Meta Network Research Articles

Opossums as opportunistic visitors of chiropterophilous flowers: Interaction between Didelphis albiventris (Marsupialia) and Hymenaea cangaceira (Fabaceae)

AbstractBat‐pollinated plants feature open flowers, facilitating easy access to their abundant nectar for various animal groups. The Caatinga, a Brazilian Seasonally Tropical Dry Forest, stands out globally as a hotspot for bat‐pollinated plants. This study presents the first documentation of the opossum Didelphis albiventris (Marsupialia) engaging in the consumption of nectar on flowers of Hymenaea cangaceira (Fabaceae), a bat‐pollinated tree species within the Caatinga. The observation spanned five nights and occurred unexpectedly during an experiment involving bagged flowers to assess nectar dynamics. Initially, torn bags were discovered containing flowers, which nectar had been consumed. Subsequent nights revealed the opossum actively exploring bagged flowers, likely guided by the fragrant nectar of H. cangaceira, and unmanipulated flowers located in the canopy's upper reaches. Given the opossum's contact with reproductive structures during these visits, it emerges as a potential pollinator. Following this observation, a comprehensive literature review on Didelphis interactions with other plants was conducted, leading to the construction of a meta network. This network unequivocally illustrates that the majority of plants visited by Didelphis indeed have bats as their primary pollinators. Owing to the nocturnal, arboreal and stealthy nature of opossums, documenting their interactions with flowers proves highly challenging. Consequently, our record highlights significant knowledge gaps in comprehending the interplay between opossums and chiropterophilous flowers. These gaps include the under‐sampling of this process in the Caatinga, the role of olfactory signalling and the broader consequences of this unique interaction.

Multi-task prompt tuning with soft context sharing for vision–language models

Vision–language models have recently shown great potential on many tasks in computer vision. Meanwhile, prior work demonstrates prompt tuning designed for vision–language models could acquire superior performance on few-shot image recognition compared to linear probe, a strong baseline. In practice, many few-shot tasks are inherently correlated, particularly within specialized domains. However, such information is overlooked previously. Inspired by the fact that modeling task relationship by multi-task learning can usually boost performance, we propose a novel method SoftCPT (Soft Context Sharing for Prompt Tuning) to tune pre-trained vision–language models on multiple target few-shot tasks jointly. Specifically, we design a task-shared meta network to generate prompt context for each task using task name together with a learnable task context as input. The parameters of this meta network as well as the task context are tuned on the joint training set of all tasks. As such, the prompt context of all tasks will be shared in a soft manner. Extensive experiments across four multi-task few-shot datasets covering 44 tasks and 1593 categories demonstrate that SoftCPT significantly outperforms single-task prompt tuning methods, highlighting the effectiveness of multi-task learning for vision–language prompt tuning.

Fault Diagnosis of Harmonic Drives Using Multimodal Collaborative Meta Network With Severely Missing Modality

Fault Diagnosis of Harmonic Drives Using Multimodal Collaborative Meta Network With Severely Missing Modality

Cross-Domain Recommendation To Cold-Start Users Via Categorized Preference Transfer

Abstract Most existing cross-domain recommendation (CDR) systems apply the embedding and mapping idea to tackle the cold-start user problem and, to this end, they learn a common bridge function to transfer the user preferences from the source domain into the target domain. However, sharing a bridge function for all users inevitably leads to biased recommendations. This paper proposes a novel method, named CDR to cold-start users via categorized preference transfer (CDRCPT), to overcome the shortcomings of existing approaches. First, the embeddings of users and items in both the source and target domain are learned through pretraining and we utilize preference encoder to obtain the preference embeddings of users in the source domain. Second, mini-batch clustering is applied in the source domain to group users according to their preferences; here, each cluster identifies a specific class of users, and each cluster is represented by its center. Finally, the general representation is fed into a meta network to learn a bridge function for each available class of users. Experiments on two real data sets show that our CDRCPT method is effective in improving the accuracy and robustness of recommendations.

Untangling biodiversity interactions: A meta network on pollination in Earth's most diverse tropical savanna.

Pollination is vital for ecosystem functioning, especially in biodiversity-rich regions like the Brazilian Cerrado. Our research establishes a comprehensive meta network of pollinator-plant interactions within this biome. We quantified the importance of different pollinator groups, identifying keystone species. We examined potential biases in sampling effort and the spatial behavior of interactions within the heterogeneous Cerrado plant physiognomies. Our investigation uncovered 1499 interactions among 293 plant species and 386 visitor species, with legitimate pollination accounting for 42.4% of the interactions. The network exhibited modularity, driven by bees and insects, with vertebrates bridging diurnal and nocturnal modules. While a generalized pattern emerged, high specialization existed within modules due to habitat diversity. Bees, particularly Apis mellifera (exotic) and Trigona spinipes (native), played central roles as network hubs. Hummingbirds and bats, engaged in specialized interactions showing strong connectivity within and between modules. Interestingly, invertebrate-vertebrate modules were more prevalent than expected in the meta network. However, a bias was evident, primarily within specific biogeographical districts with fragmented landscapes and intrusion from other biomes. Variations in plant species and endemism rates influenced pollinator occurrence and the Cerrado network topology. Our study offers valuable insights into pollinator-plant interactions within the Cerrado, encompassing both invertebrates and vertebrates. The modeled network represents a significant step in understanding the structural complexity of pollination networks, integrating partial networks from diverse pollination systems within heterogeneous habitats. Nevertheless, a biogeographical bias could limit a comprehensive understanding of network functionality across the Cerrado.

Improving Semantic Analysis in Visualization with Meta Network Representation and Parsing Algorithm

This article aims to advance semantic analysis models, particularly in visualization, by proposing a novel semantic representation method utilizing the semantic Meta Network (MNet). MNet is a complex framework comprising semantic elements, internal and external relationships, and feature attributes, defined hierarchically through recursive processes, aiming to depict the comprehensive semantic space from phrase-level components to complete texts. The methodology involves the development of a general construction algorithm for MNet, encompassing meta relationships, tree structures, and network structures, and a Parsing method for specific semantic analysis problems, including a bottom-up specification-based MNet semantic dependency tree construction algorithm and a network construction algorithm tailored for natural language interface parsing. Empirical experiments confirm the effectiveness of these algorithms, particularly in parsing natural language control interface instructions in Supervisory Control and Data Acquisition (SCADA) systems, bridging specific semantic analysis problems with the general construction and parsing processes of MNet, accounting for internal semantics concerning language unit structures and foreign language meanings in the linguistic context, thereby contributing significantly to the field of natural language semantic analysis.

Distilling base-and-meta network with contrastive learning for few-shot semantic segmentation

Current studies in few-shot semantic segmentation mostly utilize meta-learning frameworks to obtain models that can be generalized to new categories. However, these models trained on base classes with sufficient annotated samples are biased towards these base classes, which results in semantic confusion and ambiguity between base classes and new classes. A strategy is to use an additional base learner to recognize the objects of base classes and then refine the prediction results output by the meta learner. In this way, the interaction between these two learners and the way of combining results from the two learners are important. This paper proposes a new model, namely Distilling Base and Meta (DBAM) network by using self-attention mechanism and contrastive learning to enhance the few-shot segmentation performance. First, the self-attention-based ensemble module (SEM) is proposed to produce a more accurate adjustment factor for improving the fusion of two predictions of the two learners. Second, the prototype feature optimization module (PFOM) is proposed to provide an interaction between the two learners, which enhances the ability to distinguish the base classes from the target class by introducing contrastive learning loss. Extensive experiments have demonstrated that our method improves on the PASCAL-5i under 1-shot and 5-shot settings, respectively.

From Instance to Metric Calibration: A Unified Framework for Open-World Few-Shot Learning.

Robust few-shot learning (RFSL), which aims to address noisy labels in few-shot learning, has recently gained considerable attention. Existing RFSL methods are based on the assumption that the noise comes from known classes (in-domain), which is inconsistent with many real-world scenarios where the noise does not belong to any known classes (out-of-domain). We refer to this more complex scenario as open-world few-shot learning (OFSL), where in-domain and out-of-domain noise simultaneously exists in few-shot datasets. To address the challenging problem, we propose a unified framework to implement comprehensive calibration from instance to metric. Specifically, we design a dual-networks structure composed of a contrastive network and a meta network to respectively extract feature-related intra-class information and enlarged inter-class variations. For instance-wise calibration, we present a novel prototype modification strategy to aggregate prototypes with intra-class and inter-class instance reweighting. For metric-wise calibration, we present a novel metric to implicitly scale the per-class prediction by fusing two spatial metrics respectively constructed by the two networks. In this way, the impact of noise in OFSL can be effectively mitigated from both feature space and label space. Extensive experiments on various OFSL settings demonstrate the robustness and superiority of our method. Our source codes is available at https://github.com/anyuexuan/IDEAL.

A meta network pruning framework for remaining useful life prediction of rocket engine bearings with temporal distribution discrepancy

Accurate remaining useful life prediction (RUL) is important for the reliability and safety of liquid rocket engines. In this paper, a meta network pruning framework with attention augmented convolutions is proposed for RUL prediction. To address the problem of distribution discrepancy in engineering data under transient working conditions, a data-driven distribution matching strategy is designed. Besides, in view of the prediction accuracy and computation complexity of the model, an iterative meta network pruning algorithm, which automatically calculates the meta-gradients of each convolutional kernel according to the chain rule, is developed to identify, and then delete the unimportant connections in the framework. The method is verified on a high-precision cryogenic rocket engine bearing experiment platform under liquid nitrogen and received better performance than benchmark algorithms.

Accelerating physics-informed neural network based 1D arc simulation by meta learning

Physics-informed neural networks (PINNs) have a wide range of applications as an alternative to traditional numerical methods in plasma simulation. However, in some specific cases of PINN-based modeling, a well-trained PINN may require tens of thousands of optimizing iterations during training stage for complex modeling and huge neural networks, which is sometimes very time-consuming. In this work, we propose a meta-learning method, namely Meta-PINN, to reduce the training time of PINN-based 1D arc simulation. In Meta-PINN, the meta network is first trained by a two-loop optimization on various training tasks of plasma modeling, and then used to initialize the PINN-based network for new tasks. We demonstrate the power of Meta-PINN by four cases corresponding to 1D arc models at different boundary temperatures, arc radii, arc pressures, and gas mixtures. We found that a well-trained meta network can produce good initial weights for PINN-based arc models even at conditions slightly outside of training range. The speed-up in terms of relative L2 error by Meta-PINN ranges from 1.1× to 6.9× in the cases we studied. The results indicate that Meta-PINN is an effective method for accelerating the PINN-based 1D arc simulation.

Implementation of an Artificial Intelligence Approach to GPR Systems for Landmine Detection

Artificial Neural Network (ANN) approaches are applied to detect and determine the object class using a special set of the UltraWideBand (UWB) pulse Ground Penetrating Radar (GPR) sounding results. It used the results of GPR sounding with the antenna system, consisting of one radiator and four receiving antennas located around the transmitting antenna. The presence of four receiving antennas and, accordingly, the signals received from four spatially separated positions of the antennas provide a collection of signals received after reflection from an object at different angles and, due to this, to determine the location of the object in a coordinate system, connected to the antenna. We considered the sums and differences of signals received by two of the four antennas in six possible combinations: (1 and 2, 1 and 3, 2 and 3, 1 and 4, etc.). These combinations were then stacked sequentially one by one into one long signal. Synthetic signals constructed in such a way contain many more notable differences and specific information about the class to which the object belongs as well as the location of the searched object compared to the signals obtained by an antenna system with just one radiating and one receiving antenna. It therefore increases the accuracy in determining the object’s coordinates and its classification. The pulse radiation, propagation, and scattering are numerically simulated by the finite difference time domain (FDTD) method. Results from the experiment on mine detection are used to examine ANN too. The set of signals from different objects having different distances from the GPR was used as a training and testing dataset for ANN. The training aims to recognize and classify the detected object as a landmine or other object and to determine its location. The influence of Gaussian noise added to the signals on noise immunity of ANN was investigated. The recognition results obtained by using an ANN ensemble are presented. The ensemble consists of fully connected and recurrent neural networks, gated recurrent units, and a long-short term memory network. The results of the recognition by all ANNs are processed by a meta network to provide a better quality of underground object classification.

Learning the Effect of Registration Hyperparameters with HyperMorph

We introduce HyperMorph, a framework that facilitates efficient hyperparameter tuning in learning-based deformable image registration. Classical registration algorithms perform an iterative pair-wise optimization to compute a deformation field that aligns two images. Recent learning-based approaches leverage large image datasets to learn a function that rapidly estimates a deformation for a given image pair. In both strategies, the accuracy of the resulting spatial correspondences is strongly influenced by the choice of certain hyperparameter values. However, an effective hyperparameter search consumes substantial time and human effort as it often involves training multiple models for different fixed hyperparameter values and may lead to suboptimal registration. We propose an amortized hyperparameter learning strategy to alleviate this burden by learning the impact of hyperparameters on deformation fields. We design a meta network, or hypernetwork, that predicts the parameters of a registration network for input hyperparameters, thereby comprising a single model that generates the optimal deformation field corresponding to given hyperparameter values. This strategy enables fast, high-resolution hyperparameter search at test-time, reducing the inefficiency of traditional approaches while increasing flexibility. We also demonstrate additional benefits of HyperMorph, including enhanced robustness to model initialization and the ability to rapidly identify optimal hyperparameter values specific to a dataset, modality, task, or even anatomical region, all without the need to retrain models. We make our code publicly available at <a href='http://hypermorph.voxelmorph.net'>http://hypermorph.voxelmorph.net</a>

Domain-aware meta network for radar HRRP target recognition with missing aspects

Conventional high resolution range profile (HRRP) based radar automatic target recognition (RATR) methods are based on a complete template library assumption. However, such an assumption can barely hold in real world applications since it is difficult to observe target echoes from all aspects, especially for noncooperative targets. To alleviate the target-aspect missing problem, we develop a domain-aware meta network (DOAMN) for HRRP-based RATR. Specifically, the DOAMN first uses a domain-aware (DOM) module to distinguish whether HRRPs come from the seen or unseen target aspects, then a meta network (MNet) is employed to learn a generalized parameter initialization that is able to achieve fast adaptation among target aspects. To enable the proposed DOAMN to be trained in an end-to-end manner, we further present an effective iterative hybrid optimization method. Experiments on simulated HRRP dataset demonstrate the effectiveness and efficiency of the proposed model.

Show, Attend and Distill: Knowledge Distillation via Attention-based Feature Matching

Knowledge distillation extracts general knowledge from a pretrained teacher network and provides guidance to a target student network. Most studies manually tie intermediate features of the teacher and student, and transfer knowledge through predefined links. However, manual selection often constructs ineffective links that limit the improvement from the distillation. There has been an attempt to address the problem, but it is still challenging to identify effective links under practical scenarios. In this paper, we introduce an effective and efficient feature distillation method utilizing all the feature levels of the teacher without manually selecting the links. Specifically, our method utilizes an attention based meta network that learns relative similarities between features, and applies identified similarities to control distillation intensities of all possible pairs. As a result, our method determines competent links more efficiently than the previous approach and provides better performance on model compression and transfer learning tasks. Further qualitative analyses and ablative studies describe how our method contributes to better distillation.

Fast human motion transfer based on a meta network

The paper proposes a new approach to the task of unsupervised human motion style transfer which transforms the style of an unlabeled input human motion into the style of another unlabeled motion sequence and preserves the behavior of the primitive motion at the same time. For supervised stylistic motion transfer task, constructing a database including a large quantity of training sample pairs is expensive and tedious for artists. Therefore, we focus on unsupervised methods. However, recent works need to retrain a network for each new character with different styles, which is a time-consuming and resource-intensive process. To tackle this problem, we utilize a meta network which builds a direct-mapping from the style motion sequence to the transformation network, namely, producing a corresponding motion transformation network in the meta network by a feed-forward propagation. The model transfers multiple styles of motion clips to different motions and generates natural stylized motions in real-time without a re-training for every style. Besides, our model can transfer styles between two movements depending on different skeletons. We also explore important features of the transfer network manifold by interpolation between the hidden states from meta network. Experiments validate the flexibility and effectiveness of our data-driven model.

Asynchronous framework with Reptile+ algorithm to meta learn partially observable Markov decision process

Meta-learning has recently received much attention in a wide variety of deep reinforcement learning (DRL). In non-meta-learning, we have to train a deep neural network as a controller to learn a specific control task from scratch using a large amount of data. This way of training has shown many limitations in handling different related tasks. Therefore, meta-learning on control domains becomes a powerful tool for transfer learning on related tasks. However, it is widely known that meta-learning requires massive computation and training time. This paper will propose a novel DRL framework, which is called HCGF-R2-DDPG (Hybrid CPU/GPU Framework for Reptile+ and Recurrent Deep Deterministic Policy Gradient). HCGF-R2-DDPG will integrate meta-learning into a general asynchronous training architecture. The proposed framework will allow utilising both CPU and GPU to boost the training speed for the meta network initialisation. We will evaluate HCGF-R2-DDPG on various Partially Observable Markov Decision Process (POMDP) domains.

Dynamic interaction networks for image-text multimodal learning

Recently, there is a surge of interest in image-text multimodal representation learning, and many neural network based models have been proposed aiming to capture the interaction between two modalities with different forms of functions. Despite their success, a potential limitation of these methods is insufficient to model all kinds of interactions with a set of static parameters. To alleviate this problem, we present a dynamic interaction network, in which the parameters of the interaction function are dynamically generated by a meta network. Additionally, to provide necessary multimodal features that the meta network needs, we propose a new neural module called Multimodal Transformer. Experimentally, we not only make a comprehensively quantitative evaluation on four image-text tasks, but also show some interpretable analyses of our models, revealing the internal working mechanism of the dynamic parameter learning.

Algorithms of the Minimum Spanning Tree in Gas Supply Systems

In this paper, a method for tracing gas networks using the Kruskal algorithm and the Steiner algorithm is considered. The result of the work of algorithms on the example of a bio meta network between agricultural enterprises located on the territory of the Belgorod region is demonstrated. The economic result is determined using Steiner’s refinement algorithm.

Multilevel meta network analysis with application to studying network dynamics of network interventions

In this paper, I introduce new methods for multilevel meta network analysis. The new methods can combine results from multiple network models, assess the effects of predictors at network or higher levels and account for both within- and cross-network correlations of the parameters in the network models. To demonstrate the new methods, I studied network dynamics of a smoking prevention intervention that was implemented in 76 classes of six middle schools in China. The results show that as compared to random intervention (i.e., that targets random students), smokers’ popularity was significantly reduced in the classes with network interventions (i.e., those target central students or students with their friends together). The findings highlight the importance of examining network outcomes in evaluating social and health interventions, the role of social selection in managing social influence, and the potential of using network methods to design more effective interventions.

Creating, Using and Updating Thesauri Files for Automap and ORA

Abstract : AutoMap [1] is text analysis software that performs Network Text Analysis by running an automated process on a corpus of raw text data to generate one or more meta-networks which include the nodes and links representing relations among entities described. Automap uses thesaurus files [1] when creating meta-networks. These thesaurus files are list which allows the association of words or phrases found in texts with abstract concepts and/or node classes used in the extracted meta-networks. Over time, a large number of thesauri have been created. Many of the extant thesauri contain entries that are relevant to new text analysis projects. But thesaurus re-use is difficult due to the number of thesauri. In this report, we describe one approach to making thesaurus re-use easier by combining and reconciling multiple thesauri into one under user control. With this approach, the process of creating a Meta network out of a raw corpus of text data is more efficient and the user is able to perform a more accurate analysis of the Meta network, as the individual thesauri files can be merged to create a single and large Universal or Master Thesaurus containing all the general abstract concepts, along with several different Domain-specific thesauri. In the following report, we first discuss the differences between a Universal thesaurus and the domain or the project specific thesauri. We then go on to discuss the evolution in the formats of the thesauri used by AutoMap, followed by a discussion of the standard Dynamic Network Analysis (DNA) meta-ontology [1]. We then detail the process used to create a single universal/master thesaurus and several different Domain thesauri. The process involves a mix of two major processes which we refer to as the Split routine and the Merge routine. We shall discuss the Split routine and the merge routine algorithm along with the process that has been used to merge and create a single thesaurus file by combining a large number of thesauri files.