Machine learning with applications最新文献

Machine learning models for near collision detection in autonomous vehicles promise enhanced predictive power. However, training on these large datasets presents storage and computational challenges, particularly when operated on conventional computing systems. This paper addresses the problem of training anomaly detection models from large-scale vehicle trajectory datasets and adopts a reservoir sampling-based data sketching technique. Predetermined subset sizes ranging from 0.4% to 100% of the original data are utilized, A single-pass reservoir sampling algorithm is then applied to construct these data subsets efficiently. Subsequently, a Support Vector Machine (SVM) model is trained on these subsets, and its performance is assessed by various metrics, including accuracy, precision, recall, and F1-score. Experimental outcomes on the HighD dataset, a comprehensive real-world collection of vehicle trajectories, confirm that our approach can achieve robust near-collision detection. With a full dataset, our model achieved an F1-score of 0.9998 for class 0 and 0.9984 for class 1. When the data was reduced to as low as 0.4% of the original size, the F1-score for class 0 remained at 0.9998 and 0.7143 for class 1. This demonstrates a capability to maintain a relatively high performance even with a 99.6% reduction in data size. Moreover, precision and recall values ranged from 71.3% to 0.999 across varying sketch sizes.

The as-rolled mechanical properties of microalloyed steels result from their chemical composition and thermomechanical processing history. Accurate predictions of the mechanical properties would reduce the need for expensive and time-consuming testing. At the same time, understanding the interplay between process variables and alloy composition will help reduce product variability and facilitate future alloy design. This paper provides an artificial neural network methodology to predict lower yield strength (LYS) and ultimate tensile strength (UTS). The proposed method uses feature engineering to transform raw data into features typically used in physical metallurgy to better utilize the artificial neural network model in understanding the process. SHAP values are used to reveal the effect of thermomechanical controlled processing, which can be rationalized by physical metallurgy theory.

Wearable biometric tracking devices are becoming increasingly common, providing users with physiological metrics such as heart rate variability (HRV) and skin conductance. We hypothesize that these metrics can be used as inputs for machine learning models to detect independent variables, such as target prevalence or hours awake, objective task performance, and metacognitive states. Over the course of 1–25 h awake, 40 participants completed four sessions of a simulated mine hunting task while non-invasive wearables collected physiological and behavioral data. The collected data were used to generate multiple machine learning models to detect the independent variables of the experiment (e.g., time awake and target prevalence), objective task performance, or metacognitive states. The strongest generated model was the time awake detection model (area under the curve = 0.92). All other models performed much closer to chance (area under the curve = 0.57–0.66), suggesting the model architecture used in this paper can detect time awake but falls short in other domains.

Clustering techniques are used to group observations and discover interesting patterns within data. Model-based clustering is one such method that is often an attractive choice due to the specification of a generative model for the given data and the ability to calculate model-selection criteria, which is in turn used to select the number of clusters. However, when only distances between observations are available, model-based clustering can no longer be used, and heuristic algorithms without the aforementioned advantages are usually used instead. As a solution, Oh and Raftery (2007) suggest a Bayesian model-based clustering method (named BMCD) that only requires a dissimilarity matrix as input, while also accounting for the measurement error that may be present within the observed data. In this paper, we extend the BMCD framework by proposing several additional models, alternative model selection criteria, and strategies for reducing computing time of the algorithm. These extensions ensure that the algorithm is effective even in high-dimensional spaces and provides a wide range of choices to the practitioner that can be used with a variety of data. Additionally, a publicly available software implementation of the algorithm is provided as a package in the R programming language.

The concept of the Butterfly Effect, derived from chaos theory, highlights how seemingly minor changes can lead to significant, unpredictable outcomes in complex systems. This phenomenon is particularly pertinent in the realm of AI fairness and bias. Factors such as subtle biases in initial data, deviations during algorithm training, or shifts in data distribution from training to testing can inadvertently lead to pronounced unfair results. These results often disproportionately impact marginalized groups, reinforcing existing societal inequities. Furthermore, the Butterfly Effect can magnify biases in data or algorithms, intensify feedback loops, and heighten susceptibility to adversarial attacks. Recognizing the complex interplay within AI systems and their societal ramifications, it is imperative to rigorously scrutinize any modifications in algorithms or data inputs for possible unintended effects. This paper proposes a combination of algorithmic and empirical methods to identify, measure, and counteract the Butterfly Effect in AI systems. Our approach underscores the necessity of confronting these challenges to foster equitable outcomes and ensure responsible AI evolution.

Corporate bankruptcy and financial distress prediction is a topic of interest for a variety of stakeholders, including businesses, financial institutions, investors, regulatory bodies, auditors, and academics. Various statistical and artificial intelligence methodologies have been devised to produce more accurate predictions. As more researchers are now focusing on this growing field of interest, this paper provides an up-to-date comprehensive survey, classification, and critical analysis of the literature on corporate bankruptcy and financial distress predictions, including definitions of bankruptcy and financial distress, prediction methodologies and models, data pre-processing, feature selection, model implementation, performance criteria and their measures for assessing the performance of classifiers or prediction models, and methodologies for the performance evaluation of prediction models. Finally, a critical analysis of the surveyed literature is provided to inspire possible future research directions.

Artificial intelligence (AI) has made remarkable strides, giving rise to the development of large language models such as ChatGPT. The chatbot has garnered significant attention from academia, industry, and the general public, marking the beginning of a new era in AI applications. This work explores how well ChatGPT can write source code. To this end, we performed a series of experiments to assess the extent to which ChatGPT is capable of solving general programming problems. Our objective is to assess ChatGPT’s capabilities in two different programming languages, namely C++ and Java, by providing it with a set of programming problem, encompassing various types and difficulty levels. We focus on evaluating ChatGPT’s performance in terms of code correctness, run-time efficiency, and memory usage. The experimental results show that, while ChatGPT is good at solving easy and medium programming problems written in C++ and Java, it encounters some difficulties with more complicated tasks in the two languages. Compared to code written by humans, the one generated by ChatGPT is of lower quality, with respect to runtime and memory usage.

Self-service shopping technologies have become commonplace in modern society. Although various innovative solutions have been adopted, there is still a gap in providing efficient services to consumers. Recent developments in mobile application technologies and internet-of-things devices promote information and knowledge dissemination by integrating innovative services to meet users’ needs. We argue that object retrieval applications can be used to provide effective online or self-service shopping. Therefore, to fill this technological void, this study aims to propose an object retrieval system using a fusion-based salient object detection (SOD) method. The SOD has attracted significant attention, and recently many heuristic computational models have been developed for object detection. It has been widely used in object detection and retrieval applications. This work proposes an instance retrieval system based on the SOD to find the objects from the commodity datasets. A prediction about the object’s position is made using the saliency detection system through a saliency model, and the proposed SOD-based retrieval (SODRet) framework uses saliency maps for retrieving the searched items. The method proposed in this work is evaluated on INSTRE and Flickr32 datasets. Our proposed work outperforms state-of-the-art object retrieval methods and can further be employed for large-scale self-service shopping-based points of sales.