SEARCH · Search arXiv

Results for “transformer architecture”

Search indexed arXiv papers on artificial intelligence, large language models, computer vision and robotics. Read source abstracts and follow links to arXiv.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 1,243 records · Page 69Linked to original sources

arXiv2021

Improved RAMEN: Towards Domain Generalization for Visual Question Answering

Currently nearing human-level performance, Visual Question Answering (VQA) is an emerging area in artificial intelligence. Established as a multi-disciplinary field in machine learning, both computer vision and natural language processing communities are working together to achieve state-of-the-art (SOTA) performance. However, there is a gap between the SOTA results and real world applications. This is due to the lack of model generalisation. The RAMEN model \cite{Shrestha2019} aimed to achieve domain generalization by obtaining the highest score across two main types of VQA datasets. This study provides two major improvements to the early/late fusion module and aggregation module of the RAMEN architecture, with the objective of further strengthening domain generalization. Vector operations based fusion strategies are introduced for the fusion module and the transformer architecture is introduced for the aggregation module. Improvements of up to five VQA datasets from the experiments conducted are evident. Following the results, this study analyses the effects of both the improvements on the domain generalization problem. The code is available on GitHub though the following link \url{https://github.com/bhanukaManesha/ramen}.

cs.CV↗

arXiv2022

Transformers in Medical Image Analysis: A Review

Transformers have dominated the field of natural language processing, and recently impacted the computer vision area. In the field of medical image analysis, Transformers have also been successfully applied to full-stack clinical applications, including image synthesis/reconstruction, registration, segmentation, detection, and diagnosis. Our paper aims to promote awareness and application of Transformers in the field of medical image analysis. Specifically, we first overview the core concepts of the attention mechanism built into Transformers and other basic components. Second, we review various Transformer architectures tailored for medical image applications and discuss their limitations. Within this review, we investigate key challenges revolving around the use of Transformers in different learning paradigms, improving the model efficiency, and their coupling with other techniques. We hope this review can give a comprehensive picture of Transformers to the readers in the field of medical image analysis.

cs.CV↗

arXiv2023

Cross Aggregation Transformer for Image Restoration

Recently, Transformer architecture has been introduced into image restoration to replace convolution neural network (CNN) with surprising results. Considering the high computational complexity of Transformer with global attention, some methods use the local square window to limit the scope of self-attention. However, these methods lack direct interaction among different windows, which limits the establishment of long-range dependencies. To address the above issue, we propose a new image restoration model, Cross Aggregation Transformer (CAT). The core of our CAT is the Rectangle-Window Self-Attention (Rwin-SA), which utilizes horizontal and vertical rectangle window attention in different heads parallelly to expand the attention area and aggregate the features cross different windows. We also introduce the Axial-Shift operation for different window interactions. Furthermore, we propose the Locality Complementary Module to complement the self-attention mechanism, which incorporates the inductive bias of CNN (e.g., translation invariance and locality) into Transformer, enabling global-local coupling. Extensive experiments demonstrate that our CAT outperforms recent state-of-the-art methods on several image restoration applications. The code and models are available at https://github.com/zhengchen1999/CAT.

cs.CV↗

arXiv2023

ViT Unified: Joint Fingerprint Recognition and Presentation Attack Detection

A secure fingerprint recognition system must contain both a presentation attack (i.e., spoof) detection and recognition module in order to protect users against unwanted access by malicious users. Traditionally, these tasks would be carried out by two independent systems; however, recent studies have demonstrated the potential to have one unified system architecture in order to reduce the computational burdens on the system, while maintaining high accuracy. In this work, we leverage a vision transformer architecture for joint spoof detection and matching and report competitive results with state-of-the-art (SOTA) models for both a sequential system (two ViT models operating independently) and a unified architecture (a single ViT model for both tasks). ViT models are particularly well suited for this task as the ViT's global embedding encodes features useful for recognition, whereas the individual, local embeddings are useful for spoof detection. We demonstrate the capability of our unified model to achieve an average integrated matching (IM) accuracy of 98.87% across LivDet 2013 and 2015 CrossMatch sensors. This is comparable to IM accuracy of 98.95% of our sequential dual-ViT system, but with ~50% of the parameters and ~58% of the latency.

cs.CV↗

arXiv2023

Lightening-Transformer: A Dynamically-operated Optically-interconnected Photonic Transformer Accelerator

The wide adoption and significant computing resource of attention-based transformers, e.g., Vision Transformers and large language models (LLM), have driven the demand for efficient hardware accelerators. There is a growing interest in exploring photonics as an alternative technology to digital electronics due to its high energy efficiency and ultra-fast processing speed. Photonic accelerators have shown promising results for CNNs, which mainly rely on weight-static linear operations. However, they encounter issues when efficiently supporting Transformer architectures, questioning the applicability of photonics to advanced ML tasks. The primary hurdle lies in their inefficiency in handling unique workloads in Transformers, i.e., dynamic and full-range tensor multiplication. In this work, we propose Lightening-Transformer, the first light-empowered, high-performance, and energy-efficient photonic Transformer accelerator. To overcome prior designs' fundamental limitations, we introduce a novel dynamically-operated photonic tensor core, DPTC, a crossbar array of interference-based optical vector dot-product engines supporting highly parallel, dynamic, and full-range matrix multiplication. Furthermore, we design a dedicated accelerator that integrates our novel photonic computing cores with photonic interconnects for inter-core data broadcast, fully unleashing the power of optics. Comprehensive evaluations show that ours achieves >2.6x energy and >12x latency reductions compared to prior photonic accelerators and delivers the lowest energy cost and 2 to 3 orders of magnitude lower energy-delay product compared to electronic Transformer accelerators, all while maintaining digital-comparable accuracy. Our work highlights the immense potential of photonics for advanced ML workloads, such as Transformer-backboned LLM. Our work is available at https://github.com/zhuhanqing/Lightening-Transformer.

cs.ET↗

arXiv2023

Collect-and-Distribute Transformer for 3D Point Cloud Analysis

Remarkable advancements have been made recently in point cloud analysis through the exploration of transformer architecture, but it remains challenging to effectively learn local and global structures within point clouds. In this paper, we propose a new transformer network equipped with a collect-and-distribute mechanism to communicate short- and long-range contexts of point clouds, which we refer to as CDFormer. Specifically, we first employ self-attention to capture short-range interactions within each local patch, and the updated local features are then collected into a set of proxy reference points from which we can extract long-range contexts. Afterward, we distribute the learned long-range contexts back to local points via cross-attention. To address the position clues for short- and long-range contexts, we additionally introduce the context-aware position encoding to facilitate position-aware communications between points. We perform experiments on five popular point cloud datasets, namely ModelNet40, ScanObjectNN, ShapeNetPart, S3DIS and ScanNetV2, for classification and segmentation. Results show the effectiveness of the proposed CDFormer, delivering several new state-of-the-art performances on point cloud classification and segmentation tasks. The source code is available at \url{https://github.com/haibo-qiu/CDFormer}.

cs.CV↗

arXiv2023

Audio-aware Query-enhanced Transformer for Audio-Visual Segmentation

The goal of the audio-visual segmentation (AVS) task is to segment the sounding objects in the video frames using audio cues. However, current fusion-based methods have the performance limitations due to the small receptive field of convolution and inadequate fusion of audio-visual features. To overcome these issues, we propose a novel \textbf{Au}dio-aware query-enhanced \textbf{TR}ansformer (AuTR) to tackle the task. Unlike existing methods, our approach introduces a multimodal transformer architecture that enables deep fusion and aggregation of audio-visual features. Furthermore, we devise an audio-aware query-enhanced transformer decoder that explicitly helps the model focus on the segmentation of the pinpointed sounding objects based on audio signals, while disregarding silent yet salient objects. Experimental results show that our method outperforms previous methods and demonstrates better generalization ability in multi-sound and open-set scenarios.

cs.SD↗

arXiv2023

Event-based Vision for Early Prediction of Manipulation Actions

Neuromorphic visual sensors are artificial retinas that output sequences of asynchronous events when brightness changes occur in the scene. These sensors offer many advantages including very high temporal resolution, no motion blur and smart data compression ideal for real-time processing. In this study, we introduce an event-based dataset on fine-grained manipulation actions and perform an experimental study on the use of transformers for action prediction with events. There is enormous interest in the fields of cognitive robotics and human-robot interaction on understanding and predicting human actions as early as possible. Early prediction allows anticipating complex stages for planning, enabling effective and real-time interaction. Our Transformer network uses events to predict manipulation actions as they occur, using online inference. The model succeeds at predicting actions early on, building up confidence over time and achieving state-of-the-art classification. Moreover, the attention-based transformer architecture allows us to study the role of the spatio-temporal patterns selected by the model. Our experiments show that the Transformer network captures action dynamic features outperforming video-based approaches and succeeding with scenarios where the differences between actions lie in very subtle cues. Finally, we release the new event dataset, which is the first in the literature for manipulation action recognition. Code will be available at https://github.com/DaniDeniz/EventVisionTransformer.

cs.CV↗

arXiv2024

The I/O Complexity of Attention, or How Optimal is Flash Attention?

Self-attention is at the heart of the popular Transformer architecture, yet suffers from quadratic time and memory complexity. The breakthrough FlashAttention algorithm revealed I/O complexity as the true bottleneck in scaling Transformers. Given two levels of memory hierarchy, a fast cache (e.g. GPU on-chip SRAM) and a slow memory (e.g. GPU high-bandwidth memory), the I/O complexity measures the number of accesses to memory. FlashAttention computes attention using $\frac{N^2d^2}{M}$ I/O operations where $N$ is the dimension of the attention matrix, $d$ the head-dimension and $M$ the cache size. However, is this I/O complexity optimal? The known lower bound only rules out an I/O complexity of $o(Nd)$ when $M=Θ(Nd)$, since the output that needs to be written to slow memory is $Ω(Nd)$. This leads to the main question of our work: Is FlashAttention I/O optimal for all values of $M$? We resolve the above question in its full generality by showing an I/O complexity lower bound that matches the upper bound provided by FlashAttention for any values of $M \geq d^2$ within any constant factors. Further, we give a better algorithm with lower I/O complexity for $M < d^2$, and show that it is optimal as well. Moreover, our lower bounds do not rely on using combinatorial matrix multiplication for computing the attention matrix. We show even if one uses fast matrix multiplication, the above I/O complexity bounds cannot be improved. We do so by introducing a new communication complexity protocol for matrix compression, and connecting communication complexity to I/O complexity. To the best of our knowledge, this is the first work to establish a connection between communication complexity and I/O complexity, and we believe this connection could be of independent interest and will find many more applications in proving I/O complexity lower bounds in the future.

cs.LG↗

arXiv2024

A Review of Intelligent Device Fault Diagnosis Technologies Based on Machine Vision

This paper provides a comprehensive review of mechanical equipment fault diagnosis methods, focusing on the advancements brought by Transformer-based models. It details the structure, working principles, and benefits of Transformers, particularly their self-attention mechanism and parallel computation capabilities, which have propelled their widespread application in natural language processing and computer vision. The discussion highlights key Transformer model variants, such as Vision Transformers (ViT) and their extensions, which leverage self-attention to improve accuracy and efficiency in visual tasks. Furthermore, the paper examines the application of Transformer-based approaches in intelligent fault diagnosis for mechanical systems, showcasing their superior ability to extract and recognize patterns from complex sensor data for precise fault identification. Despite these advancements, challenges remain, including the reliance on extensive labeled datasets, significant computational demands, and difficulties in deploying models on resource-limited devices. To address these limitations, the paper proposes future research directions, such as developing lightweight Transformer architectures, integrating multimodal data sources, and enhancing adaptability to diverse operational conditions. These efforts aim to further expand the application of Transformer-based methods in mechanical fault diagnosis, making them more robust, efficient, and suitable for real-world industrial environments.

cs.CV↗

arXiv2025

Hashed Watermark as a Filter: Defeating Forging and Overwriting Attacks in Weight-based Neural Network Watermarking

As valuable digital assets, deep neural networks necessitate robust ownership protection, positioning neural network watermarking (NNW) as a promising solution. Among various NNW approaches, weight-based methods are favored for their simplicity and practicality; however, they remain vulnerable to forging and overwriting attacks. To address those challenges, we propose NeuralMark, a robust method built around a hashed watermark filter. Specifically, we utilize a hash function to generate an irreversible binary watermark from a secret key, which is then used as a filter to select the model parameters for embedding. This design cleverly intertwines the embedding parameters with the hashed watermark, providing a robust defense against both forging and overwriting attacks. An average pooling is also incorporated to resist fine-tuning and pruning attacks. Furthermore, it can be seamlessly integrated into various neural network architectures, ensuring broad applicability. Theoretically, we analyze its security boundary. Empirically, we verify its effectiveness and robustness across 13 distinct Convolutional and Transformer architectures, covering five image classification tasks and one text generation task. The source codes are available at https://github.com/AIResearch-Group/NeuralMark.

cs.CR↗

arXiv2025

QU-NLP at CheckThat! 2025: Multilingual Subjectivity in News Articles Detection using Feature-Augmented Transformer Models with Sequential Cross-Lingual Fine-Tuning

This paper presents our approach to the CheckThat! 2025 Task 1 on subjectivity detection, where systems are challenged to distinguish whether a sentence from a news article expresses the subjective view of the author or presents an objective view on the covered topic. We propose a feature-augmented transformer architecture that combines contextual embeddings from pre-trained language models with statistical and linguistic features. Our system leveraged pre-trained transformers with additional lexical features: for Arabic we used AraELECTRA augmented with part-of-speech (POS) tags and TF-IDF features, while for the other languages we fine-tuned a cross-lingual DeBERTa~V3 model combined with TF-IDF features through a gating mechanism. We evaluated our system in monolingual, multilingual, and zero-shot settings across multiple languages including English, Arabic, German, Italian, and several unseen languages. The results demonstrate the effectiveness of our approach, achieving competitive performance across different languages with notable success in the monolingual setting for English (rank 1st with macro-F1=0.8052), German (rank 3rd with macro-F1=0.8013), Arabic (rank 4th with macro-F1=0.5771), and Romanian (rank 1st with macro-F1=0.8126) in the zero-shot setting. We also conducted an ablation analysis that demonstrated the importance of combining TF-IDF features with the gating mechanism and the cross-lingual transfer for subjectivity detection. Furthermore, our analysis reveals the model's sensitivity to both the order of cross-lingual fine-tuning and the linguistic proximity of the training languages.

cs.CL↗

arXiv2025

Anatomically Constrained Transformers for Echocardiogram Analysis

Video transformers have recently demonstrated strong potential for echocardiogram (echo) analysis, leveraging self-supervised pre-training and flexible adaptation across diverse tasks. However, like other models operating on videos, they are prone to learning spurious correlations from non-diagnostic regions such as image backgrounds. To overcome this limitation, we propose the Video Anatomically Constrained Transformer (ViACT), a novel framework that integrates anatomical priors directly into the transformer architecture. ViACT represents a deforming anatomical structure as a point set and encodes both its spatial geometry and corresponding image patches into transformer tokens. During pre-training, ViACT follows a masked autoencoding strategy that masks and reconstructs only anatomical patches, enforcing that representation learning is focused on the anatomical region. The pre-trained model can then be fine-tuned for tasks localized to this region. In this work we focus on the myocardium, demonstrating the framework on echo analysis tasks such as left ventricular ejection fraction (EF) regression and cardiac amyloidosis (CA) detection. The anatomical constraint focuses transformer attention within the myocardium, yielding interpretable attention maps aligned with regions of known CA pathology. Moreover, ViACT generalizes to myocardium point tracking without requiring task-specific components such as correlation volumes used in specialized tracking networks.

cs.CV↗

arXiv2025

Beyond CLIP: Knowledge-Enhanced Multimodal Transformers for Cross-Modal Alignment in Diabetic Retinopathy Diagnosis

Diabetic retinopathy (DR) is a leading cause of preventable blindness worldwide, demanding accurate automated diagnostic systems. While general-domain vision-language models like Contrastive Language-Image Pre-Training (CLIP) perform well on natural image tasks, they struggle in medical domain applications, particularly in cross-modal retrieval for ophthalmological images. We propose a novel knowledge-enhanced joint embedding framework that integrates retinal fundus images, clinical text, and structured patient data through a multimodal transformer architecture to address the critical gap in medical image-text alignment. Our approach employs separate encoders for each modality: a Vision Transformer (ViT-B/16) for retinal images, Bio-ClinicalBERT for clinical narratives, and a multilayer perceptron for structured demographic and clinical features. These modalities are fused through a joint transformer with modality-specific embeddings, trained using multiple objectives including contrastive losses between modality pairs, reconstruction losses for images and text, and classification losses for DR severity grading according to ICDR and SDRG schemes. Experimental results on the Brazilian Multilabel Ophthalmological Dataset (BRSET) demonstrate significant improvements over baseline models. Our framework achieves near-perfect text-to-image retrieval performance with Recall@1 of 99.94% compared to fine-tuned CLIP's 1.29%, while maintaining state-of-the-art classification accuracy of 97.05% for SDRG and 97.97% for ICDR. Furthermore, zero-shot evaluation on the unseen DeepEyeNet dataset validates strong generalizability with 93.95% Recall@1 versus 0.22% for fine-tuned CLIP. These results demonstrate that our multimodal training approach effectively captures cross-modal relationships in the medical domain, establishing both superior retrieval capabilities and robust diagnostic performance.

cs.CV↗

arXiv2026

REFLECT: Role-Aware Detection of Sentiment, Emotion, Politeness, and Formality

Detecting communicative meaning in dialogue requires modeling not only lexical content but also the participant-role structures through which sentiment, emotion, politeness, and formality (SEPF) are expressed. Existing approaches rely on contextual language representations while leaving semantic roles implicit. We introduce a role-aware transformer architecture that maps the four SEPF dimensions to theory-grounded Semantic Role Labeling (SRL) configurations, incorporating these configurations directly into the transformer's self-attention mechanism. Role-aware attention models non-directional internal states (sentiment and emotion), while directional role-aware attention captures initiator-to-recipient relations for external cues (politeness and formality). The resulting role-aware representations are fused with contextual embeddings through a learned gate that adaptively weights how much role information each input contributes, rather than appending semantic roles as an auxiliary feature vector. Across four benchmark datasets, the framework yields statistically reliable gains over strong BERT-based baselines for emotion, politeness, and formality, with gains concentrated in theoretically predicted cases such as low-frequency, complex emotions and highly polite expressions; sentiment improvements are marginal and non-significant, suggesting that sentiment is conveyed more through lexical choice than participant-role structure. Additional evaluation on real-world data demonstrates that the framework improves both predictive performance and model traceability. These results demonstrate that linguistically grounded modifications to transformer self-attention improve the detection of communicative meaning, generalize beyond benchmark datasets to real-world mental-health data, and yield traceable, role-grounded explanations that extend beyond predictive performance to interpretability.

cs.CL↗

arXiv2026

Putting Transformer Theories to the Test

Several formal theories have been developed to characterize the capabilities and limitations of the transformer architecture, each employing different abstractions that capture different features of the architecture. Due to the differences in their underlying assumptions and conceptual frameworks, these theories can disagree in their predictions about what tasks transformers find easy or hard. We map out seven theories of transformer capability (Star-free, $\mathsf{AC}^0$, $\mathsf{TC}^0$, C-RASP, MPC, BAPO, and Sensitivity) and translate their results into a common matrix of predictions across a suite of tasks covering different computational structures. We provide new theoretical results to complete this matrix. We then evaluate theoretical predictions through controlled experiments measuring in-distribution sample complexity, length generalization, and pretrained language model performance. We find that no single theory explains the observed pattern of successes and failures. When a theory does not capture the transformer behaviors relevant to a task, its predictions break down; for instance, we show that the $k$-hop induction task is predicted hard by C-RASP because it does not model content-dependent attention, but we train small transformers on this task that length-generalize. Similarly, theories assuming hard attention like $\mathsf{AC}^0$ and BAPO predict binary majority to be hard, for which we also find success in practice. Together, our results point to several specific features that a theory of transformers must capture: soft rather than hard attention, dynamic content-dependent retrieval, bounded computation at each position, an account of learnability and inductive biases, and a combination of upper and lower bounds.

cs.LG↗

arXiv2026

ARLoop: Fully-Open Pre-trained Looped Transformers with Attention Residuals

Looped transformers increase effective depth by repeatedly applying a shared set of transformer layers, offering a parameter-efficient way to scale computation. However, standard residual connections aggregate information across recurrent depth into a single stream, providing limited control over which intermediate representations from preceding layers should be reused. In this paper, we introduce ARLoop, a looped transformer architecture based on attention residuals, where the input of each layer is selectively aggregated through attention over model depth. Distinct from standard transformers, we find that looped transformers do not require access to the full residual history: sliding-window depth attention matches or outperforms full-history attention at larger recurrent depths. Across controlled experiments with multiple parameter scales (140M and 370M) and recurrent depths (R = 1 to 5), ARLoop consistently outperforms prior residual designs for looped transformers. More importantly, we further scale ARLoop to 0.6B and 1.2B parameters under a realistic LLM pretraining recipe, demonstrating that the gains transfer to realistic pretraining, with particularly strong improvements on math and coding tasks. Our 1.2B model, pretrained on only 0.5T tokens with three recurrences, outperforms Huginn-3.5B (the only fully-open looped LLM as of our release date) across all evaluated tasks and is competitive with similar-sized non-looped models such as Qwen3.5-2B despite being trained on substantially less data.

cs.CL↗

arXiv2025

Multi-Modality Transformer for E-Commerce: Inferring User Purchase Intention to Bridge the Query-Product Gap

E-commerce click-stream data and product catalogs offer critical user behavior insights and product knowledge. This paper propose a multi-modal transformer termed as PINCER, that leverages the above data sources to transform initial user queries into pseudo-product representations. By tapping into these external data sources, our model can infer users' potential purchase intent from their limited queries and capture query relevant product features. We demonstrate our model's superior performance over state-of-the-art alternatives on e-commerce online retrieval in both controlled and real-world experiments. Our ablation studies confirm that the proposed transformer architecture and integrated learning strategies enable the mining of key data sources to infer purchase intent, extract product features, and enhance the transformation pipeline from queries to more accurate pseudo-product representations.

cs.IR↗