Suitable Embedding Research Articles

Co-regularized optimal high-order graph embedding for multi-view clustering

Real-world applications frequently involve multiple data modalities in the same samples, which are regarded as multi-view data. Multi-view clustering has been studied extensively in recent years to demonstrate embedded heterogeneity. However, most existing methods emphasize low-order correlation in multiple views, whereas approaches that incorporate high-order correlation are limited by the equal view-specific significance problem or a trade-off between global and local consistency. In this paper, we propose a co-regularized optimal graph-based clustering method known as Co-MSE, which integrates the correlation of different orders. By integrating the first-order and second-order similarities, the local structure is preserved, while an optimized embedding representation for multi-view data is obtained simultaneously through co-regularization. We demonstrate that Co-MSE can aid in providing a more suitable embedding representation and further enable satisfactory clustering performance. Extensive experiments on real-world datasets confirm the effectiveness and advantages of the proposed method.

HindiPersonalityNet: Personality Detection in Hindi Conversational Data Using Deep Learning with Static Embedding

Personality detection along with other behavioral and cognitive assessment can essentially explain why people act the way they do and can be useful to various online applications such as recommender systems, job screening, matchmaking, and counseling. Additionally, psychometric natural language processing relying on textual cues and distinctive markers in writing style within conversational utterances reveals signs of individual personalities. This work demonstrates a text-based deep neural model, HindiPersonalityNet, of classifying conversations into three personality categories (ambivert, extrovert, introvert) for detecting personality in Hindi conversational data. The model utilizes a gated recurrent unit with BioWordVec embeddings for text classification and is trained/tested on a novel dataset, शख्सियत (pronounced as Shakhsiyat) curated using dialogues from an Indian crime-thriller drama series, Aarya . The model achieves an F1-score of 0.701 and shows the potential for leveraging conversational data from various sources to understand and predict a person's personality traits. It exhibits the ability to capture both semantic and long-distance dependencies in conversations and establishes the effectiveness of our dataset as a benchmark for personality detection in Hindi dialogue data. Further, a comprehensive comparison of various static and dynamic word embedding is done on our standardized dataset to ascertain the most suitable embedding method for personality detection.

Graph-Regularized Non-Negative Matrix Factorization for Single-Cell Clustering in scRNA-Seq Data.

The advent of single-cell RNA sequencing (scRNA-seq) has brought forth fresh perspectives on intricate biological processes, revealing the nuances and divergences present among distinct cells. Accurate single-cell analysis is a crucial prerequisite for in-depth investigation into the underlying mechanisms of heterogeneity. Due to various technical noises, like the impact of dropout values, scRNA-seq data remains challenging to interpret. In this work, we propose an unsupervised learning framework for scRNA-seq data analysis (aka Sc-GNNMF). Based on the non-negativity and sparsity of scRNA-seq data, we propose employing graph-regularized non-negative matrix factorization (GNNMF) algorithm for the analysis of scRNA-seq data, which involves estimating cell-cell sparse similarity and gene-gene sparse similarity through Laplacian kernels and p-nearest neighbor graphs ( p-NNG). By assuming intrinsic geometric local invariance, we use a weighted p-nearest known neighbors ( p-NKN) to optimize the scRNA-seq data. The optimized scRNA-seq data then participates in the matrix decomposition process, promoting the closeness of cells with similar types in cell-gene data space and determining a more suitable embedding space for clustering. Sc-GNNMF demonstrates superior performance compared to other methods and maintains satisfactory compatibility and robustness, as evidenced by experiments on 11 real scRNA-seq datasets. Furthermore, Sc-GNNMF yields excellent results in clustering tasks, extracting useful gene markers, and pseudo-temporal analysis.

A semantic-based model with a hybrid feature engineering process for accurate spam detection

Detecting spam emails is essential to maintaining the security and integrity of email communication. Existing research has made significant progress in developing effective spam detection models, but challenges remain in improving classification performance and adaptability to evolving spamming techniques. In this study, we propose a novel spam detection model with a comprehensive feature engineering approach that combines term frequency-inverse document frequency (TF-IDF) vectorizer and word embedding features to optimize the feature space. Our contribution lies in integrating semantic-based word embeddings, leveraging pre-existing knowledge to capture the semantic meaning of words and enhance the representation of email texts. To identify the most suitable word embedding technique for our model, we evaluated GloVe, Word2Vec, and FastText. GloVe was selected for its better performance, which is the result of its pre-training on a large and diverse text corpus. Furthermore, the model was evaluated without word embeddings, which did not exhibit the same effectiveness level as our word embedding-based model. Additionally, we utilized the support vector machine as a classifier and hyperparameter tuning technique to identify our model’s most effective parameter values. The proposed model was tested on two datasets. The experimental results showed that our model outperformed the other models discussed in the literature, achieving an accuracy of 99.5% on the SpamAssassin dataset, and 99.28% on the Enron-Spam dataset.

An approach for identifying complementary patents based on deep learning

Current studies on technology mining and analysis often focus on patent similarity, with relatively limited research on patent complementarity. Specifically, the hierarchical relationships among patents are seldom used and a standardized complementary patents dataset has not been established. In addition, it is necessary to utilize both network structure features and text content features of patents, and find the most suitable representation learning method for them. Finally, the relationships among different dimensions of feature representations are complex, making it essential to learn the contributions of each dimension considering complex interactions. Therefore, this paper first constructs a complementary patents dataset using hierarchical relationships contained in IPC numbers. Secondly, we design three types of embedding methods for patent semantic representation, including network embedding, text embedding and fusion embedding. Thirdly, we propose a deep learning framework enhanced by the CBAM (Convolutional Block Attention Module) to deal with the complex interactions between different dimensions of patent representation. The result shows that the proposed method CompGCN combined with ESimCSE_Attention performs best for complementary patent identification and the F1 score reaches 95.76 %. In addition, HeGAN and ESimCSE_Attention are the most suitable embedding methods for network structure and text content respectively. These results not only validate the effectiveness of the proposed approach, but also provide helpful and useful suggestions for method selection and complex relationships mining.

Human Versus Machine Intelligence: Assessing Natural Language Generation Models Through Complex Systems Theory.

The introduction of Transformer architectures - with the self-attention mechanism - in automatic Natural Language Generation (NLG) is a breakthrough in solving general task-oriented problems, such as the simple production of long text excerpts that resemble ones written by humans. While the performance of GPT-X architectures is there for all to see, many efforts are underway to penetrate the secrets of these black-boxes in terms of intelligent information processing whose output statistical distributions resemble that of natural language. In this work, through the complexity science framework, a comparative study of the stochastic processes underlying the texts produced by the English version of GPT-2 with respect to texts produced by human beings, notably novels in English and programming codes, is offered. The investigation, of a methodological nature, consists first of all of an analysis phase in which the Multifractal Detrended Fluctuation Analysis and the Recurrence Quantification Analysis - together with Zipf's law and approximate entropy - are adopted to characterize long-term correlations, regularities and recurrences in human and machine-produced texts. Results show several peculiarities and trends in terms of long-range correlations and recurrences in the last case. The synthesis phase, on the other hand, uses the complexity measures to build synthetic text descriptors - hence a suitable text embedding - which serve to constitute the features for feeding a machine learning system designed to operate feature selection through an evolutionary technique. Using multivariate analysis, it is then shown the grouping tendency of the three analyzed text types, allowing to place GTP-2 texts in between natural language texts and computer codes. Similarly, the classification task demonstrates that, given the high accuracy obtained in the automatic discrimination of text classes, the proposed set of complexity measures is highly informative. These interesting results allow us to add another piece to the theoretical understanding of the surprising results obtained by NLG systems based on deep learning and let us to improve the design of new informetrics or text mining systems for text classification, fake news detection, or even plagiarism detection.

Improving static and temporal knowledge graph embedding using affine transformations of entities

To find a suitable embedding for a knowledge graph (KG) remains a big challenge nowadays. By measuring the distance or plausibility of triples and quadruples in static and temporal knowledge graphs, many reliable knowledge graph embedding (KGE) models are proposed. However, these classical models may not be able to represent and infer various relation patterns well, such as TransE cannot represent symmetric relations, DistMult cannot represent inverse relations, RotatE cannot represent multiple relations, etc.. In this paper, we improve the ability of these models to represent various relation patterns by introducing the affine transformation framework. Specifically, we first utilize a set of affine transformations related to each relation or timestamp to operate on entity vectors, and then these transformed vectors can be applied not only to static KGE models, but also to temporal KGE models. The main advantage of using affine transformations is their good geometry properties with interpretability. Our experimental results demonstrate that the proposed intuitive design with affine transformations provides a statistically significant increase in performance with adding a few extra processing steps and keeping the same number of embedding parameters. Taking TransE as an example, we employ the scale transformation (the special case of an affine transformation). Surprisingly, it even outperforms RotatE to some extent on various datasets. We also introduce affine transformations into RotatE, Distmult, ComplEx, TTransE and TComplEx respectively, and experiments demonstrate that affine transformations consistently and significantly improve the performance of state-of-the-art KGE models on both static and temporal knowledge graph benchmarks.

Multimodal few-shot classification without attribute embedding

Multimodal few-shot learning aims to exploit complementary information inherent in multiple modalities for vision tasks in low data scenarios. Most of the current research focuses on a suitable embedding space for the various modalities. While solutions based on embedding provide state-of-the-art results, they reduce the interpretability of the model. Separate visualization approaches enable the models to become more transparent. In this paper, a multimodal few-shot learning framework that is inherently interpretable is presented. This is achieved by using the textual modality in the form of attributes without embedding them. This enables the model to directly explain which attributes caused it to classify an image into a particular class. The model consists of a variational autoencoder to learn the visual latent representation, which is combined with a semantic latent representation that is learnt from a normal autoencoder, which calculates a semantic loss between the latent representation and a binary attribute vector. A decoder reconstructs the original image from concatenated latent vectors. The proposed model outperforms other multimodal methods when all test classes are used, e.g., 50 classes in a 50-way 1-shot setting, and is comparable for lesser number of ways. Since raw text attributes are used, the datasets for evaluation are CUB, SUN and AWA2. The effectiveness of interpretability provided by the model is evaluated by analyzing how well it has learnt to identify the attributes.

Counterfeiting in Depth Synthesis based on Digital Watermarking

The purpose of this paper is to discuss and apply digital watermarking technology to solve the forgery problem in depth synthesis. With the rapid development of deep synthesis technology and its application in various fields, it is particularly important to protect the authenticity and integrity of digital content. Based on the understanding of digital watermarking, this paper explores an experimental design, which uses watermarking embedding and extraction algorithms and forgery detection technology to solve the problem of deep forgery, protect the copyright, integrity and anti-copy of digital products. In order to improve the robustness and reliability of the watermark, a suitable watermark embedding and extraction algorithm is designed by analyzing the characteristics of deep synthesis forged media in the experimental process. Then select the data set containing the original digital media and the deep synthetic forged samples, extract the features of the two, and find out the features that distinguish the differences between the two. Finally, the forgery detection technology is used to evaluate the performance of digital watermarking technology in depth forgery detection. In this paper, digital watermarking technology is used to provide an effective solution to the problem of forgery in depth synthesis, which can be applied to protect intellectual property rights, prevent tampering and forgery, and protect the authenticity and integrity of digital media content.

Visual image encryption algorithm based on compressed sensing and 2D cosine -type logistic map

A new two-dimensional cosine-type Logistic map (2D-CTLM) is proposed in this paper. Performance analysis shows that 2D-CTLM has good ergodicity, complex behavior, and a wide range of chaotic regions. To study its application, an adaptive embedded high visual security image encryption scheme is proposed by combining the system with two-dimensional compression sensing to address the shortcomings of fixed embedding positions and low transmission efficiency in existing schemes. The embedding position is fixed when the ciphertext image is embedded into the carrier image. When embedded in an inappropriate location, it will affect the decryption quality and the imperceptibility of the steganographic image. Firstly, the initial value of 2D-CTLM system is generated by using the information entropy of plaintext image and counter to generate random measurement matrix, which is used to compress and encrypt plaintext image from two directions at the same time. Secondly, the compressed cipher image is scrambled by the initial matrix generated by the GOL (game of life) rules in two-dimensional cellular automata to obtain the ciphertext image. Finally, the most suitable embedding position in the carrier image is found by combining information entropy and edge entropy. Then the unquantized ciphertext image is embedded into it by ST (slant transform) embedding method to obtain a high visual security image. To improve the transmission efficiency, the color carrier image can be selected to transmit three gray plaintext images at the same time, maintain high-quality reconstruction effect. SHA-512 and the information entropy of plaintext image are used to generate the initial value of encryption and measurement matrix, enhance the correlation between the algorithm and plaintext image, and effectively resist known plaintext and selective plaintext attacks. Simulation results and performance analysis show that compared with the existing encryption schemes, the encryption scheme has excellent visual effect, good decryption quality and good robustness.

Chemical synthesized nano-Li3V2O5 anode for high-rate rechargeable Li-ion batteries at low temperature

Lithium-ion batteries (LIBs) are widely used and considered as an ideal power supply for different applications. However, severe power/energy loss and lithium dendrite growth at low temperatures are still major problems for the graphite-based LIBs. Here we develop a highly stable nano ω-Li3V2O5 anode (named n-LVO-H) with a suitable lithium embedding potential via a chemical method. Benefiting from a short Li+ transport path, the product n-LVO-H anode accelerates the diffusion of Li+ in the particles. In addition, a robust solid-electrolyte interface layer built on the n-LVO-H particles using 1,3-dioxolane (DOL) based electrolyte with low desolvation energy facilitates Li+ transport at the interface. Such an unconventional combination of nano-size anode material with DOL electrolyte renders the n-LVO-H cell high C-rate capacity (207.1 mAh g−1 at 10C), while enjoying cycling stability (92.93 % retention at 10C after 1000 cycles) and low-temperature tolerance (248.1 mAh g−1 at -20 °C and 0.2C) concurrently.

Prototype Theory Meets Word Embedding: A Novel Approach for Text Categorization via Granular Computing

The problem of the information representation and interpretation coming from senses by the brain has plagued scientists for decades. The same problems, from a different perspective, hold in automated Pattern Recognition systems. Specifically, in solving various NLP tasks, an ever better and richer semantic representation of text as a set of features is needed and a plethora of text embedding techniques in algebraic spaces are continuously provided by researchers. These spaces are well suited to be conceived as conceptual spaces in light of the Gärdenfors’s Conceptual Space theory, which, within the Cognitive Science paradigm, seeks a geometrization of thought that bridges the gap between an associative lower level and a symbolic higher level in which information is organized and processed and where inductive reasoning is appropriate. Granular Computing can offer the toolbox for granulating text that can be represented by more abstract entities than words, offering a good hierarchical representation of the text embedded in an algebraic space driving Machine Learning applications, specifically, in text mining tasks. In this paper, the Conceptual Space Theory, the Granular Computing approach and Machine Learning are bound in a novel common framework for solving some text categorization tasks with both standard classifiers suited for working with mathbb {R}^n vectors and a Recurrent Neural Network (RNN) — an LSTM — able to deal with sequences. Instead of working with word vectors, the algorithms process more abstract entities (concepts), where patterns, in a first approach, are obtained through the construction of a symbolic histogram starting from a suitable set of information granules, representing a document as a distribution of concepts. For the RNN case, as a further novelty, a text is represented as a random walk over prototypes within the conceptual space synthesized over a suitable text embedding procedure. A comparison of the performance and a critical discussion are offered for both a neural embedding technique and the well-known LSA, showing how the conceptual level leads also to Knowledge Discovery applications.

Supervised Relation Extraction Between Suicide-Related Entities and Drugs: Development and Usability Study of an Annotated PubMed Corpus.

Drug-induced suicide has been debated as a crucial issue in both clinical and public health research. Published research articles contain valuable data on the drugs associated with suicidal adverse events. An automated process that extracts such information and rapidly detects drugs related to suicide risk is essential but has not been well established. Moreover, few data sets are available for training and validating classification models on drug-induced suicide. This study aimed to build a corpus of drug-suicide relations containing annotated entities for drugs, suicidal adverse events, and their relations. To confirm the effectiveness of the drug-suicide relation corpus, we evaluated the performance of a relation classification model using the corpus in conjunction with various embeddings. We collected the abstracts and titles of research articles associated with drugs and suicide from PubMed and manually annotated them along with their relations at the sentence level (adverse drug events, treatment, suicide means, or miscellaneous). To reduce the manual annotation effort, we preliminarily selected sentences with a pretrained zero-shot classifier or sentences containing only drug and suicide keywords. We trained a relation classification model using various Bidirectional Encoder Representations from Transformer embeddings with the proposed corpus. We then compared the performances of the model with different Bidirectional Encoder Representations from Transformer-based embeddings and selected the most suitable embedding for our corpus. Our corpus comprised 11,894 sentences extracted from the titles and abstracts of the PubMed research articles. Each sentence was annotated with drug and suicide entities and the relationship between these 2 entities (adverse drug events, treatment, means, and miscellaneous). All of the tested relation classification models that were fine-tuned on the corpus accurately detected sentences of suicidal adverse events regardless of their pretrained type and data set properties. To our knowledge, this is the first and most extensive corpus of drug-suicide relations.

Impact of word embedding models on text analytics in deep learning environment: a review.

The selection of word embedding and deep learning models for better outcomes is vital. Word embeddings are an n-dimensional distributed representation of a text that attempts to capture the meanings of the words. Deep learning models utilize multiple computing layers to learn hierarchical representations of data. The word embedding technique represented by deep learning has received much attention. It is used in various natural language processing (NLP) applications, such as text classification, sentiment analysis, named entity recognition, topic modeling, etc. This paper reviews the representative methods of the most prominent word embedding and deep learning models. It presents an overview of recent research trends in NLP and a detailed understanding of how to use these models to achieve efficient results on text analytics tasks. The review summarizes, contrasts, and compares numerous word embedding and deep learning models and includes a list of prominent datasets, tools, APIs, and popular publications. A reference for selecting a suitable word embedding and deep learning approach is presented based on a comparative analysis of different techniques to perform text analytics tasks. This paper can serve as a quick reference for learning the basics, benefits, and challenges of various word representation approaches and deep learning models, with their application to text analytics and a future outlook on research. It can be concluded from the findings of this study that domain-specific word embedding and the long short term memory model can be employed to improve overall text analytics task performance.

Rational design of anti-Kasha photoemission from a biazulene core embedded in an antiaromatic/aromatic hybrid.

The violation of the Kasha photoemission rule in organic molecules has intrigued chemists since their discovery, being always of relevance given its connection with unique electronic properties of molecules. However, an understanding of the molecular structure-anti-Kasha property relationship in organic materials has not been well-established, possibly because of the few existing cases available, limiting their prospective exploration and ad hoc design. Here we introduce a novel strategy to design organic emitters from high excited states combining intramolecular J-coupling of anti-Kasha chromophores with the hindering of vibrationally-induced non-radiative decay channels by enforcing molecular rigidity. We apply our approach to the integration of two antiparallel azulene units bridged with one heptalene all inserted into a polycyclic conjugated hydrocarbon (PCH). With the help of quantum chemistry calculations, we identify a suitable PCH embedding structure and predict its anti-Kasha emission from the third high energy excited singlet state. Finally, steady fluorescence and transient absorption spectroscopy studies corroborate the photophysical properties in a recently synthesized chemical derivative with this pre-designed structure.

HOTTEST: Hate and Offensive content identification in Tamil using Transformers and Enhanced STemming

Offensive content or hate speech is defined as any form of communication that aims to annoy, harass, disturb, or anger an individual or community based on factors such as faith, ethnicity, appearance, or sexuality. Nowadays, offensive content posted in regional languages increased due to the popularity of social networks and other apps usage by common people. This work proposes a method to detect and identify hate speech or offensive content in Tamil. We have used the HASOC 2021 data set that contains YouTube comments in Tamil language and written in Tamil script. In this research work, an attempt is made to find suitable embedding techniques for Tamil text representation by applying TF-IDF and pre-trained transformer models like BERT, XLM-RoBERTa, IndicBERT, mBERT, TaMillion, and MuRIL. As Tamil is a morphologically rich language, a detailed analysis is made to study the performance of hate speech detection in Tamil by applying enhanced stemming algorithms. An extensive experimental study was performed with different classifiers such as logistic regression, SVM, stochastic Gradient Descent, decision tree, and ensemble learning models in combination with the above techniques. The results of this detailed experimental study show that stop word removal produces mixed results and does not guarantee improvement in the performance of the classifier to detect offensive content for Tamil data. However, the performance on stemmed data shows a significant improvement over un-stemmed data in Tamil texts. As the data is highly imbalanced, we also combined an oversampling/downsampling technique to analyze its role in designing the best offensive classifier for Tamil text. The highest performance was achieved by a combination of stemming the text data, embedding it with the multi-lingual model MuRIL and using a majority voting ensemble as the downstream classifier. We have achieved the F1-score of 84% and accuracy of 86% for detecting offensive content in Tamil YouTube comments.

The Most Likely Evolution of Diffusing and Vanishing Particles: Schrödinger Bridges with Unbalanced Marginals

Stochastic flows of an advective-diffusive nature are ubiquitous in physical sciences. Of particular interest is the problem to reconcile observed marginal distributions with a given prior posed by E. Schrodinger in 1932/32 and known as the Schrodinger Bridge Problem (SBP). Due to its fundamental significance, interest in SBP has in recent years enticed a broad spectrum of disciplines. Yet, while the mathematics and applications of SBP have been developing at a considerable pace, accounting for marginals of unequal mass has received scant attention; the problem to interpolate between unbalanced marginals has been approached by introducing source/sink terms in an Adhoc manner. Nevertheless, losses are inherent in many physical processes and, thereby, models that account for lossy transport may also need to be reconciled with observed marginals following Schrodinger's dictum; that is, to adjust the probability of trajectories of particles, including those that do not make it to the terminal observation point, so that the updated law represents the most likely way that particles may have been transported, or vanished, at some intermediate point. Thus, the purpose of this work is to develop such a natural generalization of the SBP for stochastic evolution with losses, whereupon particles are "killed" according to a probabilistic law. Through a suitable embedding, we turn the problem into an SBP for stochastic processes that combine diffusive and jump characteristics. Then, following a large-deviations formalism, given a prior law that allows for losses, we ask for the most probable evolution of particles along with the most likely killing rate as the particles transition between the specified marginals. Our approach differs sharply from previous work involving a Feynman-Kac multiplicative reweighing of the reference measure: The latter, as we argue, is far from Schrodinger's quest.

Construction of a traversable wormhole from a suitable embedding function

In this work, we construct a traversable wormhole by providing a suitable embedding function ensuring the fulfilling of the flaring-out condition. The solution contains free parameters that are reduced through the study of the acceptable conditions of a traversable wormhole. We compute both the quantifier of exotic matter and the quasi-normal modes through the 13th order WKB as a function of the remaining free parameters. We obtain that the wormhole geometry can be sustained by a finite amount of exotic matter and seems to be stable under scalar perturbations.

General embedded cluster protocol for accurate modeling of oxygen vacancies in metal-oxides.

The O vacancy (Ov) formation energy, EOv, is an important property of a metal-oxide, governing its performance in applications such as fuel cells or heterogeneous catalysis. These defects are routinely studied with density functional theory (DFT). However, it is well-recognized that standard DFT formulations (e.g., the generalized gradient approximation) are insufficient for modeling the Ov, requiring higher levels of theory. The embedded cluster method offers a promising approach to compute EOv accurately, giving access to all electronic structure methods. Central to this approach is the construction of quantum(-mechanically treated) clusters placed within suitable embedding environments. Unfortunately, current approaches to constructing the quantum clusters either require large system sizes, preventing application of high-level methods, or require significant manual input, preventing investigations of multiple systems simultaneously. In this work, we present a systematic and general quantum cluster design protocol that can determine small converged quantum clusters for studying the Ov in metal-oxides with accurate methods, such as local coupled cluster with single, double, and perturbative triple excitations. We apply this protocol to study the Ov in the bulk and surface planes of rutile TiO2 and rock salt MgO, producing the first accurate and well-converged determinations of EOv with this method. These reference values are used to benchmark exchange-correlation functionals in DFT, and we find that all the studied functionals underestimate EOv, with the average error decreasing along the rungs of Jacob's ladder. This protocol is automatable for high-throughput calculations and can be generalized to study other point defects or adsorbates.

Visual Goal Human-Robot Communication Framework With Few-Shot Learning: A Case Study in Robot Waiter System

A conventional adopted method for operating a waiter robot is based on the static position control, where predefined goal positions are marked on a map. However, this solution is not optimal in a dynamic setting, such as in a coffee shop or an outdoor catering event, because the customers often change their positions. This article explores an alternative human-robot interface design where a human operator communicates the identity of the customer to the robot instead. Inspired by how human communicates, we propose a framework for communicating a visual goal to the robot, through interactive two-way communications. The framework exploits concepts from two machine learning domains: human-in-the-loop machine learning, where active learning is used to acquire informative data, and deep metric learning, where a suitable embedding can improve the learning ability of a classifier. We also propose novel class imbalance handling techniques, which aim to actively alleviate the class imbalance problem found to be important in this mode of communication. The framework is evaluated using publicly available pedestrian datasets. We demonstrate that the proposed framework can help reduce the number of required two-way interactions and increases the robustness of the predictive model. We successfully implement the framework on a mobile robot for a delivery service in a cafe-like environment. Through the online visual goal human-robot communication, the robot can detect, recognize, and autonomously navigate to the target customer.