Learning Compositional Models for Object Categories from Small Sample Sets

An adaptive Bayesian contextual classification procedure that utilizes both spectral and spatial interpixel dependency contexts in estimation of statistics and classification is proposed. Essentially, this classifier is the constructive coupling of an adaptive classification procedure and a Bayesian contextual classification procedure. In this classifier, the joint prior probabilities of the classes of each pixel and its spatial neighbors are modeled by the Markov random field. The estimation of statistics and classification are performed in a recursive manner to allow the establishment of the positive-feedback process in a computationally efficient manner. Experiments with real hyperspectral data show that, starting with a small training sample set, this classifier can reach classification accuracies similar to that obtained by a pixelwise maximum likelihood pixel classifier with a very large training sample set. Additionally, classification maps are produced that have significantly less speckle error.

The ability to detect human actions or gestures is key for a wide range of applications that involve interactions between humans and robots. These actions are patterns that have a particular spatio-temporal structure. This paper presents an approach for encoding such patterns using spike-timing networks with axonal conductance delays. The proposed method brings the following contributions: first, it enables the encoding of patterns in an unsupervised manner. Second, it allows us to create models of specific patterns using a very small set of training samples, in contrast with standard pattern recognition approaches that typically require large amounts of training data. Based on these models, the method further enables classification of new patterns using a longest-common subsequence approach for matching between patterns of activated neurons. Third, the approach is invariant to scale and translation and thus it enables generalization across multiple scales and positions. Fourth, the approach also enables early recognition of patterns from only partial information about the pattern. The proposed method is validated on a set of gestures representing the digits from 0 to 9, extracted from video data of a human drawing the corresponding digits. The results are also compared with other state of the art pattern recognition algorithms.

Multi-agent simulations facilitate modelling the complex behaviours through simple rules codified within the agents. However, the manual exploration of effective rules is generally prohibitive or at least sub-optimal in these types of simulations. Evolutionary techniques can effectively explore the rule and parameter space for high fidelity simulations; however, the pairing of these two techniques is challenging in classes of problems where the evaluation of simulation fidelity is reliant on human judgement. In this work we present a machine learning approach to evolve high fidelity low complexity sheepdog herding simulations. A multi-objective evolutionary algorithm is applied to evolve simulations with both high fidelity and low complexity as the two objectives to be optimised. Fidelity is measured via a machine learning system trained using a small set of training samples of human judgements; whereas the complexity is measured using the number of rules and parameters used to codify the agents in the system. The experimental results demonstrate the effectiveness of the approach in evolving high fidelity and low complexity multi-agent simulations when the fidelity cannot be measured objectively.

Internet of autonomous unmanned vehicles (IAUV) is a global network of sensors, robots, and autonomous vehicles. Person re‐identification (Re‐ID) is an important intelligent transportation application in IAUV, which needs to be solved using artificial intelligence algorithms. In this study, a domain adaptation metric learning method embedded with structural information (called DAML‐ESI) is designed for person Re‐ID in IAUV. Due to the lack of labeling information in the target domain, DAML‐ESI realizes person Re‐ID with the help of the discriminative and structural information of pedestrian images of related domains. DAML‐ESI projects pedestrian images selected from different domains into a common metric space and establishes a discriminative metric learning model, which requires that the positive sample pair be mapped to a point, and the distance distribution of the negative sample pair be mapped to a fixed value. The projection matrix learned by DAML‐ESI is used to eliminate the distribution differences between different domains, and the distance metric is used to ensure that the learned metric learning model has strong discriminative ability in the metric space. To verify the effectiveness of DAML‐ESI, experimental comparisons are conducted on three person Re‐ID datasets, and DAML‐ESI achieves satisfactory recognition performance.

Adaptive online extreme learning machine by regulating forgetting factor by concept drift map

Sparsity in function and derivative approximation via the empirical feature space

The performance of visual object tracking depends largely on the target appearance model. Benefited from the success of CNN in feature extraction, recent studies have paid much attention to CNN representation learning and feature fusion model. However, the existing feature fusion models ignore the relation between the features of different layers. In this paper, we propose a deep feature fusion model based on the siamese network by considering the connection between feature maps of CNN. To tackle the limitation of different feature map sizes in CNN, we propose to fuse different resolution feature maps by introducing de-convolutional layers in the offline training stage. Specifically, a top-down modulation is adopted for feature fusion. In the tracking stage, a simple matching operation between the fused feature of the examplar and search region is conducted with the learned model, which can maintain the real-time tracking speed. Experimental results show that, the proposed method obtains favorable tracking accuracy against the state-of-the-art trackers with a real-time tracking speed.

In recent years, the quick deployment of Machine Learning in Small and Medium-sized Enterprises (SMEs) has been delayed by the complexity of model design and the lack of skilled personnel. However, traditional Automated Machine Learning (AutoML) models often restart learning from scratch for each task, resulting in high computational costs and delayed optimization. Therefore, to address these issues, a meta-learning-enhanced AutoML with Preference-Based Optimization (MetaAutoMPO) framework was proposed. Initially, it integrates a meta-learning layer that extracts taskspecific meta-features and retrieves optimal model configurations from similar tasks before starting the training process. Subsequently, regression models were constructed using a Multilayer Perceptron (MLP) with hyperparameters tuned via Bayesian optimization. The learned models are then used as objective functions in a reference-point-based Nondominated Sorting Genetic Algorithm II (NSGA-II) to generate preference-aware process parameterizations. Finally, experimental validation demonstrates that MetaAutoMPO achieves <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{3 5 \%}$</tex> faster convergence and improves productivity (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\uparrow 3.19 \%$</tex>) while reducing defect rates (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\downarrow$</tex> 2.15%) compared with conventional AutoML methods.

Abstract. Despite showing great success of applications in many commercial fields, machine learning and data science models generally show limited success in many scientific fields, including hydrology (Karpatne et al., 2017). The approach is often criticized for its lack of interpretability and physical consistency. This has led to the emergence of new modelling paradigms, such as theory-guided data science (TGDS) and physics-informed machine learning. The motivation behind such approaches is to improve the physical meaningfulness of machine learning models by blending existing scientific knowledge with learning algorithms. Following the same principles in our prior work (Chadalawada et al., 2020), a new model induction framework was founded on genetic programming (GP), namely the Machine Learning Rainfall–Runoff Model Induction (ML-RR-MI) toolkit. ML-RR-MI is capable of developing fully fledged lumped conceptual rainfall–runoff models for a watershed of interest using the building blocks of two flexible rainfall–runoff modelling frameworks. In this study, we extend ML-RR-MI towards inducing semi-distributed rainfall–runoff models. The meaningfulness and reliability of hydrological inferences gained from lumped models may tend to deteriorate within large catchments where the spatial heterogeneity of forcing variables and watershed properties is significant. This was the motivation behind developing our machine learning approach for distributed rainfall–runoff modelling titled Machine Induction Knowledge Augmented – System Hydrologique Asiatique (MIKA-SHA). MIKA-SHA captures spatial variabilities and automatically induces rainfall–runoff models for the catchment of interest without any explicit user selections. Currently, MIKA-SHA learns models utilizing the model building components of two flexible modelling frameworks. However, the proposed framework can be coupled with any internally coherent collection of building blocks. MIKA-SHA's model induction capabilities have been tested on the Rappahannock River basin near Fredericksburg, Virginia, USA. MIKA-SHA builds and tests many model configurations using the model building components of the two flexible modelling frameworks and quantitatively identifies the optimal model for the watershed of concern. In this study, MIKA-SHA is utilized to identify two optimal models (one from each flexible modelling framework) to capture the runoff dynamics of the Rappahannock River basin. Both optimal models achieve high-efficiency values in hydrograph predictions (both at catchment and subcatchment outlets) and good visual matches with the observed runoff response of the catchment. Furthermore, the resulting model architectures are compatible with previously reported research findings and fieldwork insights of the watershed and are readily interpretable by hydrologists. MIKA-SHA-induced semi-distributed model performances were compared against existing lumped model performances for the same basin. MIKA-SHA-induced optimal models outperform the lumped models used in this study in terms of efficiency values while benefitting hydrologists with more meaningful hydrological inferences about the runoff dynamics of the Rappahannock River basin.

Learning a high-dimensional dense representation for vocabulary terms, also known as a word embedding, has recently attracted much attention in natural language processing and information retrieval tasks. The embedding vectors are typically learned based on term proximity in a large corpus. This means that the objective in well-known word embedding algorithms, e.g., word2vec, is to accurately predict adjacent word(s) for a given word or context. However, this objective is not necessarily equivalent to the goal of many information retrieval (IR) tasks. The primary objective in various IR tasks is to capture relevance instead of term proximity, syntactic, or even semantic similarity. This is the motivation for developing unsupervised relevance-based word embedding models that learn word representations based on query-document relevance information. In this paper, we propose two learning models with different objective functions; one learns a relevance distribution over the vocabulary set for each query, and the other classifies each term as belonging to the relevant or non-relevant class for each query. To train our models, we used over six million unique queries and the top ranked documents retrieved in response to each query, which are assumed to be relevant to the query. We extrinsically evaluate our learned word representation models using two IR tasks: query expansion and query classification. Both query expansion experiments on four TREC collections and query classification experiments on the KDD Cup 2005 dataset suggest that the relevance-based word embedding models significantly outperform state-of-the-art proximity-based embedding models, such as word2vec and GloVe.

Based on the urgent need for agricultural intelligence in various northern countries, this study develops an algorithm for crop pest and disease identification based on the basic theory of convolutional neural networks (CNN). The existing related algorithms are basically based on platforms that can carry multiple cameras to work to (e.g., tractors or combine harvesters). Such algorithms are not applicable to cultivation scenarios such as terraces. Therefore, in this study, unlike other designs based on multiple nets, the learning model of this learning uses only a single net for the train dataset. The main research objective of this learning is to achieve satisfactory result using a simplified CNN model design. The main approach of this learning model is to use the convolutional layer in the model to learn the crop pest and disease features in the given images, then use the pooling layer to reduce the number of parameters, and finally use the Fully connected layer to classify them into two categories: diseased and healthy. After learning through the train dataset with a sample size of 7000, the test dataset was imported into the completed learned model for testing, and the result was an accuracy of 94.5% and a loss rate of 0.1385.

State-of-the-art 2D image compression schemes rely on the power of convolutional neural networks (CNNs). Although CNNs offer promising perspectives for 2D image compression, extending such models to omnidirectional images is not straightforward. First, omnidirectional images have specific spatial and statistical properties that can not be fully captured by current CNN models. Second, basic mathematical operations composing a CNN architecture, e.g., translation and sampling, are not well-defined on the sphere. In this paper, we study the learning of representation models for omnidirectional images and propose to use the properties of HEALPix uniform sampling of the sphere to redefine the mathematical tools used in deep learning models for omnidirectional images. In particular, we: i) propose the definition of a new convolution operation on the sphere that keeps the high expressiveness and the low complexity of a classical 2D convolution; ii) adapt standard CNN techniques such as stride, iterative aggregation, and pixel shuffling to the spherical domain; and then iii) apply our new framework to the task of omnidirectional image compression. Our experiments show that our proposed on-the-sphere solution leads to a better compression gain that can save 13.7% of the bit rate compared to similar learned models applied to equirectangular images. Also, compared to learning models based on graph convolutional networks, our solution supports more expressive filters that can preserve high frequencies and provide a better perceptual quality of the compressed images. Such results demonstrate the efficiency of the proposed framework, which opens new research venues for other omnidirectional vision tasks to be effectively implemented on the sphere manifold.

Many real-world problems, like microchip design, can be modeled by means of the well-known traveling salesman problem (TSP). Many instances of this problem can be found in the literature. Although several optimization algorithms have been applied to TSP instances, the selection of the more promising algorithm is, in practice, a difficult decision. In this paper, a new meta-learning-based approach is investigated for the selection of optimization algorithms for TSP instances. Essentially, a learning model is trained with TSP instances for which the performance of a set of optimization algorithms is known a priori. Then, the learned model is used to predict the best algorithm for a new TSP instance. Each instance is described by meta-features that capture characteristics of the TSP that affect the performance of the optimization algorithms. Given that the best solution for a given TSP instance can be obtained by several algorithms, the meta-learning problem is considered here to be a multi-label classification problem. Several experiments illustrate the performance of the proposed approach, with promising results.

Non-Hispanic white (White) populations are overrepresented in medical studies. Potential healthcare disparities can happen when machine learning models, used in diabetes technologies, are trained on data from primarily White patients. We aimed to evaluate algorithmic fairness in glucose predictions. This study utilized continuous glucose monitoring (CGM) data from 101 White and 104 Black participants with type 1 diabetes collected by the JAEB Center for Health Research, US. Long short-term memory (LSTM) deep learning models were trained on 11 datasets of different proportions of White and Black participants and tailored to each individual using transfer learning to predict glucose 60 minutes ahead based on 60-minute windows. Root mean squared errors (RMSE) were calculated for each participant. Linear mixed-effect models were used to investigate the association between racial composition and RMSE while accounting for age, sex, and training data size. A median of 9 weeks (IQR: 7, 10) of CGM data was available per participant. The divergence in performance (RMSE slope by proportion) was not statistically significant for either group. However, the slope difference (from 0% White and 100% Black to 100% White and 0% Black) between groups was statistically significant (p = 0.02), meaning the RMSE increased 0.04 [0.01, 0.08] mmol/L more for Black participants compared to White participants when the proportion of White participants increased from 0 to 100% in the training data. This difference was attenuated in the transfer learned models (RMSE: 0.02 [-0.01, 0.05] mmol/L, p = 0.20). The racial composition of training data created a small statistically significant difference in the performance of the models, which was not present after using transfer learning. This demonstrates the importance of diversity in datasets and the potential value of transfer learning for developing more fair prediction models.

Deep learning models have been deployed in an increasing number of edge and mobile devices to provide healthcare. These models rely on training with a tremendous amount of labeled data to achieve high accuracy. However, for medical applications such as dermatological disease diagnosis, the private data collected by mobile dermatology assistants exist on distributed mobile devices of patients, and each device only has a limited amount of data. Directly learning from limited data greatly deteriorates the performance of learned models. Federated learning (FL) can train models by using data distributed on devices while keeping the data local for privacy. Existing works on FL assume all the data have ground-truth labels. However, medical data often comes without any accompanying labels since labeling requires expertise and results in prohibitively high labor costs. The recently developed self-supervised learning approach, contrastive learning (CL), can leverage the unlabeled data to pre-train a model for learning data representations, after which the learned model can be fine-tuned on limited labeled data to perform dermatological disease diagnosis. However, simply combining CL with FL as federated contrastive learning (FCL) will result in ineffective learning since CL requires diverse data for accurate learning but each device in FL only has limited data diversity. In this work, we propose an on-device FCL framework for dermatological disease diagnosis with limited labels. Features are shared among devices in the FCL pre-training process to provide diverse and accurate contrastive information without sharing raw data for privacy. After that, the pre-trained model is fine-tuned with local labeled data independently on each device or collaboratively with supervised federated learning on all devices. Experiments on dermatological disease datasets show that the proposed framework effectively improves the recall and precision of dermatological disease diagnosis compared with state-of-the-art methods.