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 1,585 records · Page 88Linked to original sources

arXiv2024

Mesh Denoising Transformer

Mesh denoising, aimed at removing noise from input meshes while preserving their feature structures, is a practical yet challenging task. Despite the remarkable progress in learning-based mesh denoising methodologies in recent years, their network designs often encounter two principal drawbacks: a dependence on single-modal geometric representations, which fall short in capturing the multifaceted attributes of meshes, and a lack of effective global feature aggregation, hindering their ability to fully understand the mesh's comprehensive structure. To tackle these issues, we propose SurfaceFormer, a pioneering Transformer-based mesh denoising framework. Our first contribution is the development of a new representation known as Local Surface Descriptor, which is crafted by establishing polar systems on each mesh face, followed by sampling points from adjacent surfaces using geodesics. The normals of these points are organized into 2D patches, mimicking images to capture local geometric intricacies, whereas the poles and vertex coordinates are consolidated into a point cloud to embody spatial information. This advancement surmounts the hurdles posed by the irregular and non-Euclidean characteristics of mesh data, facilitating a smooth integration with Transformer architecture. Next, we propose a dual-stream structure consisting of a Geometric Encoder branch and a Spatial Encoder branch, which jointly encode local geometry details and spatial information to fully explore multimodal information for mesh denoising. A subsequent Denoising Transformer module receives the multimodal information and achieves efficient global feature aggregation through self-attention operators. Our experimental evaluations demonstrate that this novel approach outperforms existing state-of-the-art methods in both objective and subjective assessments, marking a significant leap forward in mesh denoising.

cs.CV↗

arXiv2024

Dynamical Mean-Field Theory of Self-Attention Neural Networks

Transformer-based models have demonstrated exceptional performance across diverse domains, becoming the state-of-the-art solution for addressing sequential machine learning problems. Even though we have a general understanding of the fundamental components in the transformer architecture, little is known about how they operate or what are their expected dynamics. Recently, there has been an increasing interest in exploring the relationship between attention mechanisms and Hopfield networks, promising to shed light on the statistical physics of transformer networks. However, to date, the dynamical regimes of transformer-like models have not been studied in depth. In this paper, we address this gap by using methods for the study of asymmetric Hopfield networks in nonequilibrium regimes --namely path integral methods over generating functionals, yielding dynamics governed by concurrent mean-field variables. Assuming 1-bit tokens and weights, we derive analytical approximations for the behavior of large self-attention neural networks coupled to a softmax output, which become exact in the large limit size. Our findings reveal nontrivial dynamical phenomena, including nonequilibrium phase transitions associated with chaotic bifurcations, even for very simple configurations with a few encoded features and a very short context window. Finally, we discuss the potential of our analytic approach to improve our understanding of the inner workings of transformer models, potentially reducing computational training costs and enhancing model interpretability.

cond-mat.dis-nn↗

arXiv2024

Multi-Class Abnormality Classification Task in Video Capsule Endoscopy

In this work for Capsule Vision Challenge 2024, we addressed the challenge of multiclass anomaly classification in video capsule Endoscopy (VCE)[1] with a variety of deep learning models, ranging from custom CNNs to advanced transformer architectures. The purpose is to correctly classify diverse gastrointestinal disorders, which is critical for increasing diagnostic efficiency in clinical settings. We started with a baseline CNN model and improved performance with ResNet[2] for better feature extraction, followed by Vision Transformer (ViT)[3] to capture global dependencies. We further improve the results by using Multiscale Vision Transformer (MViT)[4] for improved hierarchical feature extraction, while Dual Attention Vision Transformer (DaViT) [5] delivered best results by combining spatial and channel attention methods. Our best balanced accuracy on validation set [6] was 0.8592 and Mean AUC was 0.9932. This methodology enabled us to improve model accuracy across a wide range of criteria, greatly surpassing all other methods.Additionally, our team capsule commandos achieved 7th place ranking with a test set[7] performance of Mean AUC: 0.7314 and balanced accuracy: 0.3235

eess.IV↗

arXiv2025

Image Restoration via Multi-domain Learning

Due to adverse atmospheric and imaging conditions, natural images suffer from various degradation phenomena. Consequently, image restoration has emerged as a key solution and garnered substantial attention. Although recent Transformer architectures have demonstrated impressive success across various restoration tasks, their considerable model complexity poses significant challenges for both training and real-time deployment. Furthermore, instead of investigating the commonalities among different degradations, most existing restoration methods focus on modifying Transformer under limited restoration priors. In this work, we first review various degradation phenomena under multi-domain perspective, identifying common priors. Then, we introduce a novel restoration framework, which integrates multi-domain learning into Transformer. Specifically, in Token Mixer, we propose a Spatial-Wavelet-Fourier multi-domain structure that facilitates local-region-global multi-receptive field modeling to replace vanilla self-attention. Additionally, in Feed-Forward Network, we incorporate multi-scale learning to fuse multi-domain features at different resolutions. Comprehensive experimental results across ten restoration tasks, such as dehazing, desnowing, motion deblurring, defocus deblurring, rain streak/raindrop removal, cloud removal, shadow removal, underwater enhancement and low-light enhancement, demonstrate that our proposed model outperforms state-of-the-art methods and achieves a favorable trade-off among restoration performance, parameter size, computational cost and inference latency. The code is available at: https://github.com/deng-ai-lab/SWFormer.

eess.IV↗

arXiv2026

From Optimization to Prediction: Transformer-Based Path-Flow Estimation to the Traffic Assignment Problem

The traffic assignment problem is essential for traffic flow analysis, traditionally solved using mathematical programs under the Equilibrium principle. These methods become computationally prohibitive for large-scale networks due to non-linear growth in complexity with the number of OD pairs. This study introduces a novel data-driven approach using deep neural networks, specifically leveraging the Transformer architecture, to predict equilibrium path flows directly. By focusing on path-level traffic distribution, the proposed model captures intricate correlations between OD pairs, offering a more detailed and flexible analysis compared to traditional link-level approaches. The Transformer-based model drastically reduces computation time, while adapting to changes in demand and network structure without the need for recalculation. Numerical experiments are conducted on the Manhattan-like synthetic network, the Sioux Falls network, and the Eastern-Massachusetts network. The results demonstrate that the proposed model is orders of magnitude faster than conventional optimization. It efficiently estimates path-level traffic flows in multi-class networks, reducing computational costs and improving prediction accuracy by capturing detailed trip and flow information. The model also adapts flexibly to varying demand and network conditions, supporting traffic management and enabling rapid `what-if' analyses for enhanced transportation planning and policy-making.

cs.LG↗

arXiv2026

Scaling Self-Supervised and Cross-Modal Pretraining for Volumetric CT Transformers

We introduce SPECTRE, a fully transformer-based foundation model for volumetric computed tomography (CT). Our Self-Supervised & Cross-Modal Pretraining for CT Representation Extraction (SPECTRE) approach utilizes scalable 3D Vision Transformer architectures and modern self-supervised and vision-language pretraining strategies to learn general-purpose CT representations. Volumetric CT poses unique challenges, such as extreme token scaling, geometric anisotropy, and weak or noisy clinical supervision, that make standard transformer and contrastive learning recipes ineffective out of the box. The framework jointly optimizes a local transformer for high-resolution volumetric feature extraction and a global transformer for whole-scan context modeling, making large-scale 3D attention computationally tractable. Notably, SPECTRE is trained exclusively on openly available CT datasets, demonstrating that high-performing, generalizable representations can be achieved without relying on private data. Pretraining combines DINO-style self-distillation with SigLIP-based vision-language alignment using paired radiology reports, yielding features that are both geometrically consistent and clinically meaningful. Across multiple CT benchmarks, SPECTRE consistently outperforms prior CT foundation models in both zero-shot and fine-tuned settings, establishing SPECTRE as a scalable, open, and fully transformer-based foundation model for 3D medical imaging.

cs.CV↗

arXiv2026

Winner-Take-All Spiking Transformer for Language Modeling

Spiking Transformers, which combine the scalability of Transformers with the sparse, energy-efficient property of Spiking Neural Networks (SNNs), have achieved impressive results in neuromorphic and vision tasks and attracted increasing attention. However, existing directly trained spiking transformers primarily focus on vision tasks. For language modeling with spiking transformer, convergence relies heavily on softmax-based spiking self-attention, which incurs high energy costs and poses challenges for neuromorphic deployment. To address this issue, we introduce Winner-Take-All (WTA) mechanisms into spiking transformers and propose two novel softmax-free, spike-driven self-attention modules: WTA Spiking Self-Attention (WSSA) and Causal WTA Spiking Self-Attention (CWSSA). Based on them, we design WTA-based Encoder-only Spiking Transformer (WE-Spikingformer) for masked language modeling and WTA-based Decoder-only Spiking Transformer (WD-Spikingformer) for causal language modeling, systematically exploring softmax-free, spiking-driven Transformer architectures trained end-to-end for natural language processing tasks. Extensive experiments on 16 datasets spanning natural language understanding, question-answering tasks, and commonsense reasoning tasks validate the effectiveness of our approach and highlight the promise of spiking transformers for general language modeling and energy-efficient artificial intelligence.

cs.NE↗

arXiv2026

Geometric Asymmetry in MoE Specialization: Functional Decorrelation and Representational Overlap

Mixture-of-Experts (MoE) architectures achieve scalable capacity through sparse routing, yet the geometric structure of expert specialization remains poorly understood. We introduce a unified Jacobian-PCA-Grassmann framework for analyzing MoE layers in both function space and representation space. Across pretrained MoE Transformers (Mistral, Qwen), we find a consistent structural asymmetry: experts exhibit strong functional decorrelation (consistently low, near-zero cross-expert Jacobian alignment) while their routed representations occupy distinct but partially overlapping subspaces. This indicates that functional decorrelation and representation overlap coexist rather than coincide in MoE specialization. Controlled routing experiments further indicate that routing sparsity appears to be a key factor shaping this geometry: top-k routing induces sharper functional separation and larger subspace divergence, whereas fully soft routing yields more entangled expert structure. Together, these results suggest a geometric interpretation in which MoE layers may be viewed as implementing locally decorrelated operators over overlapping submanifolds on a shared representation manifold, and provide a general diagnostic framework for studying conditional computation in modern Transformer architectures.

cs.LG↗

arXiv2025

EuLearn: A 3D database for learning Euler characteristics

We present EuLearn, the first surface datasets equitably representing a diversity of topological types. We designed our embedded surfaces of uniformly varying genera relying on random knots, thus allowing our surfaces to knot with themselves. EuLearn contributes new topological datasets of meshes, point clouds, and scalar fields in 3D. We aim to facilitate the training of machine learning systems that can discern topological features. We experimented with specific emblematic 3D neural network architectures, finding that their vanilla implementations perform poorly on genus classification. To enhance performance, we developed a novel, non-Euclidean, statistical sampling method adapted to graph and manifold data. We also introduce adjacency-informed adaptations of PointNet and Transformer architectures that rely on our non-Euclidean sampling strategy. Our results demonstrate that incorporating topological information into deep learning workflows significantly improves performance on these otherwise challenging EuLearn datasets.

cs.CG↗

arXiv2022

Hierarchical Neural Network Approaches for Long Document Classification

Text classification algorithms investigate the intricate relationships between words or phrases and attempt to deduce the document's interpretation. In the last few years, these algorithms have progressed tremendously. Transformer architecture and sentence encoders have proven to give superior results on natural language processing tasks. But a major limitation of these architectures is their applicability for text no longer than a few hundred words. In this paper, we explore hierarchical transfer learning approaches for long document classification. We employ pre-trained Universal Sentence Encoder (USE) and Bidirectional Encoder Representations from Transformers (BERT) in a hierarchical setup to capture better representations efficiently. Our proposed models are conceptually simple where we divide the input data into chunks and then pass this through base models of BERT and USE. Then output representation for each chunk is then propagated through a shallow neural network comprising of LSTMs or CNNs for classifying the text data. These extensions are evaluated on 6 benchmark datasets. We show that USE + CNN/LSTM performs better than its stand-alone baseline. Whereas the BERT + CNN/LSTM performs on par with its stand-alone counterpart. However, the hierarchical BERT models are still desirable as it avoids the quadratic complexity of the attention mechanism in BERT. Along with the hierarchical approaches, this work also provides a comparison of different deep learning algorithms like USE, BERT, HAN, Longformer, and BigBird for long document classification. The Longformer approach consistently performs well on most of the datasets.

cs.CL↗

arXiv2019

Transfer Learning from Transformers to Fake News Challenge Stance Detection (FNC-1) Task

In this paper, we report improved results of the Fake News Challenge Stage 1 (FNC-1) stance detection task. This gain in performance is due to the generalization power of large language models based on Transformer architecture, invented, trained and publicly released over the last two years. Specifically (1) we improved the FNC-1 best performing model adding BERT sentence embedding of input sequences as a model feature, (2) we fine-tuned BERT, XLNet, and RoBERTa transformers on FNC-1 extended dataset and obtained state-of-the-art results on FNC-1 task.

cs.CL↗

arXiv2021

Do Transformer Modifications Transfer Across Implementations and Applications?

The research community has proposed copious modifications to the Transformer architecture since it was introduced over three years ago, relatively few of which have seen widespread adoption. In this paper, we comprehensively evaluate many of these modifications in a shared experimental setting that covers most of the common uses of the Transformer in natural language processing. Surprisingly, we find that most modifications do not meaningfully improve performance. Furthermore, most of the Transformer variants we found beneficial were either developed in the same codebase that we used or are relatively minor changes. We conjecture that performance improvements may strongly depend on implementation details and correspondingly make some recommendations for improving the generality of experimental results.

cs.LG↗

arXiv2021

Learning Robust Latent Representations for Controllable Speech Synthesis

State-of-the-art Variational Auto-Encoders (VAEs) for learning disentangled latent representations give impressive results in discovering features like pitch, pause duration, and accent in speech data, leading to highly controllable text-to-speech (TTS) synthesis. However, these LSTM-based VAEs fail to learn latent clusters of speaker attributes when trained on either limited or noisy datasets. Further, different latent variables start encoding the same features, limiting the control and expressiveness during speech synthesis. To resolve these issues, we propose RTI-VAE (Reordered Transformer with Information reduction VAE) where we minimize the mutual information between different latent variables and devise a modified Transformer architecture with layer reordering to learn controllable latent representations in speech data. We show that RTI-VAE reduces the cluster overlap of speaker attributes by at least 30\% over LSTM-VAE and by at least 7\% over vanilla Transformer-VAE.

cs.SD↗

arXiv2022

CoBERL: Contrastive BERT for Reinforcement Learning

Many reinforcement learning (RL) agents require a large amount of experience to solve tasks. We propose Contrastive BERT for RL (CoBERL), an agent that combines a new contrastive loss and a hybrid LSTM-transformer architecture to tackle the challenge of improving data efficiency. CoBERL enables efficient, robust learning from pixels across a wide range of domains. We use bidirectional masked prediction in combination with a generalization of recent contrastive methods to learn better representations for transformers in RL, without the need of hand engineered data augmentations. We find that CoBERL consistently improves performance across the full Atari suite, a set of control tasks and a challenging 3D environment.

cs.LG↗

arXiv2022

Applying Spatiotemporal Attention to Identify Distracted and Drowsy Driving with Vision Transformers

A 20% rise in car crashes in 2021 compared to 2020 has been observed as a result of increased distraction and drowsiness. Drowsy and distracted driving are the cause of 45% of all car crashes. As a means to decrease drowsy and distracted driving, detection methods using computer vision can be designed to be low-cost, accurate, and minimally invasive. This work investigated the use of the vision transformer to outperform state-of-the-art accuracy from 3D-CNNs. Two separate transformers were trained for drowsiness and distractedness. The drowsy video transformer model was trained on the National Tsing-Hua University Drowsy Driving Dataset (NTHU-DDD) with a Video Swin Transformer model for 10 epochs on two classes -- drowsy and non-drowsy simulated over 10.5 hours. The distracted video transformer was trained on the Driver Monitoring Dataset (DMD) with Video Swin Transformer for 50 epochs over 9 distraction-related classes. The accuracy of the drowsiness model reached 44% and a high loss value on the test set, indicating overfitting and poor model performance. Overfitting indicates limited training data and applied model architecture lacked quantifiable parameters to learn. The distracted model outperformed state-of-the-art models on DMD reaching 97.5%, indicating that with sufficient data and a strong architecture, transformers are suitable for unfit driving detection. Future research should use newer and stronger models such as TokenLearner to achieve higher accuracy and efficiency, merge existing datasets to expand to detecting drunk driving and road rage to create a comprehensive solution to prevent traffic crashes, and deploying a functioning prototype to revolutionize the automotive safety industry.

cs.CV↗

arXiv2023

Data-Efficient Image Quality Assessment with Attention-Panel Decoder

Blind Image Quality Assessment (BIQA) is a fundamental task in computer vision, which however remains unresolved due to the complex distortion conditions and diversified image contents. To confront this challenge, we in this paper propose a novel BIQA pipeline based on the Transformer architecture, which achieves an efficient quality-aware feature representation with much fewer data. More specifically, we consider the traditional fine-tuning in BIQA as an interpretation of the pre-trained model. In this way, we further introduce a Transformer decoder to refine the perceptual information of the CLS token from different perspectives. This enables our model to establish the quality-aware feature manifold efficiently while attaining a strong generalization capability. Meanwhile, inspired by the subjective evaluation behaviors of human, we introduce a novel attention panel mechanism, which improves the model performance and reduces the prediction uncertainty simultaneously. The proposed BIQA method maintains a lightweight design with only one layer of the decoder, yet extensive experiments on eight standard BIQA datasets (both synthetic and authentic) demonstrate its superior performance to the state-of-the-art BIQA methods, i.e., achieving the SRCC values of 0.875 (vs. 0.859 in LIVEC) and 0.980 (vs. 0.969 in LIVE).

cs.CV↗

arXiv2024

State Space Models for Event Cameras

Today, state-of-the-art deep neural networks that process event-camera data first convert a temporal window of events into dense, grid-like input representations. As such, they exhibit poor generalizability when deployed at higher inference frequencies (i.e., smaller temporal windows) than the ones they were trained on. We address this challenge by introducing state-space models (SSMs) with learnable timescale parameters to event-based vision. This design adapts to varying frequencies without the need to retrain the network at different frequencies. Additionally, we investigate two strategies to counteract aliasing effects when deploying the model at higher frequencies. We comprehensively evaluate our approach against existing methods based on RNN and Transformer architectures across various benchmarks, including Gen1 and 1 Mpx event camera datasets. Our results demonstrate that SSM-based models train 33% faster and also exhibit minimal performance degradation when tested at higher frequencies than the training input. Traditional RNN and Transformer models exhibit performance drops of more than 20 mAP, with SSMs having a drop of 3.76 mAP, highlighting the effectiveness of SSMs in event-based vision tasks.

cs.CV↗

arXiv2024

Dynamic Context Adaptation and Information Flow Control in Transformers: Introducing the Evaluator Adjuster Unit and Gated Residual Connections

Transformers have revolutionized various domains of artificial intelligence due to their unique ability to model long-range dependencies in data. However, they lack in nuanced, context-dependent modulation of features and information flow. This paper introduces two significant enhancements to the transformer architecture - the Evaluator Adjuster Unit (EAU) and Gated Residual Connections (GRC) - designed to address these limitations. The EAU dynamically modulates attention outputs based on the relevance of the input context, allowing for more adaptive response patterns. Concurrently, the GRC modifies the transformer's residual connections through a gating mechanism that selectively controls the information flow, thereby enhancing the network's ability to focus on contextually important features. We evaluate the performance of these enhancements across several benchmarks in natural language processing. Our results demonstrate improved adaptability and efficiency, suggesting that these modifications could set new standards for designing flexible and context-aware transformer models.

cs.LG↗