Fine-grained Information Research Articles

Bootstrapping Automated Testing for RESTful Web Services

Modern RESTful services expose RESTful APIs to integrate with diversified applications. Most RESTful API parameters are weakly typed, which greatly increases the possible input value space. Weakly-typed parameters pose difficulties for automated testing tools to generate effective test cases to reveal web service defects related to parameter validation. We call this phenomenon the type collapse problem. To remedy this problem, we introduce FET (Format-encoded Type) techniques, including the FET, the FET lattice, and the FET inference to model fine-grained information for API parameters. Inferred FET can enhance parameter validation, such as generating a parameter validator for a certain RESTful server. Enhanced by FET techniques, automated testing tools can generate targeted test cases. We demonstrate Leif, a trace-driven fuzzing tool, as a proof-of-concept implementation of FET techniques. Experiment results on 27 commercial services show that FET inference precisely captures documented parameter definitions, which helps Leif discover 11 new bugs and reduce <inline-formula><tex-math notation="LaTeX">$72\% - 86\%$</tex-math></inline-formula> fuzzing time compared to state-of-the-art fuzzers. Leveraged by the inter-parameter dependency inference, Leif saves 15% fuzzing time.

Multi-class disease detection using deep learning and human brain medical imaging

Medical imaging and deep learning methods have significantly improved the early detection of brain diseases like tumors and Ischemic stroke with higher accuracy. Machine learning methods especially neural-network based algorithms have shown huge success in medical image analysis for variety of tasks including the detection, segmentation and classification of brain tumors and Ischemic stroke. Usually, these models address one problem at a time which is considered as Artificial Weak Intelligence (AWI). There is the need to develop methods that can push the research towards strong or Artificial General AI (AGI) where a single model can solve multiple tasks.In this work, we propose convolutional neural network based integrated model to detect and classify two brain diseases simultaneously i.e., tumors and Ischemic stroke. For this, a new dataset is created by merging two open-source datasets: BRATS 2015 and ISLES 2015. The designed network is an enhancement of encoder-decoder architecture based UNET where feature maps from one encoder block are fused with output of following encoder block to keep low-level fine-grained information intact and distinguish the overlapping features during encoding process in addition to UNET skip connections. The dataset is partitioned into training and validation sets on the ratio of 80:20 respectively with proportionate image inclusion in each training batch to address class imbalance issue. Average accuracy obtained through proposed model is 99.56%, 99.99% specificity, 99.59% precision and F1- score 99.57%. Obtained performance scores shows the usability of proposed feature fusion mechanism for multi-disease detection.

TransAM: Transformer appending matcher for few-shot knowledge graph completion

Few-shot knowledge graph completion (FSKGC) refers to predicting new facts for a new relation with only few-shot observed entity pairs (triples) as support set. Existing solutions to FSKGC mainly conduct the matching process over entity pair representations. Although effective, a major concern of these models is that the entity interactions are not fully explored, based on the observation that they usually generate the pair representation before the matching stage. Such a design inherently overlooks the fine-grained information from entity interactions, leading to performance decrements in one or three shot, which require matching models to capture more sufficient semantic meanings for prediction. To remedy this issue, in this paper, we explore the entity interactions within and between different instances, i.e., the co-occurrence of two entities, for FSKGC and propose our model named TransAM, Transformer Appending Matcher. TransAM solves the FSKGC problem by computing the probability of entity sequence with a well-designed transformer matching network. Specifically, TransAM appends query entity pair to serialized reference entity sequence and utilizes transformer to calculate the probability by capturing intra- and inter- triple entity interactions. To bridge the gap between transformer and the triple structure, TransAM introduces rotary operation to preserve the head and tail roles of entity within the triple and distinguishes different triples by a separated triple position encoding. Empirical studies on two public benchmark datasets NELL-One and Wiki-One show that TransAM outperforms existing metric-learning solutions in MRR and Hits@1 with both one- and three- shot settings, and achieves comparable results on five-shot setting. Datasets and code will be public available at https://github.com/gawainx/TransAM.

On the Identifiability of Cognitive Diagnostic Models: Diagnosing Students’ Translation Ability

Background. In recent years Cognitive Diagnostic Models (CDMs) have attracted a great deal of attention from researchers in a variety of educational fields. However, they have not been taken into consideration in Translation Quality Assessment (TQA), in the aims of presenting fine-grained information about the strengths and weaknesses of translation students.&#x0D; Purpose. The present study compares the ACDM, DINO, DINA, HO-DINA, and G-DINA models, in order to define the strengths and weaknesses of Iranian translation students and to examine whether the required translation attributes are compensatory, non-compensatory, additive, or hierarchical.&#x0D; Methods. 200 BA translation students translated a two-English-text translation, which was scored by three experienced translation raters using the Translation Quality Assessment Rubric (TQAR). The professional translators, established the relationships between the TQAR items and the nine proposed target translation attributes by constructing a Q-matrix.&#x0D; Results. Based on the results, HO-DINA can be considered the best-fitting model. Bibliography and technical skills, together with work methodology skills, are shown to be the most difficult attributes for translation students.&#x0D; Conclusion. HO-DINA is a non-compensatory model, thus the study findings assert that for a correct response to a test item, all measurable attributes need to be mastered.

Traffic event detection as a slot filling problem

Social media platforms, such as Twitter, can be used to extract information related to traffic events. Previous works focused mainly on classifying tweets into predefined categories (i.e., traffic or non-traffic) without many details of traffic events. However, extracting traffic-related fine-grained information from tweets is essential to build an intelligent transportation system. In this work, we address for the first time the problem of detecting traffic events using Twitter as two subtasks: (i) identifying whether a tweet is traffic-related or not as a text classification subtask, and (ii) extracting more fine-grained information (i.e., “what”, “when”, “where”, and the “consequence” of the traffic event) as a slot filling subtask. We also publish two Dutch Traffic Twitter datasets from Belgium and the Brussels capital region. We propose using deep learning based methods that process the two subtasks separately or jointly. Experimental results indicate that the proposed architectures achieve high performance scores (i.e., more than 95% F1 score) on the constructed datasets for both subtasks, even in a transfer learning scenario. In addition, incorporating tweet-level information in each of the tokens comprising the tweet (for the BERT-based model) can lead to a performance improvement for the joint setting. Our datasets and code are available on GitHub.11https://github.com/Glovesme/TrafficEventDetectionBE.

A Fine-grained Channel State Information-based Deep Learning System for Dynamic Gesture Recognition

Indoor gesture recognition technology is concerned with making the machine accurately recognize dynamic gestures within a certain range. Remarkably, most of this technology is based on passive recognition methods. This is quite striking because the high cost is a crucial factor in active recognition methods and ignoring this aspect can increase the reality gap. In this paper, we tend to use the fine-grained channel state information (CSI) in Wi-Fi to build a dynamic CNN-GRU-Attention (CGA) model to implement a gesture recognition system and thus alleviate this problem. Firstly, we study the influence of gestures on the amplitude and phase difference in CSI, and prove the feasibility of proposed method by analyzing the fluctuation of amplitude and phase difference under different conditions. Then, we use data processing methods such as phase correction and unwrapping with a new proposed adaptive gesture action truncation algorithm to extract the phase difference and remove redundant information, thus ensuring the validity of data. Finally, we propose to segment gesture fragment into 3-channel CSI images as input information of model. Extensive comparison experiments are conducted under the influence of different people, different indoor environments, and different sampling rates. The results show that the system has high accuracy.

Semantic-visual shared knowledge graph for zero-shot learning.

Almost all existing zero-shot learning methods work only on benchmark datasets (e.g., CUB, SUN, AwA, FLO and aPY) which have already provided pre-defined attributes for all the classes. These methods thus are hard to apply on real-world datasets (like ImageNet) since there are no such pre-defined attributes in the data environment. The latest works have explored to use semantic-rich knowledge graphs (such as WordNet) to substitute pre-defined attributes. However, these methods encounter a serious "role="presentation">domain shift" problem because such a knowledge graph cannot provide detailed enough semantics to describe fine-grained information. To this end, we propose a semantic-visual shared knowledge graph (SVKG) to enhance the detailed information for zero-shot learning. SVKG represents high-level information by using semantic embedding but describes fine-grained information by using visual features. These visual features can be directly extracted from real-world images to substitute pre-defined attributes. A multi-modals graph convolution network is also proposed to transfer SVKG into graph representations that can be used for downstream zero-shot learning tasks. Experimental results on the real-world datasets without pre-defined attributes demonstrate the effectiveness of our method and show the benefits of the proposed. Our method obtains a +2.8%, +0.5%, and +0.2% increase compared with the state-of-the-art in 2-hops, 3-hops, and all divisions relatively.

AA-trans: Core attention aggregating transformer with information entropy selector for fine-grained visual classification

The task of fine-grained visual classification (FGVC) is to distinguish targets from subordinate classifications. Since fine-grained images have the inherent characteristic of large inter-class variances and small intra-class variances, it is considered an extremely difficult task. Most existing approaches adopt CNN-based networks as feature extractors, which causes the extracted discriminative regions to contain most parts of the object in this way, thus failing to locate the really important parts. Recently, the vision transformer (ViT) has demonstrated its power on a wide range of image tasks, which uses an attention mechanism to capture global contextual information to establish a remote dependency on the target and thus extract more powerful features. Nevertheless, the ViT model still focuses more on global coarse-grained information rather than local fine-grained information, which may lead to its undesirable performance in fine-grained image classification. To this end, we redesigned an attention aggregating transformer (AA-Trans) to better capture minor differences among images by improving the ViT structure in this paper. In detail, we propose a core attention aggregator (CAA), which enables better information sharing between each transformer layer. Besides, we further propose an innovative information entropy selector (IES) to guide the network in acquiring discriminative parts of the image precisely. Extensive experiments show that our proposed model structure can achieve a new state-of-the-art performance on several mainstream datasets.

Semantic-embedding Guided Graph Network for cross-modal retrieval

Many methods focus on aligning image regions with the corresponding text fragments, and ignore that images contain fragments that cannot be expressed by texts. To fully express the information of images and prevent the performance degradation caused by fine-grained information deviating from the core meaning of images, we propose Semantic-embedding Guided Graph Network (SGGN) for cross-modal retrieval. It learns the detail representations of each modality, with an integrated semantic information, by guiding local fragments to capture the internal correlation of cross-modal data and effectively convey the information. To further bridge the semantic gap between different modalities, SGGN uses adversarial network to play a game, and uses graph aggregation network to absorb complementary information of neighbor samples. We evaluate our approach on two datasets. Our method (based on R@10) achieves 97.2% on Flickr30k dataset. On MS-COCO dataset, it reaches 99.2% using 1K test set and 92.0% using 5K test set.

Hypergraph and cross-attention-based unsupervised domain adaptation framework for cross-domain myocardial infarction localization

Solving individual differences between subjects is critical for the promotion of electrocardiogram (ECG) classification algorithms in the intelligent health monitoring industry. Popular inter-subject-based solutions usually require the manual labeling of heartbeats and frequent updating of the model for new subjects. To track these problems, we propose a hypergraph and cross-attention-based unsupervised domain adaptation (HGCA-UDA) framework for the myocardial infarction localization. Specifically, we first build a hypergraph-based dual-channel network, that can simultaneously learn specific feature representations from an ECG lead and disease category levels for samples from different domains. We then design a cross-attention module to align cross-domain locally similar samples. Subsequently, a domain alignment strategy based on the Wasserstein distance is proposed to align the global edge feature distribution. Finally, a pseudo-label generation scheme is proposed to further align fine-grained category information. We conduct extensive experiments on two public benchmark datasets (the Physikalisch-Technische Bundesanstalt (PTB) and PTB_XL database), and the results show that the proposed HGCR-UDA (with unlabeled patients) achieves comparable results compared with state-of-the-art inter-patient-based methods (with labeled patients) and has excellent applications prospects in the field of intelligent health monitoring.

Incomplete mixed data-driven outlier detection based on local–global neighborhood information

Outlier detection is a crucial task for identifying unexpected patterns, errors, and behaviors; therefore, maximizing the valuable information obtained from ubiquitous, incomplete, redundant, noisy, and mixed data poses a great challenge. To achieve efficient graph-based outlier detection, we enhance the connectivity between similar objects and weaken the connectivity between heterogeneous objects. The network structure proposed in this paper is called “an incomplete local and global neighborhood information (ILGNI) network.” In this network, incomplete mixed data can be exploited considering two aspects; single-attribute local information and multi-attribute global information. Specifically, we initially utilize unsupervised attribute reduction methods to improve data quality. Then, from the perspective of local and global information, we use the level of similarity of objects to design strong-neighborhood and weak-similarity relations to deal with incomplete data. On this basis, the topology reconstruction of a high number of fine-grained neighborhood information networks can be realized. Finally, the outlier scores are calculated considering the stationary distribution of the Markov random walk model over the ILGNI network. The experiments conducted on four real-telecom fraud datasets demonstrate that the proposed algorithm can achieve enhanced outlier detection performance with low time complexity. In addition, the proposed method can effectively mine the information obtained from incomplete data and has high applicability to feature-related and feature-independent datasets.

Classification of metastatic hepatic carcinoma and hepatocellular carcinoma lesions using contrast-enhanced CT based on EI-CNNet.

For hepatocellular carcinoma and metastatic hepatic carcinoma, imaging is one of the main diagnostic methods. In clinical practice, diagnosis mainly relied on experienced imaging physicians, which was inefficient and cannot met the demand for rapid and accurate diagnosis. Therefore, how to efficiently and accurately classify the two types of liver cancer based on imaging is an urgent problem to be solved at present. The purpose of this study was to use the deep learning classification model to help radiologists classify the single metastatic hepatic carcinoma and hepatocellular carcinoma based on the enhanced features of enhanced CT (Computer Tomography) portal phase images of the liver site. In this retrospective study, 52 patients with metastatic hepatic carcinoma and 50 patients with hepatocellular carcinoma were among the patients who underwent preoperative enhanced CT examinations from 2017-2020. A total of 565 CT slices from these patients were used to train and validate the classification network (EI-CNNet, training/validation: 452/113). First, the EI block was used to extract edge information from CT slices to enrich fine-grained information and classify them. Then, ROC (Receiver Operating Characteristic) curve was used to evaluate the performance, accuracy, and recall of the EI-CNNet. Finally, the classification results of EI-CNNet were compared with popular classification models. By utilizing 80% data for model training and 20% data for model validation, the average accuracy of this experiment was 98.2% ± 0.62 (mean ± standard deviation (SD)), the recall rate was 97.23%±2.77, the precision rate was 98.02%±2.07, the network parameters were 11.83 MB, and the validation time was 9.83s/sample. The classification accuracy was improved by 20.98% compared to the base CNN network and the validation time was10.38s/sample. Compared with other classification networks, the InceptionV3 network showed improved classification results, but the number of parameters was increased and the validation time was 33s/sample, and the classification accuracy was improved by 6.51% using this method. EI-CNNet demonstrated promised diagnostic performance and has potential to reduce the workload of radiologists and may help distinguish whether the tumor is primary or metastatic in time; otherwise, it may be missed or misjudged. This article is protected by copyright. All rights reserved.

A multi-resolution unet algorithm based on data augmentation and multi-center training for polyp automatic segmentation

In clinical practice, segmenting polyps from colonoscopy images plays an important role in the diagnosis and treatment of colorectal cancer since it provides valuable information. However, accurate polyp segmentation is full of changes due to the following reasons: (1) the small training datasets with a limited number of samples and the lack of data variability; (2) the same type of polyps with a variation in texture, size, and color; (3) the weak boundary between a polyp and its surrounding mucosa. To address these challenges, we propose a novel robust deep neural network based on data augmentation, called Robust Multi-center Multi-resolution Unet (RMMSUNet), for the polyp segmentation task. Data augmentation and Multi-center training are both utilized to increase the amount and diversity of training dataset. The new multi-resolution blocks make up for the lack of fine-grained information in U-Net, and ensures the generation of more accurate pixel-level segmentation prediction graphs. Region-based refinement is added as the post-processing for the network output, to correct some wrongly predicted pixels and further refine the segmentation results. Quantitative and qualitative evaluations on the challenging polyp dataset show that our RMMSUNet improves the segmentation accuracy significantly, when comparing to other SOTA algorithms.

A Jointly Guided Deep Network for Fine-Grained Cross-Modal Remote Sensing Text–Image Retrieval

Remote sensing (RS) cross-modal text–image retrieval has great application value in many fields such as military and civilian. Existing methods utilize the deep network to project the images and texts into a common space and measure the similarity. However, the majority of those methods only utilize the inter-modality information between different modalities, which ignores the rich semantic information within the specific modality. In addition, due to the complexity of the RS images, there exists a lot of interference relation information within the extracted representation from the original features. In this paper, we propose a jointly guided deep network for fine-grained cross-modal RS text–image retrieval. First, we capture the fine-grained semantic information within the specific modality and then guide the learning of another modality of representation, which can make full use of the intra- and inter-modality information. Second, to filter out the interference information within the representation extracted from the two modalities of data, we propose an interference filtration module based on the gated mechanism. According to our experimental results, significant improvements in terms of retrieval tasks can be achieved compared with state-of-the-art algorithms. The source code is available at https://github.com/CQULab/JGDN .

Dynamic heterogeneous graph representation learning with neighborhood type modeling

Graph representation learning aims to learn the representations of graph structured data in low-dimensional space, and has a wide range of applications in graph analysis tasks. Real-world networks are generally heterogeneous and dynamic, which contain multiple types of nodes and edges, and the graph may evolve at a high speed over time. The complex heterogeneous properties and rapidly evolving graph structures make it difficult to learn high-quality graph representations for dynamic heterogeneous graphs. Currently, studies concentrated on representation learning of temporal heterogeneous networks are insufficient. Existing methods either rely on meta-paths where the embedding quality heavily depending on experts’ selection, or use network snapshots where the fine-grained temporal information cannot be captured. In this paper, we propose a novel graph neural network model–node signature based Temporal Heterogeneous Graph Attention Network, termed as THGAT, for learning the representations of dynamic heterogeneous networks. THGAT improves the aggregation way of neighborhood information, and pays attention to the enlightenment of the importance of neighbor nodes by heterogeneous information and temporal information that cannot be ignored in the network. We also innovatively propose three node signature methods for encoding the heterogeneous information of the nodes and use the time encoding technique suitable for real-time networks to directly represent the temporal information, so as to overcome the limitations of existing methods. We conduct experiments on four real-world datasets, and the results demonstrate that THGAT improves the representation learning quality significantly, in aspects of link prediction, node classification, and node clustering, compared to the state-of-the-art methods. To make the work more complete, we also analyze the applicable scenarios of the three node signature methods through experiments, respectively.

3U: Joint Design of UAV-USV-UUV Networks for Cooperative Target Hunting

A swarm of unmanned vehicles can provide fine-grained spatial-temporal information acquisition and monitoring in comparison to a single agent, which is beneficial in terms of environment mapping, terrain exploration, and target hunting. However, the cooperation of single type of unmanned vehicles may be not qualified for fulfilling complex underwater tasks considering the motion constraints. In this paper, a joint design of the unmanned aerial/surface/underwater vehicle (UAV-USV-UUV) network, also referred to as 3U network, is proposed for cooperative underwater target hunting. We first introduce the advantages of this 3U heterogeneous system in multi-task cooperation and portray its system model. Moreover, we propose an energy-oriented target hunting model by jointly optimizing the UAV's position, the UUV's trajectory as well as their inter-connectivity. Finally, DQN algorithms are conceived to solve the proposed target hunting problem. Simulation results show the proposed scheme is suitable for underwater target hunting with a high success rate considering a trade-off between the system energy consumption and inter-connectivity.

The Value of Social Media Data in Fashion Forecasting

Problem definition: How to use social media to predict style color and jeans fit sales for a retailer. Academic/practical relevance: Neither retail practice nor the academic literature provides a method for using social media to predict style color and jeans fit sales for a retailer. We present and validate a systematic approach for doing that. Methodology: Demand forecasting in the fashion industry is challenging due to short product lifetimes, long manufacturing lead times, and constant innovation of fashion products. We investigate the value of social media information for color trends and jeans fit forecasting. We partner with three multinational retailers, two apparel and one footwear, and combine their proprietary data sets with web-crawled publicly available data on Twitter and the Google Search Volume Index. We implement a variety of machine learning models to develop forecasts that can be used in setting the initial shipment quantity for an item, arguably the most important decision for fashion retailers. Results: Our findings show that fine-grained social media information has significant predictive power in forecasting color and fit demands months in advance of the sales season, and therefore greatly helps in making the initial shipment quantity decision. The predictive power of including social media features, measured by the improvement of the out-of-sample mean absolute deviation over current practice ranges from 24% to 57%. Managerial implications: To our knowledge, this study is the first to explore and demonstrate the value of social media information in fashion demand forecasting in a way that is practical and operable for fashion retailers. With consistent results across all three retailers, we demonstrate the robustness of our findings over market and geographic heterogeneity, and different forecast horizons. Moreover, we discuss potential mechanisms that might be driving this significant predictive power. Our results suggest that changes in fashion demand are driven more by “bottom-up” changes in consumer preferences than by “top-down” influence from the fashion industry. Funding: This work was supported by Wharton School Fishman-Davidson Center for Service and Operations Management, the Wharton School Baker Retailing Center, and the Wharton School Risk Management Center Russell Ackoff Doctoral Student Fellowship. Supplemental Material: The online appendix is available at https://doi.org/10.1287/msom.2023.1193 .

Language-Augmented Pixel Embedding for Generalized Zero-Shot Learning

Zero-shot Learning (ZSL) aims to recognize novel classes through seen knowledge. The canonical approach to ZSL leverages a visual-to-semantic embedding to map the global features of an image sample to its semantic representation. These global features usually overlook the fine-grained information which is vital for knowledge transfer between seen and unseen classes, rendering these features sub-optimal for ZSL task, especially the more realistic Generalized Zero-shot Learning (GZSL) task where global features of similar classes could hardly be separated. To provide a remedy to this problem, we propose Language-Augmented Pixel Embedding (LAPE) that directly bridges the visual and semantic spaces in a pixel-based manner. To this end, we map the local features of each pixel to different attributes and then extract each semantic attribute from the corresponding pixel. However, the lack of pixel-level annotation conduces to an inefficient pixel-based knowledge transfer. To mitigate this dilemma, we adopt the text information of each attribute to augment the local features of image pixels which are related to the semantic attributes. Experiments on four ZSL benchmarks demonstrate that LAPE outperforms current state-of-the-art methods. Comprehensive ablation studies and analyses are provided to dissect what factors lead to this success.

A Deep Subdomain Adaptation Network With Attention Mechanism for Malware Variant Traffic Identification at an IoT Edge Gateway

The prevailing of malware variants in ubiquitous Internet of Things (IoT) devices causes enormous losses. Accurate and timely identification of malware variant traffic at an IoT edge gateway can effectively reduce the loss. TransNet, the state-of-the-art technology for malware variant traffic detection, considers only global domain adaptation and ignores the alignment of distributions between different subdomains, which fails to capture the fine-grained information of classification targets. Besides, TransNet converges very slowly, which may use up precious resources in IoT devices. This article proposes a deep subdomain adaptation network with attention mechanism (DSAN-AT) to accurately and efficiently identify malware variant traffic at an IoT edge gateway. DSAN-AT utilizes local maximum mean discrepancy (LMMD) to align the traffic feature distributions of subdomains in the source and target domains. It also exploits channel and spatial attention mechanisms to accelerate learning traffic features between different subdomains to save precious computing resources at the IoT edge gateway. Our experimental study demonstrates that DSAN-AT achieves an average accuracy of 97.15% (96.37% for TransNet) and converges fast without using a large target domain training data set. DSAN-AT has strong practicality for identifying malware variant traffic at an edge IoT gateway.

Hybrid multi-fluid-particle simulations of the cosmic neutrino background

Simulation of the cosmic clustering of massive neutrinos is a daunting task, due both to their large velocity dispersion and to their weak clustering power becoming swamped by Poisson shot noise. We present a new approach, the multi-fluid hybrid-neutrino simulation, which partitions the neutrino population into multiple flows, each of which is characterised by its initial momentum and treated as a separate fluid. These fluid flows respond initially linearly to nonlinear perturbations in the cold matter, but slowest flows are later converted to a particle realisation should their clustering power exceed some threshold. After outlining the multi-fluid description of neutrinos, we study the conversion of the individual flows into particles, in order to quantify transient errors, as well as to determine a set of criteria for particle conversion. Assembling our results into a total neutrino power spectrum, we demonstrate that our multi-fluid hybrid-neutrino simulation is convergent to < 3% if conversion happens at z = 19 and agrees with more expensive simulations in the literature for neutrino fractions as high as Ω νh 2 = 0.005. Moreover, our hybrid-neutrino approach retains fine-grained information about the neutrinos' momentum distribution. However, the momentum resolution is currently limited by free-streaming transients excited by missing information in the neutrino particle initialisation procedure, which restricts the particle conversion to z ≳ 19 if percent-level resolution is desired.