In recent years, pre-trained large language models (LLMs) have demonstrated remarkable efficiency in achieving an inference-time few-shot learning capability known as in-context learning. However, existing literature ...
In recent years, pre-trained large language models (LLMs) have demonstrated remarkable efficiency in achieving an inference-time few-shot learning capability known as in-context learning. However, existing literature has highlighted the sensitivity of this capability to the selection of few-shot demonstrations. Current understandings of the underlying mechanisms by which this capability arises from regular language model pretraining objectives remain disconnected from the real-world LLMs. This study aims to examine the in-context learning phenomenon through a Bayesian lens, viewing real-world LLMs as latent variable models. On this premise, we propose an algorithm to select optimal demonstrations from a set of annotated data with a small LM, and then directly generalize the selected demonstrations to larger LMs. We demonstrate significant improvement over baselines, averaged over eight GPT models on eight real-world text classification datasets. We also demonstrate the real-world usefulness of our algorithm on GSM8K, a math word problem dataset. Our empirical findings support our hypothesis that LLMs implicitly infer a latent variable containing task information. Code: https://***/WANGXinyiLinda/concept-based-demonstration-selection.
Technological developments have resulted in a trend of cryptocurrencies that use a technology called blockchain to create and record all transactions made into a digital ledger. Along with the emergence of the trend o...
详细信息
Human activity recognition (HAR) is the process of using mobile sensor data to determine the physical activities performed by individuals. HAR is the backbone of many mobile healthcare applications, such as passive he...
详细信息
The concept of object-oriented (OO) serves is a fundamental approach in the development of models. The stages associated with this method contribute significantly to ensuring that the resulting model is both lucid and...
The concept of object-oriented (OO) serves is a fundamental approach in the development of models. The stages associated with this method contribute significantly to ensuring that the resulting model is both lucid and transparent. The primary objective of the study is to create a decision model for evaluating student performance. Floating fuzzy logic (FFL) is employed as a technique to handle fluctuating data within the model. Moreover, OO conception plays a central role in analyzing, designing, and constructing the model through the utilization of four distinct types of Unified Modeling Language (UML) diagrams: object, activity, state-machine, and sequence diagrams. The model itself is crafted using the Python programming language and executed in the Google Colab platform. Additionally, this model has the capability to simulate changes in students' performance on a semester-by -semester basis, exhibiting a variance of 15 % when compared to the conventional fuzzy logic model.
In real life, many activities are performed sequentially. These activities must be carried out sequentially, such as the assembly process in the manufacturing production process. This series of activities cannot be re...
In real life, many activities are performed sequentially. These activities must be carried out sequentially, such as the assembly process in the manufacturing production process. This series of activities cannot be reduced or added so that the main goal of the series of activities is achieved. Apart from that, there are also time series events that occur naturally, such as rainy and hot conditions in a certain area. The classification process of time series activities is very important to see the possibility of anomalies occurring. The significant development of machine learning models in recent years has made the process of classifying time series data increasingly researched. Several previous studies stated that deep learning models were more accurate in classifying time series data. In this paper, we will compare Convolutional Neural Network (CNN) and Transformer deep learning models in classifying time series data. Experimental results using the same public datasets for CNN and Transformer model show that the CNN model is more accurate than the Transformer model. The results of measuring accuracy using confusion matrix show that CNN has an accuracy of 92% and Transformer has an accuracy of 80%.
Love is a central theme in modern music, but do women and men differ in their expressions of love? Results from empirical studies on gender differences in love attitudes have evolved from showing consistent difference...
详细信息
Data mining is an analytical process of knowledge discovery in large and complex data sets. Many studies wish to explore data, to find information so that knowledge can be obtained through the grouping process, classi...
详细信息
Insurance claim is a fascinating issue to study. A potential loss for the insurance company is major coming from this issue. Thus, many studies performed already to answer such an issue. This study takes aim to develo...
Insurance claim is a fascinating issue to study. A potential loss for the insurance company is major coming from this issue. Thus, many studies performed already to answer such an issue. This study takes aim to develop a computational decision model based on service technology. Four fundamental methods operated in constructing the model and its implementation in service technology. The analytical hierarchical process (AHP) and fuzzy logic methods are two methods benefited to construct the model; where the AHP used to prioritize thirteen parameters considered and the fuzzy logic with its inference capability operated to generate the decision. Object oriented is an analysis and design method to analyze and design the model implanting it in service oriented architecture (SOA). Then, SOA conception functioned to deploy the model in the service architecture. Ultimately, the suggested framework comprising three layers of service-oriented architecture (SOA), namely business process, service interface, and application, has been established, alongside the integration of eight essential services that connect these three applications. The model demonstrates simulation outcomes indicating that 31.47% of claims are categorized as low risk and have been approved, 17.64% of claims are considered moderate risk with currently pending decision status (requiring additional investigation), while 50.89% of claims are classified as high risk with also pending decision status.
Opening or closing dam-gate activities manually conducted in Manggarai dam to control the dam water level. The controlling action operated to avoid the flood possibility occurring in Jakarta city (the Indonesian capit...
Opening or closing dam-gate activities manually conducted in Manggarai dam to control the dam water level. The controlling action operated to avoid the flood possibility occurring in Jakarta city (the Indonesian capital). The study was conducted to develop a smart model for flood controlling based on service or called a service-oriented smart model (SOSM). The water-flow algorithm (WFA), fuzzy logic, object and service-oriented are four main methods operated in the study. The WFA is a central method to model the real water flow in the river coming from Katulampa dam (in Bogor city) until Manggarai dam (in Jakarta). The fuzzy logic functioned to simulate the dam’s water level and the gate open/close decision should be decided by avoiding the bias value. The object-oriented model analysis and design approach, where the unified modelling language (UML) tools are operated to analyze and design the constructed model. Then, the service-oriented conception is used to integrate all sides in implementing the model. Finally, the constructed model can simulate the flood status in Jakarta via status value in decimal numbers with 6 numbers behind the point.
The point of Agile Methodology is continuous improvement, delivering a small feature quickly without sacrificing the feature quality; every sprint must be better than the previous sprint, and better can be fewer bugs,...
The point of Agile Methodology is continuous improvement, delivering a small feature quickly without sacrificing the feature quality; every sprint must be better than the previous sprint, and better can be fewer bugs, faster development, and testing. We will present how we reduce production bugs by customizing our sprint iteration. As we know, bugs are unavoidable, there is no software engineer that can make software without a bug; however, we can reduce bugs in production if we can find bugs in lower environments as early as possible. The case study in this paper was taken from one of technology company in Indonesia, the activity was done by the Quality Engineer (QA) Team. We will show that shift-left testing can help us reduce bugs in production. Testing is part of agile methodology, and the main idea of shift-left testing is to move testing early and could be done by any team member, not only QA. We include shift-left testing in our agile methodology for one year in 2022 and compare the result in the previous year.
暂无评论