Tag Performance Research Articles

Numerical Modeling, Analysis, and Optimization of RFID Tags Functioning at Low Frequencies

RFID (radio frequency identification) tags play a crucial role in a wide range of applications, from wireless communications to personal tracking and smart city infrastructure. These tags come in various shapes and sizes, prompting the authors to review the specialized literature and focus on optimizing planar designs with different geometries. This study prioritizes reducing the size of the most commonly used tags while enhancing their reliability. The primary objective of this article is to understand and improve the performance of planar RFID tags operating at 13.56 MHz through numerical simulations based on structures generated by algorithms developed in MATLAB. Building on previous research, the methodology is validated, followed by a detailed description of the algorithm designed and implemented by the authors in MATLAB to identify all possible structures that meet the design criteria. The authors compared various analyzed structures, considering different inductor shapes, dielectric materials, and thicknesses while examining their effects on gain and resonant frequency. The study also provides thermal analysis of the structures, and experimental validation of the studied designs. Finally, the researchers conclude with recommendations on the optimal structure for RFID tags.

Optimizing Passive Tag Performance With Reconfigurable Intelligent Surfaces in Bistatic Backscatter Networks

Optimizing Passive Tag Performance With Reconfigurable Intelligent Surfaces in Bistatic Backscatter Networks

The answer, my friend, is blowin' in the wind: Blow sampling provides a new dimension to whale population monitoring.

Marine mammals play a fundamental role in the functioning of healthy marine ecosystems and are important indicator species. Studying their biology, distributions, behaviour and health are still technically and logistically demanding for researchers. However, the efforts and commitment have not been in vain, since we are witnessing constant and exponential advancement in the study of these animals, thanks to technological progress in numerous fields. These include miniaturization and performance of biologger tags, which are equipped with sensors for measuring physiological parameters, hydrophones, accelerometers, time-depth records and spatial locations; the use of high throughput 'Next Generation' Sequencing to gain genetic information about communities and individual species from nucleic acids in environmental samples at miniscule concentrations; through, to the possibility of monitoring species with autonomous aerial and underwater vehicles. In parallel advances in computing and statistical modelling frameworks support the analysis of increasingly large and complex data sets. In this issue, O'Mahony etal.(2024) draw from at least two of these innovations: (a) the collection of biological material retrieved from large whales' blows using a modified drone and (b) the use of the samples to infer a wide spectrum of genetic information (both nuclear and mitochondrial) about the target animal/population. The methodology is not completely novel, but the study shows an impressive advancement in the amount of data obtained compared to preceding studies using the same approach. In the wake of these promising results, future perspectives are evaluated in relation to alternative sampling methodologies currently in use. It is possible to speculate that, in the next few years, the combination of non-invasive molecular profiling and enhanced drone technology (e.g. assembling increasingly smaller components, thus expanding capacity for autonomous operation) will open up perspectives that were unimaginable at the beginning of this millennium.

Is infrared-collinear safe information all you need for jet classification?

Machine learning-based jet classifiers are able to achieve impressive tagging performance in a variety of applications in high-energy and nuclear physics. However, it remains unclear in many cases which aspects of jets give rise to this discriminating power, and whether jet observables that are tractable in perturbative QCD such as those obeying infrared-collinear (IRC) safety serve as sufficient inputs. In this article, we introduce a new classifier, Jet Flow Networks (JFNs), in an effort to address the question of whether IRC unsafe information provides additional discriminating power in jet classification. JFNs are permutation-invariant neural networks (deep sets) that take as input the kinematic information of reconstructed subjets. The subjet radius and a cut on the subjet’s transverse momenta serve as tunable hyperparameters enabling a controllable sensitivity to soft emissions and nonperturbative effects. We demonstrate the performance of JFNs for quark vs. gluon and Z vs. QCD jet tagging. For small subjet radii and transverse momentum cuts, the performance of JFNs is equivalent to the IRC-unsafe Particle Flow Networks (PFNs), demonstrating that infrared-collinear unsafe information is not necessary to achieve strong discrimination for both cases. As the subjet radius is increased, the performance of the JFNs remains essentially unchanged until physical thresholds that we identify are crossed. For relatively large subjet radii, we show that the JFNs may offer an increased model independence with a modest tradeoff in performance compared to classifiers that use the full particle information of the jet. These results shed new light on how machines learn patterns in high-energy physics data.

Neural networks for boosted di-τ identification

We train several neural networks and boosted decision trees to discriminate fully-hadronic boosted di-τ topologies against background QCD jets, using calorimeter and tracking information. Boosted di-τ topologies consisting of a pair of highly collimated τ-leptons, arise from the decay of a highly energetic Standard Model Higgs or Z boson or from particles beyond the Standard Model. We compare the tagging performance for different neural-network models and a boosted decision tree, the latter serving as a simple benchmark machine learning model. The code used to obtain the results presented in this paper is available on GitHub.

Analysis of Reader Orientation on Detection Performance of Hilbert Curve-based Fractal Chipless RFID Tags

The role of the orientation of the radio frequency identification (RFID) reader is vital in the RFID-based communication system. This study presents the design and analysis of the impact of the orientation of the RFID reader (angle of incidence of the plane wave) on the detection and sensitivity characteristics of fractal chipless RFID tags. Four fractal (irregular) shaped tags are developed using four iterations of the Hilbert curve filling algorithm. The full-wave EM analysis of the designed tags in Matlab is performed by exporting them in Computer Simulation Technologies Micro-Wave Studio (CST MWS) in the frequency range of 2 to 20 GHz. Firstly, the performance of the tags is analyzed by observing the radar cross-section (RCS) of the tag for the fixed orientation of the incident plane wave in three different polarizations (horizontal, vertical, and oblique). Later, the variations in the EM spectrum (RCS results) are analyzed for oblique polarization by varying the incidence plane wave in both elevations (for two cases of 0∘ and 90∘) and azimuth planes (sweeping from 0∘ to 180∘ with a step size of 10∘). The analysis of the proposed aggregated RCS response for all cases in oblique polarization produces higher coding capacity (169 bits), coding spatial capacity (16.504 bits/cm2), coding spectral capacity (9.826 bits/GHz), and coding density (0.960 bits/GHz/cm2) for the realized highly irregular tag using fourth-iteration (4T) of Hilbert curve filling algorithm. The proposed procedure of detection based on aggregated response makes the developed RFID communication system more secure and reliable.

High sensitivity profiling of N-glycans from mouse serum using fluorescent imidazolium tags by HILIC electrospray ionisation spectrometry

N-linked glycosylation is a ubiquitous protein post-translational modification in which aberrant glycan biosynthesis has been linked to severe conditions like cancer. Accurate qualitative and quantitative analysis of N-glycans are crucial for investigating their physiological functions. Owing to the intrinsic absence of chromophores and high polarity of the glycans, current detection methods are restricted to liquid chromatography and mass spectrometry. Herein, we describe three new imidazolium-based glycan tags: 2’GITag, 3’GITag, and 4’GITag, that significantly improve both the limit of detection and limit of quantification of derivatized oligosaccharides, in terms of fluorescence intensity and ionisation efficiency. Our top-performing derivatisation agent, 4’GITag, shifted the detection sensitivity range from high femtomole to sub-femtomole levels in ESI-MS compared to traditional glycan label, 2AB, enabling the identification of 24 N-glycans in mouse serum, including those bearing sialic acids. Additionally, 4’GITag stabilized Na-salt forms of sialic acids, simplifying the simultaneous analysis of neutral and negative charged N-glycans significantly, avoiding the need for complex derivatisation procedures typically required for the detection of sialylated species. Overall, the favorable performance of imidazolium tags in the derivatisation and sensitive profiling of glycans has the potential for labeling tissue or live cells to explore disease biomarkers and for developing new targeted therapeutic strategies.

Bridging large language model disparities: Skill tagging of multilingual educational content

AbstractThe adoption of large language models (LLMs) in education holds much promise. However, like many technological innovations before them, adoption and access can often be inequitable from the outset, creating more divides than they bridge. In this paper, we explore the magnitude of the country and language divide in the leading open‐source and proprietary LLMs with respect to knowledge of K‐12 taxonomies in a variety of countries and their performance on tagging problem content with the appropriate skill from a taxonomy, an important task for aligning open educational resources and tutoring content with state curricula. We also experiment with approaches to narrowing the performance divide by enhancing LLM skill tagging performance across four countries (the USA, Ireland, South Korea and India–Maharashtra) for more equitable outcomes. We observe considerable performance disparities not only with non‐English languages but with English and non‐US taxonomies. Our findings demonstrate that fine‐tuning GPT‐3.5 with a few labelled examples can improve its proficiency in tagging problems with relevant skills or standards, even for countries and languages that are underrepresented during training. Furthermore, the fine‐tuning results show the potential viability of GPT as a multilingual skill classifier. Using both an open‐source model, Llama2‐13B, and a closed‐source model, GPT‐3.5, we also observe large disparities in tagging performance between the two and find that fine‐tuning and skill information in the prompt improve both, but the closed‐source model improves to a much greater extent. Our study contributes to the first empirical results on mitigating disparities across countries and languages with LLMs in an educational context. Practitioner notesWhat is already known about this topic Recent advances in generative AI have led to increased applications of LLMs in education, offering diverse opportunities. LLMs excel predominantly in English and exhibit a bias towards the US context. Automated content tagging has been studied using English‐language content and taxonomies. What this paper adds Investigates the country and language disparities in LLMs concerning knowledge of educational taxonomies and their performance in tagging content. Presents the first empirical findings on addressing disparities in LLM performance across countries and languages within an educational context. Improves GPT‐3.5's tagging accuracy through fine‐tuning, even for non‐US countries, starting from zero accuracy. Extends automated content tagging to non‐English languages using both open‐source and closed‐source LLMs. Implications for practice and/or policy Underscores the importance of considering the performance generalizability of LLMs to languages other than English. Highlights the potential viability of ChatGPT as a skill tagging classifier across countries.

Improving the Hydrodynamic Performance of Underwater Tags for Blue Shark Monitoring

The use of tag devices in marine environments has become indispensable in attaining a better understanding of marine life and contributing to conservation efforts. However, the successful deployment and operation of underwater tags both depend significantly on their hydrodynamic characteristics, particularly their resistance to motion and stability in various environmental conditions. Herein, a comprehensive study on the hydrodynamic characteristics and optimization of an underwater tag designed for monitoring blue sharks is presented. Firstly, a validation process is conducted by comparing the computational fluid dynamics (CFD) results with the experimental data from Myring’s study, focusing on the resistance characteristics of the tag’s body and the impact of various operational conditions. Subsequently, the validated CFD model is applied to assess the hydrodynamic performance of the tag under different flow conditions, velocities, and angles of attack. Through iterative simulations, including mesh independence studies and boundary condition adjustments, the study identifies key parameters influencing the tag’s resistance and stability. Furthermore, the paper proposes and implements design modifications, including the incorporation of stabilizing fins, aimed at minimizing resistance and improving the tag’s equilibrium position. The effectiveness of these design enhancements is demonstrated through a comparative analysis of resistance and pitching moments for both preliminary and optimized tag configurations. Overall, the study provides valuable insights into the hydrodynamic behavior of underwater tags and offers practical recommendations for optimizing their design to minimize interference with the movement of tagged marine animals.

Validation of tag SNPs for multiple sclerosis HLA risk alleles across the 1000 genomes panel

Currently, the genetic variants strongly associated with risk for Multiple Sclerosis (MS) are located in the Major Histocompatibility Complex. This includes DRB1*15:01 and DRB1*15:03 alleles at the HLA-DRB1 locus, the latter restricted to African populations; the DQB1*06:02 allele at the HLA-DQB1 locus which is in high linkage disequilibrium (LD) with DRB1*15:01; and protective allele A*02:01 at the HLA-A locus. HLA allele identification is facilitated by co-inherited (‘tag’) single nucleotide polymorphisms (SNPs); however, SNP validation is not typically done outside of the discovery population. We examined 19 SNPs reported to be in high LD with these alleles in 2,502 healthy subjects included in the 1000 Genomes panel having typed HLA data. Examination of 3 indices (LD R2 values, sensitivity and specificity, minor allele frequency) revealed few SNPs with high tagging performance. All SNPs examined that tag DRB1*15:01 were in perfect LD in the British population; three showed high tagging performance in 4 of the 5 European, and 2 of the 4 American populations. For DQB1*06:02, with no previously validated tag SNPs, we show that rs3135388 has high tagging performance in one South Asian, one American, and one European population. We identify for the first time that rs2844821 has high tagging performance for A*02:01 in 5 of 7 African populations including African Americans, and 4 of the 5 European populations. These results provide a basis for selecting SNPs with high tagging performance to assess HLA alleles across diverse populations, for MS risk as well as for other diseases and conditions.

Deep Learning based UPoS Tagger for Assamese Religious Text

Religious texts are known to be with specific patterns of writing, and also involve specific vocabularies. These are also known to follow specific style of writing. Thereby these texts are enriched with typical semantic and syntactic characteristics, demanding special attention for Natural Language Processing (NLP) tasks. This research paper focuses on the application of Deep Learning (DL) techniques for Parts of Speech (PoS) tagging focusing on Assamese language religious texts. We have created a specialized dataset comprising approximately 11,000 sentences extracted from various sources including web crawling and filtering religious texts from existing corpora. The dataset was manually validated by linguists to ensure accuracy, errors, and corrections required. A performance matrix was constructed to analyze the performance of the initial tagging using a pre-existing DL-based model trained for Assamese Universal Parts of Speech (UPoS) tagger. Following this, we utilized a subset of the dataset for manual evaluation, and the validated dataset is then considered as a gold standard training dataset for training other DL models using GRU, RNN and Bidirectional LSTM (BiLSTM) architectures. Training accuracies were recorded and presented, demonstrating the effectiveness of the proposed approach. Accuracies, Precision, and Recall were recorded for all the three Models. F1 scores also have been calculated. Comparison of training and testing accuracies are depicted with performance graphs.

Tagging Items with Emerging Tags: A Neural Topic Model Based Few-Shot Learning Approach

The tagging system has become a primary tool to organize information resources on the Internet, which benefits both users and the platforms. To build a successful tagging system, automatic tagging methods are desired. With the development of society, new tags keep emerging. The problem of tagging items with emerging tags is an open challenge for an automatic tagging system, and it has not been well studied in the literature. We define this problem as a tag-centered cold-start problem in this study and propose a novel neural topic model based few-shot learning method named NTFSL to solve the problem. In our proposed method, we innovatively fuse the topic modeling task with the few-shot learning task, endowing the model with the capability to infer effective topics to solve the tag-centered cold-start problem with the property of interpretability. Meanwhile, we propose a novel neural topic model for the topic modeling task to improve the quality of inferred topics, which helps enhance the tagging performance. Furthermore, we develop a novel inference method based on the variational auto-encoding framework for model inference. We conducted extensive experiments on two real-world datasets, and the results demonstrate the superior performance of our proposed model compared with state-of-the-art machine learning methods. Case studies also show the interpretability of the model.

Novel Microsatellite Tags Hold Promise for Illuminating the Lost Years in Four Sea Turtle Species.

After hatching, sea turtles leave the nest and disperse into the ocean. Many years later, they return to their natal coastlines. The period between their leaving and their returning to natal areas, known as the "Lost Years", is poorly understood. Satellite tracking studies aimed at studying the "Lost Years" are challenging due to the small size and prolonged dispersal phases of young individuals. Here, we summarize preliminary findings about the performance of prototype microsatellite tags deployed over a three-year period on 160 neonate to small juvenile sea turtles from four species released in the North Atlantic Ocean. We provide an overview of results analyzing tag performance with metrics to investigate transmission characteristics and causes of tag failure. Our results reveal that, despite certain unfavorable transmission features, overall tag performance was satisfactory. However, most track durations were shorter than those observed on individuals of similar size in other studies and did not allow for detailed analyses of trajectories and turtle behavior. Our study further suggests that tracking durations are correlated with the targeted species, highlighting a lack of robustness against some neritic behaviors. Unprecedented diving data obtained for neonate sea turtles in this study suggest that the vertical behaviors of early juveniles are already too strenuous for these miniaturized tags. Our findings will help to inform the biologging research community, showcasing recent technological advances for the species and life stages within our study.

Effect of Beverage Composition on Radio Frequency Identification (RFID) Performance Using Polyethylene Terephthalate (PET) Bottles for Smart Food Packaging Applications.

Radio frequency identification (RFID) technology is crucial in revolutionizing the food supply chain and combating global food waste. However, this technology faces challenges in full integration due to disruptive effects on tags caused by the dielectric properties of food and beverage ingredients, chemical constituents, and their packaging. This paper aims to demonstrate the effect of packaging and beverage contents on RFID tag performance. Three commercially available ultra-high frequency (UHF) RFID tags with different designs were tested on polyethylene terephthalate (PET) bottles, measuring tag performance through sensitivity, backscatter, and read range in the presence of various water-based solutions and commercially available beverages. The results highlight the substantial impact of the beverage type and tag design on RFID performance. The results of this study showed that tag 3 was the most consistent and readable tag amongst those tested in the presence of different beverage contents. Tag 3 resulted in a sensitivity ranging from -0.49 to -2.01 dBm, backscatter from -38.16 to 43.59 dBm, and read range from 1.58 to 1.88 m, while tag 1 performed the best in the presence of an empty PET bottle resulting in a sensitivity of -20.78 dBm, backscatter of -23.65 dBm, and read range of 16.34 m. The results of this study can be used for further investigations to develop a mathematical model that predicts the RFID tag performance based on the food composition. This model will be helpful for the design of the tags while facilitating the adoption of smart packaging for food traceability.

ParticleNet and its application on CEPC jet flavor tagging

Quarks (except top quarks) and gluons produced in collider experiments hadronize and fragment into sprays of stable particles, called jets. Identification of quark flavor is desired for collider experiments in high-energy physics, relying on flavor tagging algorithms. In this study, using a full simulation of the Circular Electron Positron Collider (CEPC), we investigate the flavor tagging performance of two different algorithms: ParticleNet, based on a Graph Neural Network, and LCFIPlus, based on the Gradient Booted Decision Tree. Compared to LCFIPlus, ParticleNet significantly enhances flavor tagging performance, resulting in a significant improvement in benchmark measurement accuracy, i.e., a 36% improvement for σ(ZH)·Br(Z→νν¯,H→cc¯)\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\sigma (ZH)\\cdot Br(Z\\rightarrow \ u \\bar{\ u }, H\\rightarrow c\\bar{c})$$\\end{document} measurement and a 75% improvement for |Vcb|\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$|V_{cb}|$$\\end{document} measurement via W boson decay, respectively, when the CEPC operates as a Higgs factory at the center-of-mass energy of 240 GeV and collects an integrated luminosity of 5.6 ab-1\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$^{-1}$$\\end{document}. We compare the performance of ParticleNet and LCFIPlus at different vertex detector configurations, observing that the inner radius is the most sensitive parameter, followed by material budget and spatial resolution.

Tri-relational multi-faceted graph neural networks for automatic question tagging

Automatic question tagging is a crucial task in Community Question Answering (CQA) systems such as Zhihu or Quora, as it can significantly enhance the user experience by improving the efficiency of question answering and expert recommendations. Graph-based collaborative filtering models show promising performance on this task, as they can exploit not only the semantics of text content but also the existing relations between questions and tags. However, existing approaches typically encode each question into a single vector, which may not be able to capture the diverse semantic facets of questions in CQA systems. To address this challenge, we propose a novel question-tagging framework, named Tri-Relational Multi-Faceted Graph Neural Networks (TRMFG) for Automatic Question Tagging. In TRMFG, a tri-relational graph structure is designed to better model the question-tag relations. We also propose tri-relational question-tag GNN to extract hidden latent representations of questions and tags. Specially, the Multi-Faceted Question GNN helps capture the diverse semantics of questions from relevant tags. Then we build a multiple matching component to capture more complex matching patterns of the questions based on the diverse semantics. Our experimental results on three benchmark datasets demonstrate that TRMFG significantly improves question tagging performance for CQA, outperforming the state-of-the-art methods.

Improved quantification of carbonyl sub-metabolome by liquid chromatography mass spectrometry using a fragment controlled multiplexed isotopic tag

BackgroundCarbonyl-containing metabolites are a class of key intermediate in metabolism, which has potentials to be biomarkers. Since their poor ionization, derivatization reagents, such as dansylhydrazine, are usually used to improve the sensitivity and/or to facilitate quantification. However, most current carbonyl derivatization reagents only have two channels, one is isotopically labeled and the other one is non-labeled. To quantify more samples in a run and using data-independent acquisition (DIA) mode to get comprehensive and unbiased mass fragmentation, we proposed a fragment-controlled isotopic tag, called DiMe-FP-NHNH2 (FP) which has five channels: Δ0, Δ3, Δ6, Δ9, and Δ12, thus up to 5 samples can be analyzed in a run. ResultsThe most important improvement is that the FP tag can produce multiple characteristic signals in tandem mass, diagnostic ions and neutral losses, which helps to selectively detect aldehydes/ketones for targeted and untargeted analysis. To exhibit all capabilities of the FP tag, we mimicked an untargeted metabolomics experiment, which comprises two steps. First, discovery step, using Data-Independent Analysis (SWATH-MS) and the labeling of two channels (Δ0 and Δ3), we picked out aldehyde/ketone from the pooled urine samples based on three characteristic signals, including isotope patterns, diagnostic ions, and neutral losses. Second, five-plex quantification, relative and absolute quantification were achieved in a single LC-MS analysis. Notably, because of different nominal masses, the FP tag can be used on any low or high resolution mass spectrometers. SignificanceThe benefits and performance of the FP tag are demonstrated by the analysis of urine samples collected from patients from a prostate cancer study, in which more than a thousand features were found based on MS1 fingerprint, but only around 120 aldehyde/ketone candidates were confirmed with characteristic signals and nine of which were quantified showing significant differences from healthy and reference urine samples.

Quark/gluon discrimination and top tagging with dual attention transformer

Jet tagging is a crucial classification task in high energy physics. Recently the performance of jet tagging has been significantly improved by the application of deep learning techniques. In this study, we introduce a new architecture for jet tagging: the particle dual attention transformer (P-DAT). This novel transformer architecture stands out by concurrently capturing both global and local information, while maintaining computational efficiency. Regarding the self attention mechanism, we have extended the established attention mechanism between particles to encompass the attention mechanism between particle features. The particle attention module computes particle level interactions across all the particles, while the channel attention module computes attention scores between particle features, which naturally captures jet level interactions by taking all particles into account. These two kinds of attention mechanisms can complement each other. Further, we incorporate both the pairwise particle interactions and the pairwise jet feature interactions in the attention mechanism. We demonstrate the effectiveness of the P-DAT architecture in classic top tagging and quark–gluon discrimination tasks, achieving competitive performance compared to other benchmark strategies.

Auxiliary self-supervision to metric learning for music similarity-based retrieval and auto-tagging.

In the realm of music information retrieval, similarity-based retrieval and auto-tagging serve as essential components. Similarity-based retrieval involves automatically analyzing a music track and fetching analogous tracks from a database. Auto-tagging, on the other hand, assesses a music track to deduce associated tags, such as genre and mood. Given the limitations and non-scalability of human supervision signals, it becomes crucial for models to learn from alternative sources to enhance their performance. Contrastive learning-based self-supervised learning, which exclusively relies on learning signals derived from music audio data, has demonstrated its efficacy in the context of auto-tagging. In this work, we propose a model that builds on the self-supervised learning approach to address the similarity-based retrieval challenge by introducing our method of metric learning with a self-supervised auxiliary loss. Furthermore, diverging from conventional self-supervised learning methodologies, we discovered the advantages of concurrently training the model with both self-supervision and supervision signals, without freezing pre-trained models. We also found that refraining from employing augmentation during the fine-tuning phase yields better results. Our experimental results confirm that the proposed methodology enhances retrieval and tagging performance metrics in two distinct scenarios: one where human-annotated tags are consistently available for all music tracks, and another where such tags are accessible only for a subset of music tracks.

E-PANNs: Sound Recognition Using Efficient Pre-trained Audio Neural Networks

Sounds carry an abundance of information about activities and events in our everyday environment, such as traffic noise, road works, music, or people talking. Recent machine learning methods, such as convolutional neural networks (CNNs), have been shown to be able to automatically recognize sound activities, a task known as audio tagging. One such method, pre-trained audio neural networks (PANNs), provides a neural network which has been pre-trained on over 500 sound classes from the publicly available AudioSet dataset, and can be used as a baseline or starting point for other tasks. However, the existing PANNs model has a high computational complexity and large storage requirement. This could limit the potential for deploying PANNs on resource-constrained devices, such as on-the-edge sound sensors, and could lead to high energy consumption if many such devices were deployed. In this paper, we reduce the computational complexity and memory requirement of the PANNs model by taking a pruning approach to eliminate redundant parameters from the PANNs model. The resulting Efficient PANNs (E-PANNs) model, which requires 36% less computations and 70% less memory, also slightly improves the sound recognition (audio tagging) performance. The code for the E-PANNs model has been released under an open source license.