Vahid Seydi

Conservation of marine ecosystems can be improved through a better understanding of ecosystem functioning, particularly the cryptic underwater behaviours and interactions of marine predators. Image‐based bio‐logging devices (including images, videos and active acoustic) are increasingly used to monitor wildlife movements, foraging behaviours and their environment, but generate complex datasets needing efficient analytical tools. We review advances in image‐based bio‐logging technology for ecological studies on marine fauna. Emphasis is placed on the diversity of data collected, merging research questions, challenges in image processing, and integration of Artificial Intelligence (AI) methods. Image‐based system issues, such as exposure, focus, blurriness, colour balance, moving background, perspective and scale variability are even more challenging in underwater images where conditions change constantly and cannot be controlled. We list computer vision tools and algorithms available for analyses of underwater images, including enhanced tracking algorithms that recognise objects and treat images as a time series. Although AI and computer vision methods offer ample and robust analytical solutions for (semi‐) automated image processing, their uptake by marine ecologists has been slow. Collaboration among ecologists, modellers, statisticians, engineers and computer scientists is needed to integrate ecological questions, data selection and computational methodology. We propose a four‐phase framework for image data processing and analysis (video checking and manipulation, image processing, image labelling and model development) accompanied by detailed python code. We also outline the additional complications in aligning the diverse scalar movement metrics from bio‐loggers along with image‐based data, such as acceleration, depth and location, which typically are collected at different resolutions. Building analytical frameworks for on‐board image data collection (e.g. lightweight models) is also explored. We advocate for a collaborative research community at the Ecology‐AI interface, emphasising sharing and exchange of both data and tools to drive cross‐disciplinary innovation. Beyond the Ecology‐AI interface, we pave the path for the application of insights from image‐based bio‐logging technology enabling collaboration among scientists, conservation managers, and policymakers. Systematic applications of computer vision tools to image‐based bio‐logging technology will enhance the power these data hold, informing about the status of marine ecosystems, testing and developing ecological theory and aiding conservation.

Recent advancements in semi-supervised deep learning have introduced effective strategies for leveraging both labeled and unlabeled data to improve classification performance. This work proposes a semi-supervised framework that utilizes a distance-based weighting mechanism to prioritize critical training samples based on their proximity to test data. By focusing on the most informative examples, the method enhances model generalization and robustness, particularly in challenging scenarios with noisy or imbalanced datasets. Building on techniques such as uncertainty consistency and graph-based representations, the approach addresses key challenges of limited labeled data while maintaining scalability. Experiments on twelve benchmark datasets demonstrate significant improvements across key metrics, including accuracy, precision, and recall, consistently outperforming existing methods. This framework provides a robust and practical solution for semi-supervised learning, with potential applications in domains such as healthcare and security where data limitations pose significant challenges.

Collecting a large amount of labeled data in machine learning is always challenging. Often, even with sufficient data, domain differences can cause a shift or bias in data distribution, affecting model performance during testing. Domain adaptation methods, especially adversarial techniques, are effective solutions for these challenges. The goal is to learn a classifier for an unlabeled target dataset using a labeled source dataset, enhancing resistance to domain shifts. However, existing methods sometimes struggle with adapting the joint feature distribution across domains, resulting in negative transfer. To address this, we propose a method that forms class-specific clusters to prevent negative transfer. This method is encapsulated in an unsupervised adversarial domain adaptation framework based on a variational auto-encoder. Our structure is designed to enhance invariant and discriminative feature representation. We process source and target data through a VAE to establish a smooth latent representation. In our method, source and target data are fed into a variational auto-encoder, which produces a smooth latent representation. The feature extractor then plays an adversarial minimax game with the discriminator to learn domain-invariant features, while the feature extractor is shared between the reconstructed source and reconstructed target data. In addition, we proposed a second structure in which the domain discriminator part of the prior structure is eliminated to demonstrate the influence of the variational auto-encoder in domain adaptation. On numerous unsupervised domain adaptation benchmarks, our results indicate that our proposed model outperforms or is comparable to state-of-the-art outcomes.

The object segmentation mask’s observation sequence shows the trend of changes in the object’s observable geometric form, and predicting them may assist in solving various difficulties in multi-object tracking and segmentation (MOTS). With this aim, we propose the entangled appearance and motion structures network (EAMSN), which can predict the object segmentation mask at the pixel level by integrating VAE and LSTM. Regardless of the surroundings, each EAMSN keeps complete knowledge about the sequence of probable changes in the seen map of the object and its related dynamics. It suggests that EAMSN understands the item meaningfully and is not reliant on instructive examples. As a result, we propose a novel MOTS algorithm. By employing different EAMSNs for each kind of item and training them offline, ambiguities in the segmentation mask discovered for that object may be recovered, and precise estimation of the real boundaries of the object at each step. We analyze our tracker using the KITTI MOTS and MOTS challenges datasets, which comprise car and pedestrian objects, to illustrate the usefulness of the suggested technique. As a result, we developed distinct EAMSNs for cars and pedestrians, trained using the MODELNET40 and Human3.6 M datasets, respectively. The discrepancy between training and testing data demonstrates that EAMSN is not dependent on training data. Finally, we compared our strategy to a variety of other ways. Compared to the published findings, our technique gets the best overall performance.

Deep learning methods demand enormous amounts of labeled data. Although collecting labeled data is a challenge but it can be accomplished with difficulty. Nevertheless, domain shift or bias may occur due to some conditions. Recollecting data under similar conditions is too expensive or impossible. Domain adaptation is an effective technique to deal with this problem. Among domain adaptation methods, adversarial learning models handled domain shifts well. Adversarial learning’s biggest issue is matching the features to the appropriate classes. In other words, domain adaptation may lead to negative transfer. Utilizing adversarial learning and Variational Auto-Encoder (VAE) cluster-forming capabilities, we propose a method that overcomes these limitations. Our structure attempts to generate a smooth latent representation based on both the target and the source data using a variational auto-encoder. Then the affected data are fed into an adversarial learning component. The component includes an encoder and a discriminator. The discriminator aims to identify the source and target encoders’ features. The target encoder uses inverted domain labels to mislead the discriminator. To emphasize the desired effect of VAE on domain adaptation, we test the model performance without adversarial training. In comparison to the most prevalent adversarial domain adaptation benchmarks, our method yields approving and comparable results.

Decision trees are a fundamental statistical learning tool for addressing classification and regression problems through a recursive partitioning approach that effectively accommodates numerical and categorical data [1, 2]. The Classification and regression tree (CART) algorithm underlies modern Boosting methodologies such as Gradient boosting machine (GBM), Extreme gradient boosting (XGBoost), and Light gradient boosting machine (LightGBM). However, the standard CART algorithm may require improvement due to its inability to learn from unlabeled data. This study proposes several modifications to incorporate test data into the training phase. Specifically, we introduce a method based on Graph-based semi-supervised learning called “Distance-based Weighting,” which calculates and removes irrelevant records from the training set to accelerate the training process and improve performance. We present Semi-supervised classification and regression tree (Semi-Cart), a new implementation of CART that constructs a decision tree using weighted training data. We evaluated its performance on thirteen datasets from various domains. Our results demonstrate that Semi-Cart outperforms standard CART methods and contributes to statistical learning.

Community detection in complex networks is an important task for discovering hidden information in network analysis. Neighborhood density between nodes is one of the fundamental indicators of community presence in the network. A community with a high edge density will have correlations between nodes that extend beyond their immediate neighbors, denoted by motifs. Motifs are repetitive patterns of edges observed with high frequency in the network. We proposed the PCDMS method (Probabilistic Community Detection with Motif Structure) that detects communities by estimating the triangular motif in the network. This study employs structural density between nodes, a key concept in graph analysis. The proposed model has the advantage of using a probabilistic generative model that calculates the latent parameters of the probabilistic model and determines the community based on the likelihood of triangular motifs. The relationship between observing two pairs of nodes in multiple communities leads to an increasing likelihood estimation of the existence of a motif structure between them. The output of the proposed model is the intensity of each node in the communities. The efficiency and validity of the proposed method are evaluated through experimental work on both synthetic and real-world networks; the findings will show that the community identified by the proposed method is more accurate and dense than other algorithms with modularity, NMI, and F1score evaluation metrics.

Due to the growth of using various devices and applications in modern life, the amount of data available is skyrocketing, but labeling all of this data is beyond the reach of data scientists. Thus, it is necessary to categorize data with a small amount of labeled data. In fact, it should be possible to prioritize data for labeling. To achieve this goal in this study, we have used few-shot learning with active learning and also used the power of graph convolutional networks in classifying data with a graphical structure. To implement the proposed model, we use two graph convolutional networks in parallel to calculate the embedding and the importance of each node. Using the output of both networks, we create prototypes of classes, and then, we classify them according to the distance of each node of these prototypes. We have also used active learning to select data more intelligently, which improves the overall model performance. As well as this, we have tested our proposed model in the field of electronic commerce for tagging goods in big online stores, which encounter a large number of diverse products, where high accuracy categorization in a short time without the interference of human factor and with the help of artificial intelligence is needed to reduce costs. The results of implementing the model on the Amazon dataset and its comparison with the state-of-the-art models in this field show the superiority of our method.

Community detection in networks is a useful tool for detecting the behavioral and inclinations of users to a specific topic or title. Weighted, unweighted, directed, and undirected networks can all be used for detecting communities depending on the network structure and content. The proposed model framework for community detection is based on weighted networks. We use two important and effective concepts in graph analysis. The structural density between nodes is the first concept, and the second is the weight of edges between nodes. The proposed model advantage is using a probabilistic generative model that estimates the latent parameters of the probabilistic model and detecting the community based on the probability of the presence or absence of weighted edge. The output of the proposed model is the intensity of belonging each weighted node to the communities. A relationship between the observation of a pair of nodes in multiple communities and the probability of an edge with a high weight between them, is one of the important outputs that interpret the detected communities by finding relevancy between membership of nodes to communities and edge weight. Experiments are performed on real-world weighted networks and synthetic weighted networks to evaluate the performance and accuracy of the proposed algorithm. The results will show that the proposed algorithm is more density and accurate than other algorithms in weighted community detection.

Having a lot of labeled data is always a problem in machine learning issues. Even by collecting lots of data hardly, shift in data distribution might emerge because of differences in source and target domains. The shift would make the model to face with problems in test step. Therefore, the necessity of using domain adaptation emerges. There are three techniques in the field of domain adaptation namely discrepancy based, adversarial based and reconstruction based methods. For domain adaptation, adversarial learning approaches showed state-of-the-art performance. Although there are some comprehensive surveys about domain adaptation, we technically focus on adversarial based domain adaptation methods. We examine each proposed method in detail with respect to their structures and objective functions. The common aspect of proposed methods besides domain adaptation is considering the target labels are predicted as accurately as possible. It can be represented by some methods such as metric learning and multi-adversarial discriminators as are used in some of the papers. Also, we address the negative transfer issue for dissimilar distributions and propose the addition of clustering heuristics to the underlying structures for future research.