks in most real-world scenarios.
文摘The springing up of large language models(LLMs)has shifted the community from single-task-orientated natural language processing(NLP)research to a holistic end-to-end multi-task learning paradigm.Along this line of research endeavors in the area,LLM-based prompting methods have attracted much attention,partially due to the technological advantages brought by prompt engineering(PE)as well as the underlying NLP principles disclosed by various prompting methods.Traditional supervised learning usually requires training a model based on labeled data and then making predictions.In contrast,PE methods directly use the powerful capabilities of existing LLMs(e.g.,GPT-3 and GPT-4)via composing appropriate prompts,especially under few-shot or zero-shot scenarios.Facing the abundance of studies related to the prompting and the ever-evolving nature of this field,this article aims to 1)illustrate a novel perspective to review existing PE methods within the well-established communication theory framework,2)facilitate a better/deeper understanding of developing trends of existing PE methods used in three typical tasks,and 3)shed light on promising research directions for future PE methods.
