Token Group Research Articles

Fusion Attention for Action Recognition: Integrating Sparse-Dense and Global Attention for Video Action Recognition

Conventional approaches to video action recognition perform global attention over the entire video patches, which may be ineffective due to the temporal redundancy of video frames. Recent works on masked video modeling adopt a high-ratio tube masking and reconstruction strategy as a pre-training method to mitigate the problem of focusing on spatial features well but not on temporal features. Inspired by this pre-training method, we propose Fusion Attention for Action Recognition (FAR), which fuses the sparse-dense attention patterns specialized for temporal features with global attention during fine-tuning. FAR has three main components: head-split sparse-dense attention (HSDA), token–group interaction, and group-averaged classifier. First, HSDA splits the head of multi-head self-attention to fuse global and sparse-dense attention. The sparse-dense attention is divided into groups of tube-shaped patches to focus on temporal features. Second, token–group interaction is used to improve information exchange between divided patch groups. Finally, the group-averaged classifier uses spatio-temporal features from different patch groups to improve performance. The proposed method uses the weight parameters that are pre-trained with VideoMAE and MVD, and achieves higher performance (+0.1–0.4%) with less computation than models fine-tuned with global attention on Something-Something V2 and Kinetics-400. Moreover, qualitative comparisons show that FAR captures temporal features quite well in highly redundant video frames. The FAR approach demonstrates improved action recognition with efficient computation, and exploring its adaptability across different pre-training methods presents an interesting direction for future research.

Selective Information Flow for Transformer Tracking

Fully transformer-based one-stream trackers have demonstrated outstanding performance on challenging benchmark datasets over the past three years. These trackers enable bidirectional information flow between all target template and search region tokens to compute attention features without assessing its impact on the tracker’s discriminative capability. Our study found that the information flow from a large portion of background tokens in the search region diminishes the importance of the target-specific features of the template. Moreover, previous transformer-based trackers fail to consider cues from the dynamic background region, even though it contains information about distracting similar objects. To address the identified limitations in transformer tracking, we propose a novel Selective Information Flow Tracking (SIFTrack) framework to enhance the tracker’s discriminative capability by selectively allowing information flow between different types of token groups. In the early encoder layers of the proposed SIFTrack, interactions from all search tokens to target template tokens are blocked to enrich target-specific feature extraction. In the deeper encoder layers, search tokens are partitioned into target and non-target tokens based on their attention scores. Then, bidirectional flow from target search tokens to template tokens is allowed to capture the appearance changes of the target. In addition, by including tokens from the dynamic background, SIFTrack effectively avoids distractor objects by capturing cues from the area surrounding the target. The proposed SIFTrack demonstrated outstanding performance in challenging benchmarks, particularly excelling in the one-shot tracking benchmark GOT-10k, achieving an average overlap of 74.6%. The code, models, and results of this work are available at https://github.com/JananiKugaa/SIFTrack.

Semantic-aware Message Broadcasting for Efficient Unsupervised Domain Adaptation.

Vision transformer has demonstrated great potential in abundant vision tasks. However, it also inevitably suffers from poor generalization capability when the distribution shift occurs in testing (i.e., out-of-distribution data). To mitigate this issue, we propose a novel method, Semantic-aware Message Broadcasting (SAMB), which enables more informative and flexible feature alignment for unsupervised domain adaptation (UDA). Particularly, we study the attention module in the vision transformer and notice that the alignment space using one global class token lacks enough flexibility, where it interacts information with all image tokens in the same manner but ignores the rich semantics of different regions. In this paper, we aim to improve the richness of the alignment features by enabling semantic-aware adaptive message broadcasting. Particularly, we introduce a group of learned group tokens as nodes to aggregate the global information from all image tokens, but encourage different group tokens to adaptively focus on the message broadcasting to different semantic regions. In this way, our message broadcasting encourages the group tokens to learn more informative and diverse information for effective domain alignment. Moreover, we systematically study the effects of adversarial-based feature alignment (ADA) and pseudo-label based self-training (PST) on UDA. We find that one simple two-stage training strategy with the cooperation of ADA and PST can further improve the adaptation capability of the vision transformer. Extensive experiments on DomainNet, OfficeHome, and VisDA-2017 demonstrate the effectiveness of our methods for UDA.

Proposal of a Token-Based Node Selection Mechanism for Node Distribution of Mobility IoT Blockchain Nodes.

Various elements, such as evolutions in IoT services resulting from sensoring by vehicle parts and advances in small communication technology devices, have significantly impacted the mass spread of mobility services that are provided to users in need of limited resources. In particular, business models are progressing away from one-off costs towards longer-term costs, as represented by shared services utilizing kick-boards or bicycles and subscription services for vehicle software. Advances in shared mobility services, as described, are calling for solutions that can enhance the reliability of data aggregated by users leveraging mobility services in the next-generation mobility areas. However, the mining process to renew status ensures continued network communication, and block creation demands high performance in the public block chain. Therefore, easing the mining process for state updates in public blockchains is a way to alleviate the high-performance process requirements of public blockchains. The proposed mechanism assigns token-based block creation authority instead of the mining method, which provides block creation authority to nodes that provide many resources. Blocks are created only by a group of participants with tokens, and after creation, tokens are updated and delivered to new nodes to form a new token group. Additionally, tokens are updated in each block after their initial creation, making it difficult to disguise the tokens and preventing resource-centered centralization.

TSNet: Token Sparsification for Efficient Video Transformer

In the domain of video recognition, video transformers have demonstrated remarkable performance, albeit at significant computational cost. This paper introduces TSNet, an innovative approach for dynamically selecting informative tokens from given video samples. The proposed method involves a lightweight prediction module that assigns importance scores to each token in the video. Tokens with top scores are then utilized for self-attention computation. We apply the Gumbel-softmax technique to sample from the output of the prediction module, enabling end-to-end optimization of the prediction module. We aim to extend our method on hierarchical vision transformers rather than single-scale vision transformers. We use a simple linear module to project the pruned tokens, and the projected result is then concatenated with the output of the self-attention network to maintain the same number of tokens while capturing interactions with the selected tokens. Since feedforward networks (FFNs) contribute significant computation, we also propose linear projection for the pruned tokens to accelerate the model, and the existing FFN layer progresses the selected tokens. Finally, in order to ensure that the structure of the output remains unchanged, the two groups of tokens are reassembled based on their spatial positions in the original feature map. The experiments conducted primarily focus on the Kinetics-400 dataset using UniFormer, a hierarchical video transformer backbone that incorporates convolution in its self-attention block. Our model demonstrates comparable results to the original model while reducing computation by over 13%. Notably, by hierarchically pruning 70% of input tokens, our approach significantly decreases 55.5% of the FLOPs, while the decline in accuracy is confined to 2%. Additional testing of wide applicability and adaptability with other transformers such as the Video Swin Transformer was also performed and indicated its progressive potentials in video recognition benchmarks. By implementing our token sparsification framework, video vision transformers can achieve a remarkable balance between enhanced computational speed and a slight reduction in accuracy.

An Integrated Transformer with Collaborative Tokens Mining for Fine-Grained Recognition

Fine-grained recognition mainly classifies subclass images into hundreds of subcategorical labels by locating the discriminative regions (e.g., Cape May warbler or Magnolia warbler bird). Due to the high complexity and non-differentiation of region locations through the traditional backbone architecture, most existing approaches utilize multi-level reinforcement learning to distinguish the similar appearance among sub-categories. These methods explore incomplete information through only the intra-class informative regions in one image or the inter-class and intra-class relationship in pairwise images, leading to the tendency for overlapped region locations. Since the inter-class correlations and new backbone with complete contextual semantic information play important roles in distinguishing fine-grained classes, we propose a novel transformer with the collaborative token mining (TCTM) scheme by fully exploiting the relationships between inter-class and intra-class regions. The proposed TCTM scheme with a new transformer backbone consists of two modules that collaboratively explore the spatially aware tokens: the Pyramid Tokens Multiplication (PTM) module which exploits the integrated multi-stage inter-class and intra-class correlations from new transformer architecture and the Tokens Proposals Generation (TPG) module which captures two groups of top-four discriminative tokens. The two PTMs extract contrastive tokens for each image and learn to rank these tokens, assuming that those from the same class and the same module should have smaller distances. The TPGs further sort and update the candidate tokens from the extracted attention tokens by ranking their probabilities with ground truth subcategorical labels. Through the collaboration between the PTM and TPG, our TCTM scheme can take the integrated correlations into account and mine the discriminative tokens for final fine-grained classification. Extensive experiments on four popular benchmarks show that our proposed TCTM outperforms the state-of-the-art methods by a large margin.

The Multi-Hot Representation-Based Language Model to Maintain Morpheme Units

Natural language models brought rapid developments to Natural Language Processing (NLP) performance following the emergence of large-scale deep learning models. Language models have previously used token units to represent natural language while reducing the proportion of unknown tokens. However, tokenization in language models raises language-specific issues. One of the key issues is that separating words by morphemes may cause distortion to the original meaning; also, it can prove challenging to apply the information surrounding a word, such as its semantic network. We propose a multi-hot representation language model to maintain Korean morpheme units. This method represents a single morpheme as a group of syllable-based tokens for cases where no matching tokens exist. This model has demonstrated similar performance to existing models in various natural language processing applications. The proposed model retains the minimum unit of meaning by maintaining the morpheme units and can easily accommodate the extension of semantic information.

NFTs and copyright: challenges and opportunities

The attention paid to blockchain technology has always been shadowed by the ever-popular Bitcoin. Bitcoin is not the only life-altering outcome of the blockchain revolution. Each day we come across another useful aspect of blockchain technology: first, it was smart contracts, then initial coin offerings, then security token offerings and, since the beginning of March this year, there has been a lot of hype around non-fungible tokens (NFTs). While different digital art and collectibles are tokenized and sold for mind-boggling amounts, there are some legal questions, including regarding copyright law, that need to be tackled. This article will develop as follows: In Part II, we first describe the main characteristics of blockchain technology, and its important features that impact NFTs. Then, we consider where NFTs should be classified within the different groups of tokens. We define NFTs and consider why they are so popular nowadays and how they interact with...

LET-Decoder: A WFST-Based Lazy-Evaluation Token-Group Decoder With Exact Lattice Generation

We propose a novel lazy-evaluation token-group decoding algorithm with on-the-fly composition of weighted finite-state transducers (WFSTs) for large vocabulary continuous speech recognition. In the standard on-the-fly composition decoder, a base WFST and one or more incremental WFSTs are composed during decoding, and then token passing algorithm is employed to generate the lattice on the composed search space, resulting in substantial computation overhead. To improve speed, the proposed algorithm adopts 1) a token-group method, which groups tokens with the same state in the base WFST on each frame and limits the capacity of the group and 2) a lazy-evaluation method, which does not expand a token group and its source token groups until it processes a word label during decoding. Experiments show that the proposed decoder works notably up to 3 times faster than the standard on-the-fly composition decoder.

Концепт «етнічність» в сучасних британських словниках (на матеріалі словника OALD).

The study of the essence and features of the representation of different groups of tokens related to the concept of ethnicity, will demonstrate how language contributes to social and psychological processes in society and its development, and how it affects the formation of ethnic identity. Examining the linguistic features of varieties of ethno-labeled tokens and the ways of their use in the languages of different ethnic groups, we can find out how paradigms change in the social approach to understanding interethnic communication and tolerant coexistence of different nations and peoples.

Intellectual Innovations in Georgia (11th-9th Centuries BC)

Abstract The multilayer archaeological site Grakliani hill is located in one of the main contact zones of Transcaucasia. It shows a very intensive level of communication with other cultures from the 2nd millenium BC till the 3rd century BC and a high level of development of Kartli (Iberia Caucasica) society. Two inscriptions made in unknown script (probably a local version of Aramaic script) were discovered in a shrine of the 11th-9th centuries BC. A group of weights and tokens, also of the 11th-9th centuries BC was discovered in the excavations of Grakliani Hill and the satellite site of Tsina Gora. The Achaemenid period remains are also of special interest, including Greek style architecture details.

The Influence of Incidental Tokenism on Private Evaluations of Stereotype-Typifying Products

Findings from five studies demonstrate that being an incidental token member in a transient group (e.g., a woman in a group of mostly men in a store line) lowers individuals’ private evaluations of products that typify the negative stereotypes of the tokenized identity. Incidental tokenism activates negative stereotypes associated with the tokenized identity, which subsequently leads to a desire to disassociate specifically from identity-linked products that typify those stereotypes as opposed to all identity-linked products in general. Consistent with this theorizing, similar results emerge when negative stereotypes are activated directly, and the effect is attenuated when tokenized individuals are self-affirmed. These results demonstrate the largely unexamined consequences of being a token group member on private evaluations (vs. public behavior) in subjective, preference-based (vs. objective, performance-based) domains.

COLOR NAMES "WHITE" AND "BLACK" IN DESCRIPTIONS OF PERSON IN THE KAZAKH LANGUAGE VS. ENGLISH AND RUSSIAN

This article is dedicated to the problem of color perception in the language worldview in describing a person according to the «black and white» color palette in Kazakh, English and Russian; it also examines phraseological units that feature these colors. Color is one of the principles of culture; many cultural phenomena cannot be understood without taking into account the importance of color. In particular, in descriptions of person one can distinguish a number of general associative color fields with similarities and differences. A study of phraseological dictionaries of the Kazakh, English and Russian languages, as well as works of fiction, has revealed that the color designation "ақ-белый-white" and "қара-черный-black" differ in their versatility in describing one’s appearance, character and social status. Subject: color perception in the language worldview of Russian, Kazakh and English speakers. The purpose of the work: to identify common and differential manifestations of white and black in different linguocultures using lexemes "ақ – белый – white" and "қара – черный – black" in phraseological units. Method of investigation: semantic and communicative analysis. In Kazakh, Russian, and English, white and black colors in description of person have a wide scope of application. Their frequency means that these color codes are part of a group of tokens with a high potential for phraseological activity. These colors are actively used both in everyday life and fiction when a person is described with the help of color contrasts. Material of the research: the paper features phraseological units with the elements of color identification "white", "black" and "red" from verbal folklore and Kazakh literature classics, as well as works by Russian and English writers.

Developmental network structure and support: gendered consequences for work–family strain and work–parenting strain in the Australian mining industry

Developmental networks enhance career success through the support received by the protégé via the network structure. This paper extends developmental network research by exploring the extent to which strain is associated with developmental network structure and support in the Australian mining industry, a highly volatile and unique context. Our research tests the popular notion of ‘the more support you get, the better’ which is in need of further exploration in the developmental networks literature particularly in specific work contexts with strain (vs. success variables) as outcomes. Results indicate that bigger, broader networks with more career, psychosocial and role modelling support are not always beneficial for protégés in this context. A smaller network with a broader range of developers is associated with reduced work–parenting strain, but not work–family strain. Increased career support reduced work–family strain, but this was not the case for increased psychosocial support and role modelling support. Further, gender moderated the relationship between psychosocial support and work–family strain possibly due to token group effects.

Perceptually relevant grouping of image tokens on the basis of constraint propagation from local binary patterns

Grouping image tokens is an intermediate step needed to arrive at meaningful image representation and summarization. Usually, perceptual cues, for instance, gestalt properties inform token grouping. However, they do not take into account structural continuities that could be derived from other tokens belonging to similar structures irrespective of their location. We propose an image representation that encodes structural constraints emerging from local binary patterns (LBP), which provides a long-distance measure of similarity but in a structurally connected way. Our representation provides a grouping of pixels or larger image tokens that is free of numeric similarity measures and could therefore be extended to nonmetric spaces. The representation lends itself nicely to ubiquitous image processing applications such as connected component labeling and segmentation. We test our proposed representation on the perceptual grouping or segmentation task on the popular Berkeley segmentation dataset (BSD500) that with respect to human segmented images achieves an average F-measure of 0.559. Our algorithm achieves a high average recall of 0.787 and is therefore well-suited to other applications such as object retrieval and category-independent object recognition. The proposed merging heuristic based on levels of singular tree component has shown promising results on the BSD500 dataset and currently ranks 12th among all benchmarked algorithms, but contrary to the others, it requires no data-driven training or specialized preprocessing.

Extreme Violence and the Media: Challenges of Reporting Terrorism in Nigeria

Reporting terrorism or extreme violence presents a myriad of challenges and dilemmas to media professionals, information managers, and other state actors who are saddled with the responsibilities of objectively, responsibly, and accurately purveying information to ensure effective development communication in society. By the same token, insurgent or terrorist groups spread their inordinate causes, transmit their messages of radicalisation, and garner support, recognition, and legitimacy from the populace through media channels such as handbills, internet, radio, and film. The paradox, therefore, is that in their informational offerings, the media or media professionals have inadvertently become accomplices or victims/endangered species in terror acts. In spite of this, they are duty-bound to report the events regardless of the consequences on the audience(s). Indeed, it has been argued that the sensational reportage and overly dramatization of the activities of extremists groups in the media further propagates terrorist acts. This article examines media reportage of terrorism occasioned by the activities of Boko Haram terrorist sect(s) in Nigeria and the challenges media professionals are confronted with in the line of duty. Mooring itself on the agenda setting and gate-keeping theories of the media, it uses historical-analytic method to interrogate the complex relationship between the media and Boko Haram terrorists as well as the dangers posed to Nigerian media professionals and the collective security of the Nigerian state and even the neighbouring countries.

CapSG를 이용한 IoT 서비스 접근제어 플랫폼

사물인터넷(Internet of Things, IoT) 환경의 다양한 프로토콜, 도메인, 애플리케이션 위에서 유연한 서비스 제공을 위한 효율적인 사용자 권한 관리 방법이 필요하다. 본 논문은 IoT 환경의 다양한 서비스에 대한 효율적인 접근제어를 제공하기 위하여 CapSG(Capability Service Gateway)를 이용한 IoT 서비스 플랫폼을 제안하였다. CapSG는 인증과 접근 권한을 포함하는 토큰을 사용하여, 서비스 주체에 대한 인증 및 접근제어를 수행하여 서비스를 제공한다. 또한, 서비스 토큰 관리를 위한 생성, 위임, 폐기, 거절 등의 기능을 제공한다. 본 논문은 각 도메인에 대한 토큰 그룹을 사용함으로써 도메인별 서비스 접근제어가 가능하며, 토큰 그룹 내의 특정 서비스 토큰을 이용한 접근제어 수행도 가능하도록 설계하여 IoT의 다양한 서비스 요구에 대한 접근제어의 유연성과 효율성을 제공한다. There is great need for efficient user rights management method to provide a flexible service on variety protocols, domains, applications of IoT environments. In this paper, we propose a IoT service platform with CapSG to provide efficient access control for IoT various services of the environment. CapSG uses a token including authentication and access rights to perform authentication and access control service entity providing services. In addition, the generated token for service management, delegation, revocation, and provides a function such as denied. Also, it provides functions such as generation, delegation, disposal and rejection for service token management. In this paper, it provides the flexibility and efficiency of the access control for various services require of the IoT because of it is available to access control specific domain service by using the token group for each domain and is designed to access control using specific service token of tokens group.

Primary and secondary effects of processing instruction on Spanish clitic pronouns

AbstractThis study investigates the effects of token item frequency in Structured Input activities on both a primary target form (Spanish accusative clitics) and a secondary target form (Spanish dative clitics). Participants included 460 adult learners enrolled in a beginning-level Spanish language course and they were exposed to either 40, 60, 80, 100, 120, or 140 target form tokens. This study included a pretest, immediate posttest, and a delayed posttest measuring interpretation and production of both primary and secondary target forms. Findings reveal that primary form interpretation effects across all frequencies, however, production findings present themselves with the 60 and 80 token groups only. Secondary form interpretation findings reveal themselves across all frequency levels with the exception of the lowest frequency investigated (40 tokens) and secondary form production mirror those found in previous studies on the same forms. As such, we discuss the theoretical and methodological ramifications of these findings as well as directions for future research.

A Fuzzy Logic Approach to Wrapping PDF Documents

The PDF format represents the de facto standard for print-oriented documents. In this paper, we address the problem of wrapping PDF documents, which raises new challenges in several contexts of text data management. Our proposal is based on a novel bottom-up hierarchical wrapping approach that exploits fuzzy logic to handle the “uncertainty” which is intrinsic to the structure and presentation of PDF documents. A PDF wrapper is defined by specifying a set of group type definitions that impose a target structure to groups of tokens containing the required information. Constraints on token groupings are formulated as fuzzy conditions, which are defined on spatial and content predicates of tokens. We define a formal semantics for PDF wrappers and propose an algorithm for wrapper evaluation working in polynomial time with respect to the size of a PDF document. The proposed approach has been implemented in a wrapper generation system that offers visual capabilities to assist the designer in specifying and evaluating a PDF wrapper. Experimental results have shown good accuracy and applicability of our system to PDF documents of various domains.

First Passage Time Computation in Tagged GSPNs with Queue Places

This paper presents an extension of the generalized stochastic Petri net (GSPN) formalism that enables the computation of first passage time distributions. The tagged customer technique typical of queuing networks is adapted to the GSPN context by providing a formal definition and an automatic computation of the groups of tokens that can be identified as customers, i.e. classes of homogeneous entities behaving in a similar manner. Passage times are identified through the concept of events that correspond to the firing of transitions placed at the boundaries of a subnet. The extended model obtained with this specifications is translated into an ordinary GSPN by isolating a customer from the group and highlighting its path through the net thus obtaining a representation suited for the passage time analysis. Proofs are provided to show the equivalence between these models with respect to their steady-state distributions. An important and original aspect treated in this paper is the possibility of specifying several scheduling policies of tokens at places, an information not present in ordinary GSPN models, but that is vital for the precise computation of first passage time distributions as shown by a few results computed for a simple Flexible Manufacturing application.