Zero-shot learning is dedicated to solving the classification problem of unseen categories, while generalized zero-shot learning aims to classify the samples selected from both seen classes and unseen classes, in which “seen” and “unseen” classes indicate whether they can be used in the training process, and if so, they indicate seen classes, and vice versa. Nowadays, with the promotion of deep learning technology, the performance of zero-shot learning has been greatly improved. Generalized zero-shot learning is a challenging topic that has (...) promising prospects in many realistic scenarios. Although the zero-shot learning task has made gratifying progress, there is still a strong deviation between seen classes and unseen classes in the existing methods. Recent methods focus on learning a unified semantic-aligned visual representation to transfer knowledge between two domains, while ignoring the intrinsic characteristics of visual features which are discriminative enough to be classified by itself. To solve the above problems, we propose a novel model that uses the discriminative information of visual features to optimize the generative module, in which the generative module is a dual generation network framework composed of conditional VAE and improved WGAN. Specifically, the model uses the discrimination information of visual features, according to the relevant semantic embedding, synthesizes the visual features of unseen categories by using the learned generator, and then trains the final softmax classifier by using the generated visual features, thus realizing the recognition of unseen categories. In addition, this paper also analyzes the effect of the additional classifiers with different structures on the transmission of discriminative information. We have conducted a lot of experiments on six commonly used benchmark datasets. The experimental results show that our model outperforms several state-of-the-art methods for both traditional as well as generalized zero-shot learning. (shrink)

Neural network models have recently made striking progress in natural language processing, but they are typically trained on orders of magnitude more language input than children receive. What can these neural networks, which are primarily distributional learners, learn from a naturalistic subset of a single child's experience? We examine this question using a recent longitudinal dataset collected from a single child, consisting of egocentric visual data paired with text transcripts. We train both language-only and vision-and-language neural networks and analyze the (...) linguistic knowledge they acquire. In parallel with findings from Jeffrey Elman's seminal work, the neural networks form emergent clusters of words corresponding to syntactic (nouns, transitive and intransitive verbs) and semantic categories (e.g., animals and clothing), based solely on one child's linguistic input. The networks also acquire sensitivity to acceptability contrasts from linguistic phenomena, such as determiner-noun agreement and argument structure. We find that incorporating visual information produces an incremental gain in predicting words in context, especially for syntactic categories that are comparatively more easily grounded, such as nouns and verbs, but the underlying linguistic representations are not fundamentally altered. Our findings demonstrate which kinds of linguistic knowledge are learnable from a snapshot of a single child's real developmental experience. (shrink)

Cognitive Science, Volume 46, Issue 6, June 2022.

Silent gestures consist of complex multi‐articulatory movements but are now primarily studied through categorical coding of the referential gesture content. The relation of categorical linguistic content with continuous kinematics is therefore poorly understood. Here, we reanalyzed the video data from a gestural evolution experiment (Motamedi, Schouwstra, Smith, Culbertson, & Kirby, 2019), which showed increases in the systematicity of gesture content over time. We applied computer vision techniques to quantify the kinematics of the original data. Our kinematic analyses demonstrated that gestures (...) become more efficient and less complex in their kinematics over generations of learners. We further detect the systematicity of gesture form on the level of thegesture kinematic interrelations, which directly scales with the systematicity obtained on semantic coding of the gestures. Thus, from continuous kinematics alone, we can tap into linguistic aspects that were previously only approachable through categorical coding of meaning. Finally, going beyond issues of systematicity, we show how unique gesture kinematic dialects emerged over generations as isolated chains of participants gradually diverged over iterations from other chains. We, thereby, conclude that gestures can come to embody the linguistic system at the level of interrelationships between communicative tokens, which should calibrate our theories about form and linguistic content. (shrink)

The quality and quantity of training data are crucial to the performance of a deep-learning-based brain-computer interface system. However, it is not practical to record EEG data over several long calibration sessions. A promising time- and cost-efficient solution is artificial data generation or data augmentation. Here, we proposed a DA method for the motor imagery EEG signal called brain-area-recombination. For the BAR, each sample was first separated into two ones by left/right brain channels, and the artificial samples were generated by (...) recombining the half-samples. We then designed two schemas corresponding to the single- and multi-subject scenarios. Extensive experiments using the classifier of EEGnet were conducted on two public datasets under various training set sizes. In both schemas, the BAR method can make the EEGnet have a better performance of classification. To make a comparative investigation, we selected two common DA methods, and the BAR method beat them. Further, using the proposed BAR for augmentation, EEGnet achieved up to 8.3% improvement than a typical decoding algorithm CSP-SVM, note that both the models were trained on the augmented dataset. This study shows that BAR usage can significantly improve the classification ability of deep learning to MI-EEG signals. To a certain extent, it may promote the development of deep learning technology in the field of BCI. (shrink)

The automated identification of national implementations of European directives by text similarity techniques has shown promising preliminary results. Previous works have proposed and utilized unsupervised lexical and semantic similarity techniques based on vector space models, latent semantic analysis and topic models. However, these techniques were evaluated on a small multilingual corpus of directives and NIMs. In this paper, we utilize word and paragraph embedding models learned by shallow neural networks from a multilingual legal corpus of European directives and national legislation (...) to develop unsupervised semantic similarity systems to identify transpositions. We evaluate these models and compare their results with the previous unsupervised methods on a multilingual test corpus of 43 Directives and their corresponding NIMs. We also develop supervised machine learning models to identify transpositions and compare their performance with different feature sets. (shrink)

We argue that cognitive models can provide a common ground between human users and deep reinforcement learning (Deep RL) algorithms for purposes of explainable artificial intelligence (AI). Casting both the human and learner as cognitive models provides common mechanisms to compare and understand their underlying decision-making processes. This common grounding allows us to identify divergences and explain the learner's behavior in human understandable terms. We present novel salience techniques that highlight the most relevant features in each model's decision-making, as well (...) as examples of this technique in common training environments such as Starcraft II and an OpenAI gridworld. (shrink)

Cognitive Science, Volume 46, Issue 4, April 2022.

Brain–computer interfaces (BCIs) utilizing machine learning techniques are an emerging technology that enables a communication pathway between a user and an external system, such as a computer. Owing to its practicality, electroencephalography (EEG) is one of the most widely used measurements for BCI. However, EEG has complex patterns and EEG-based BCIs mostly involve a cost/time-consuming calibration phase; thus, acquiring sufficient EEG data is rarely possible. Recently, deep learning (DL) has had a theoretical/practical impact on BCI research because of its use (...) in learning representations of complex patterns inherent in EEG. Moreover, algorithmic advances in DL facilitate short/zero-calibration in BCI, thereby suppressing the data acquisition phase. Those advancements include data augmentation (DA), increasing the number of training samples without acquiring additional data, and transfer learning (TL), taking advantage of representative knowledge obtained from one dataset to address the so-called data insufficiency problem in other datasets. In this study, we review DL-based short/zero-calibration methods for BCI. Further, we elaborate methodological/algorithmic trends, highlight intriguing approaches in the literature, and discuss directions for further research. In particular, we search forgenerative model-based andgeometric manipulation-based DA methods. Additionally, we categorize TL techniques in DL-based BCIs intoexplicitandimplicitmethods. Our systematization reveals advances in the DA and TL methods. Among the studies reviewed herein, ~45% of DA studies used generative model-based techniques, whereas ~45% of TL studies used explicit knowledge transferring strategy. Moreover, based on our literature review, we recommend an appropriate DA strategy for DL-based BCIs and discuss trends of TLs used in DL-based BCIs. (shrink)

Almost all biomedical research to date has relied upon mean measurements from cell populations, however it is well established that what it is observed at this macroscopic level can be the result of many interactions of several different single cells. Thus, the observable macroscopic ‘average’ cannot outright be used as representative of the ‘average cell’. Rather, it is the resulting emerging behaviour of the actions and interactions of many different cells. Single‐cell RNA sequencing (scRNA‐Seq) enables the comparison of the transcriptomes (...) of individual cells. This provides high‐resolution maps of the dynamic cellular programmes allowing us to answer fundamental biological questions on their function and evolution. It also allows to address medical questions such as the role of rare cell populations contributing to disease progression and therapeutic resistance. Furthermore, it provides an understanding of context‐specific dependencies, namely the behaviour and function that a cell has in a specific context, which can be crucial to understand some complex diseases, such as diabetes, cardiovascular disease and cancer. Here, we provide an overview of scRNA‐Seq, including a comparative review of emerging technologies and computational pipelines. We discuss the current and emerging applications and focus on tumour heterogeneity a clear example of how scRNA‐Seq can provide new understanding of a complex disease. Additionally, we review the limitations and highlight the need of powerful computational pipelines and reproducible protocols for the broader acceptance of this technique in basic and clinical research. (shrink)

In rebellion against low‐dimensional (e.g., two‐factor) theories in psychology, the authors make the case for high‐dimensional theories. This change in perspective requires a shift towards a focus on computation and quantitative reasoning.

Standard models in quantitative linguistics assume that word usage follows a fixed frequency distribution, often Zipf's law or a close relative. This view, however, does not capture the near daily variations in topics of conversation, nor the short-term dynamics of language change. In order to understand the dynamics of human language use, we present a corpus of daily word frequency variation scraped from online news sources every 20 min for more than 2 years. We construct a simple time-varying model with (...) a latent state, which is observed via word frequency counts. We use Bayesian techniques to infer the parameters of this model for 20,000 words, allowing us to convert complex word-frequency trajectories into low-dimensional parameters in word usage. By analyzing the inferred parameters of this model, we quantify the relative mobility and drift of words on a day-to-day basis, while accounting for sampling error. We quantify this variation and show evidence against “rich-get-richer” models of word use, which have been previously hypothesized to explain statistical patterns in language. (shrink)

The language we use over the course of conversation changes as we establish common ground and learn what our partner finds meaningful. Here we draw upon recent advances in natural language processing to provide a finer‐grained characterization of the dynamics of this learning process. We release an open corpus (>15,000 utterances) of extended dyadic interactions in a classic repeated reference game task where pairs of participants had to coordinate on how to refer to initially difficult‐to‐describe tangram stimuli. We find that (...) different pairs discover a wide variety of idiosyncratic but efficient and stable solutions to the problem of reference. Furthermore, these conventions are shaped by the communicative context: words that are more discriminative in the initial context (i.e., that are used for one target more than others) are more likely to persist through the final repetition. Finally, we find systematic structure in how a speaker's referring expressions become more efficient over time: Syntactic units drop out in clusters following positive feedback from the listener, eventually leaving short labels containing open‐class parts of speech. These findings provide a higher resolution look at the quantitative dynamics of ad hoc convention formation and support further development of computational models of learning in communication. (shrink)

This project uses machine learning and computer vision techniques and a novel interactive visualization tool to provide street-level characterization of urban spaces such as safety and maintenance in urban neighborhoods. This is achieved by collecting and annotating street-view images, extracting objective metrics through computer vision techniques, and using crowdsourcing to statistically model the perception of subjective metrics such as safety and maintenance. For modeling human perception and scaling it up with a predictive algorithm, we evaluate perception predictions across two points (...) in time separated by economic changes in the urban core of Raleigh, North Carolina, in the aftermath of the 2008 Great Recession. We hypothesize specific socioeconomic processes can be substantially reflected in the built environment of cities and, thus, render themselves visible at the street level. This paper describes the process of incorporating subjective visual ratings across two datasets of temporally separated street-view images, an algorithm, and a visualization tool. This work serves as a case study for utilizing AI and visualization techniques in a richer characterization of urban spaces that includes both objective metrics such as income (that operates at a broader scale) and subjective metrics such as perception of individuals (that operates at a narrower scale at specific locations). We outline an interdisciplinary methodology to test this hypothesis in streetscape data from Raleigh, NC, from 2008 to 2020. We describe the results of training algorithms that utilized image features with crowdsourced human perception ratings. We provide a comparison of the results with income data. The analysis and interpretation of this comparison provide insight into the challenges and opportunities for using AI technology in characterizing changes in urban environments. One challenge is the ability of human domain experts to interpret the output of algorithms through manipulation and to integrate these results into their workflow. This is addressed with a novel interface designed for interactive analysis and visualization. We conclude with a discussion of some of the benefits and limitations of integrating AI models in the human expert’s decision-making process in the presence of both subjective and objective metrics. (shrink)