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 1,063 records · Page 59Linked to original sources

arXiv2021

UniDrop: A Simple yet Effective Technique to Improve Transformer without Extra Cost

Transformer architecture achieves great success in abundant natural language processing tasks. The over-parameterization of the Transformer model has motivated plenty of works to alleviate its overfitting for superior performances. With some explorations, we find simple techniques such as dropout, can greatly boost model performance with a careful design. Therefore, in this paper, we integrate different dropout techniques into the training of Transformer models. Specifically, we propose an approach named UniDrop to unites three different dropout techniques from fine-grain to coarse-grain, i.e., feature dropout, structure dropout, and data dropout. Theoretically, we demonstrate that these three dropouts play different roles from regularization perspectives. Empirically, we conduct experiments on both neural machine translation and text classification benchmark datasets. Extensive results indicate that Transformer with UniDrop can achieve around 1.5 BLEU improvement on IWSLT14 translation tasks, and better accuracy for the classification even using strong pre-trained RoBERTa as backbone.

cs.CL↗

arXiv2021

PILOT: Introducing Transformers for Probabilistic Sound Event Localization

Sound event localization aims at estimating the positions of sound sources in the environment with respect to an acoustic receiver (e.g. a microphone array). Recent advances in this domain most prominently focused on utilizing deep recurrent neural networks. Inspired by the success of transformer architectures as a suitable alternative to classical recurrent neural networks, this paper introduces a novel transformer-based sound event localization framework, where temporal dependencies in the received multi-channel audio signals are captured via self-attention mechanisms. Additionally, the estimated sound event positions are represented as multivariate Gaussian variables, yielding an additional notion of uncertainty, which many previously proposed deep learning-based systems designed for this application do not provide. The framework is evaluated on three publicly available multi-source sound event localization datasets and compared against state-of-the-art methods in terms of localization error and event detection accuracy. It outperforms all competing systems on all datasets with statistical significant differences in performance.

cs.SD↗

arXiv2021

Vision Transformer with Progressive Sampling

Transformers with powerful global relation modeling abilities have been introduced to fundamental computer vision tasks recently. As a typical example, the Vision Transformer (ViT) directly applies a pure transformer architecture on image classification, by simply splitting images into tokens with a fixed length, and employing transformers to learn relations between these tokens. However, such naive tokenization could destruct object structures, assign grids to uninterested regions such as background, and introduce interference signals. To mitigate the above issues, in this paper, we propose an iterative and progressive sampling strategy to locate discriminative regions. At each iteration, embeddings of the current sampling step are fed into a transformer encoder layer, and a group of sampling offsets is predicted to update the sampling locations for the next step. The progressive sampling is differentiable. When combined with the Vision Transformer, the obtained PS-ViT network can adaptively learn where to look. The proposed PS-ViT is both effective and efficient. When trained from scratch on ImageNet, PS-ViT performs 3.8% higher than the vanilla ViT in terms of top-1 accuracy with about $4\times$ fewer parameters and $10\times$ fewer FLOPs. Code is available at https://github.com/yuexy/PS-ViT.

cs.CV↗

arXiv2023

Analysing similarities between legal court documents using natural language processing approaches based on Transformers

Recent advances in Artificial Intelligence (AI) have leveraged promising results in solving complex problems in the area of Natural Language Processing (NLP), being an important tool to help in the expeditious resolution of judicial proceedings in the legal area. In this context, this work targets the problem of detecting the degree of similarity between judicial documents that can be achieved in the inference group, by applying six NLP techniques based on the transformers architecture to a case study of legal proceedings in the Brazilian judicial system. The NLP transformer-based models, namely BERT, GPT-2 and RoBERTa, were pre-trained using a general purpose corpora of the Brazilian Portuguese language, and then were fine-tuned and specialised for the legal sector using 210,000 legal proceedings. Vector representations of each legal document were calculated based on their embeddings, which were used to cluster the lawsuits, calculating the quality of each model based on the cosine of the distance between the elements of the group to its centroid. We noticed that models based on transformers presented better performance when compared to previous traditional NLP techniques, with the RoBERTa model specialised for the Brazilian Portuguese language presenting the best results. This methodology can be also applied to other case studies for different languages, making it possible to advance in the current state of the art in the area of NLP applied to the legal sector.

cs.AI↗

arXiv2022

Simple Recurrence Improves Masked Language Models

In this work, we explore whether modeling recurrence into the Transformer architecture can both be beneficial and efficient, by building an extremely simple recurrent module into the Transformer. We compare our model to baselines following the training and evaluation recipe of BERT. Our results confirm that recurrence can indeed improve Transformer models by a consistent margin, without requiring low-level performance optimizations, and while keeping the number of parameters constant. For example, our base model achieves an absolute improvement of 2.1 points averaged across 10 tasks and also demonstrates increased stability in fine-tuning over a range of learning rates.

cs.CL↗

arXiv2023

Pretraining the Vision Transformer using self-supervised methods for vision based Deep Reinforcement Learning

The Vision Transformer architecture has shown to be competitive in the computer vision (CV) space where it has dethroned convolution-based networks in several benchmarks. Nevertheless, convolutional neural networks (CNN) remain the preferential architecture for the representation module in reinforcement learning. In this work, we study pretraining a Vision Transformer using several state-of-the-art self-supervised methods and assess the quality of the learned representations. To show the importance of the temporal dimension in this context we propose an extension of VICReg to better capture temporal relations between observations by adding a temporal order verification task. Our results show that all methods are effective in learning useful representations and avoiding representational collapse for observations from Atari Learning Environment (ALE) which leads to improvements in data efficiency when we evaluated in reinforcement learning (RL). Moreover, the encoder pretrained with the temporal order verification task shows the best results across all experiments, with richer representations, more focused attention maps and sparser representation vectors throughout the layers of the encoder, which shows the importance of exploring such similarity dimension. With this work, we hope to provide some insights into the representations learned by ViT during a self-supervised pretraining with observations from RL environments and which properties arise in the representations that lead to the best-performing agents. The source code will be available at: https://github.com/mgoulao/TOV-VICReg

cs.LG↗

arXiv2024

TransRUPNet for Improved Polyp Segmentation

Colorectal cancer is among the most common cause of cancer worldwide. Removal of precancerous polyps through early detection is essential to prevent them from progressing to colon cancer. We develop an advanced deep learning-based architecture, Transformer based Residual Upsampling Network (TransRUPNet) for automatic and real-time polyp segmentation. The proposed architecture, TransRUPNet, is an encoder-decoder network consisting of three encoder and decoder blocks with additional upsampling blocks at the end of the network. With the image size of $256\times256$, the proposed method achieves an excellent real-time operation speed of 47.07 frames per second with an average mean dice coefficient score of 0.7786 and mean Intersection over Union of 0.7210 on the out-of-distribution polyp datasets. The results on the publicly available PolypGen dataset suggest that TransRUPNet can give real-time feedback while retaining high accuracy for in-distribution datasets. Furthermore, we demonstrate the generalizability of the proposed method by showing that it significantly improves performance on out-of-distribution datasets compared to the existing methods. The source code of our network is available at https://github.com/DebeshJha/TransRUPNet.

eess.IV↗

arXiv2024

Point Clouds Are Specialized Images: A Knowledge Transfer Approach for 3D Understanding

Self-supervised representation learning (SSRL) has gained increasing attention in point cloud understanding, in addressing the challenges posed by 3D data scarcity and high annotation costs. This paper presents PCExpert, a novel SSRL approach that reinterprets point clouds as "specialized images". This conceptual shift allows PCExpert to leverage knowledge derived from large-scale image modality in a more direct and deeper manner, via extensively sharing the parameters with a pre-trained image encoder in a multi-way Transformer architecture. The parameter sharing strategy, combined with a novel pretext task for pre-training, i.e., transformation estimation, empowers PCExpert to outperform the state of the arts in a variety of tasks, with a remarkable reduction in the number of trainable parameters. Notably, PCExpert's performance under LINEAR fine-tuning (e.g., yielding a 90.02% overall accuracy on ScanObjectNN) has already approached the results obtained with FULL model fine-tuning (92.66%), demonstrating its effective and robust representation capability.

cs.CV↗

arXiv2023

Devignet: High-Resolution Vignetting Removal via a Dual Aggregated Fusion Transformer With Adaptive Channel Expansion

Vignetting commonly occurs as a degradation in images resulting from factors such as lens design, improper lens hood usage, and limitations in camera sensors. This degradation affects image details, color accuracy, and presents challenges in computational photography. Existing vignetting removal algorithms predominantly rely on ideal physics assumptions and hand-crafted parameters, resulting in the ineffective removal of irregular vignetting and suboptimal results. Moreover, the substantial lack of real-world vignetting datasets hinders the objective and comprehensive evaluation of vignetting removal. To address these challenges, we present Vigset, a pioneering dataset for vignetting removal. Vigset includes 983 pairs of both vignetting and vignetting-free high-resolution ($5340\times3697$) real-world images under various conditions. In addition, We introduce DeVigNet, a novel frequency-aware Transformer architecture designed for vignetting removal. Through the Laplacian Pyramid decomposition, we propose the Dual Aggregated Fusion Transformer to handle global features and remove vignetting in the low-frequency domain. Additionally, we propose the Adaptive Channel Expansion Module to enhance details in the high-frequency domain. The experiments demonstrate that the proposed model outperforms existing state-of-the-art methods. The code, models, and dataset are available at \url{https://github.com/CXH-Research/DeVigNet}.

cs.CV↗

arXiv2023

Multi-Objective Decision Transformers for Offline Reinforcement Learning

Offline Reinforcement Learning (RL) is structured to derive policies from static trajectory data without requiring real-time environment interactions. Recent studies have shown the feasibility of framing offline RL as a sequence modeling task, where the sole aim is to predict actions based on prior context using the transformer architecture. However, the limitation of this single task learning approach is its potential to undermine the transformer model's attention mechanism, which should ideally allocate varying attention weights across different tokens in the input context for optimal prediction. To address this, we reformulate offline RL as a multi-objective optimization problem, where the prediction is extended to states and returns. We also highlight a potential flaw in the trajectory representation used for sequence modeling, which could generate inaccuracies when modeling the state and return distributions. This is due to the non-smoothness of the action distribution within the trajectory dictated by the behavioral policy. To mitigate this issue, we introduce action space regions to the trajectory representation. Our experiments on D4RL benchmark locomotion tasks reveal that our propositions allow for more effective utilization of the attention mechanism in the transformer model, resulting in performance that either matches or outperforms current state-of-the art methods.

cs.LG↗

arXiv2023

Zebra: Extending Context Window with Layerwise Grouped Local-Global Attention

This paper introduces a novel approach to enhance the capabilities of Large Language Models (LLMs) in processing and understanding extensive text sequences, a critical aspect in applications requiring deep comprehension and synthesis of large volumes of information. Recognizing the inherent challenges in extending the context window for LLMs, primarily built on Transformer architecture, we propose a new model architecture, referred to as Zebra. This architecture efficiently manages the quadratic time and memory complexity issues associated with full attention in the Transformer by employing grouped local-global attention layers. Our model, akin to a zebra's alternating stripes, balances local and global attention layers, significantly reducing computational requirements and memory consumption. Comprehensive experiments, including pretraining from scratch, continuation of long context adaptation training, and long instruction tuning, are conducted to evaluate the Zebra's performance. The results show that Zebra achieves comparable or superior performance on both short and long sequence benchmarks, while also enhancing training and inference efficiency.

cs.CL↗

arXiv2024

Training Transformer Models by Wavelet Losses Improves Quantitative and Visual Performance in Single Image Super-Resolution

Transformer-based models have achieved remarkable results in low-level vision tasks including image super-resolution (SR). However, early Transformer-based approaches that rely on self-attention within non-overlapping windows encounter challenges in acquiring global information. To activate more input pixels globally, hybrid attention models have been proposed. Moreover, training by solely minimizing pixel-wise RGB losses, such as L1, have been found inadequate for capturing essential high-frequency details. This paper presents two contributions: i) We introduce convolutional non-local sparse attention (NLSA) blocks to extend the hybrid transformer architecture in order to further enhance its receptive field. ii) We employ wavelet losses to train Transformer models to improve quantitative and subjective performance. While wavelet losses have been explored previously, showing their power in training Transformer-based SR models is novel. Our experimental results demonstrate that the proposed model provides state-of-the-art PSNR results as well as superior visual performance across various benchmark datasets.

eess.IV↗

arXiv2024

Exploring the Robustness of In-Context Learning with Noisy Labels

Recently, the mysterious In-Context Learning (ICL) ability exhibited by Transformer architectures, especially in large language models (LLMs), has sparked significant research interest. However, the resilience of Transformers' in-context learning capabilities in the presence of noisy samples, prevalent in both training corpora and prompt demonstrations, remains underexplored. In this paper, inspired by prior research that studies ICL ability using simple function classes, we take a closer look at this problem by investigating the robustness of Transformers against noisy labels. Specifically, we first conduct a thorough evaluation and analysis of the robustness of Transformers against noisy labels during in-context learning and show that they exhibit notable resilience against diverse types of noise in demonstration labels. Furthermore, we delve deeper into this problem by exploring whether introducing noise into the training set, akin to a form of data augmentation, enhances such robustness during inference, and find that such noise can indeed improve the robustness of ICL. Overall, our fruitful analysis and findings provide a comprehensive understanding of the resilience of Transformer models against label noises during ICL and provide valuable insights into the research on Transformers in natural language processing. Our code is available at https://github.com/InezYu0928/in-context-learning.

cs.CL↗

arXiv2024

MoEUT: Mixture-of-Experts Universal Transformers

Previous work on Universal Transformers (UTs) has demonstrated the importance of parameter sharing across layers. By allowing recurrence in depth, UTs have advantages over standard Transformers in learning compositional generalizations, but layer-sharing comes with a practical limitation of parameter-compute ratio: it drastically reduces the parameter count compared to the non-shared model with the same dimensionality. Naively scaling up the layer size to compensate for the loss of parameters makes its computational resource requirements prohibitive. In practice, no previous work has succeeded in proposing a shared-layer Transformer design that is competitive in parameter count-dominated tasks such as language modeling. Here we propose MoEUT (pronounced "moot"), an effective mixture-of-experts (MoE)-based shared-layer Transformer architecture, which combines several recent advances in MoEs for both feedforward and attention layers of standard Transformers together with novel layer-normalization and grouping schemes that are specific and crucial to UTs. The resulting UT model, for the first time, slightly outperforms standard Transformers on language modeling tasks such as BLiMP and PIQA, while using significantly less compute and memory.

cs.LG↗

arXiv2024

Cross-Lingual Query-by-Example Spoken Term Detection: A Transformer-Based Approach

Query-by-example spoken term detection (QbE-STD) is typically constrained by transcribed data scarcity and language specificity. This paper introduces a novel, language-agnostic QbE-STD model leveraging image processing techniques and transformer architecture. By employing a pre-trained XLSR-53 network for feature extraction and a Hough transform for detection, our model effectively searches for user-defined spoken terms within any audio file. Experimental results across four languages demonstrate significant performance gains (19-54%) over a CNN-based baseline. While processing time is improved compared to DTW, accuracy remains inferior. Notably, our model offers the advantage of accurately counting query term repetitions within the target audio.

cs.LG↗

arXiv2025

ReHub: Linear Complexity Graph Transformers with Adaptive Hub-Spoke Reassignment

We present ReHub, a novel graph transformer architecture that achieves linear complexity through an efficient reassignment technique between nodes and virtual nodes. Graph transformers have become increasingly important in graph learning for their ability to utilize long-range node communication explicitly, addressing limitations such as oversmoothing and oversquashing found in message-passing graph networks. However, their dense attention mechanism scales quadratically with the number of nodes, limiting their applicability to large-scale graphs. ReHub draws inspiration from the airline industry's hub-and-spoke model, where flights are assigned to optimize operational efficiency. In our approach, graph nodes (spokes) are dynamically reassigned to a fixed number of virtual nodes (hubs) at each model layer. Recent work, Neural Atoms (Li et al., 2024), has demonstrated impressive and consistent improvements over GNN baselines by utilizing such virtual nodes; their findings suggest that the number of hubs strongly influences performance. However, increasing the number of hubs typically raises complexity, requiring a trade-off to maintain linear complexity. Our key insight is that each node only needs to interact with a small subset of hubs to achieve linear complexity, even when the total number of hubs is large. To leverage all hubs without incurring additional computational costs, we propose a simple yet effective adaptive reassignment technique based on hub-hub similarity scores, eliminating the need for expensive node-hub computations. Our experiments on LRGB indicate a consistent improvement in results over the base method, Neural Atoms, while maintaining a linear complexity. Remarkably, our sparse model achieves performance on par with its non-sparse counterpart. Furthermore, ReHub outperforms competitive baselines and consistently ranks among top performers across various benchmarks.

cs.LG↗

arXiv2025

Image-to-Text for Medical Reports Using Adaptive Co-Attention and Triple-LSTM Module

Medical report generation requires specialized expertise that general large models often fail to accurately capture. Moreover, the inherent repetition and similarity in medical data make it difficult for models to extract meaningful features, resulting in a tendency to overfit. So in this paper, we propose a multimodal model, Co-Attention Triple-LSTM Network (CA-TriNet), a deep learning model that combines transformer architectures with a Multi-LSTM network. Its Co-Attention module synergistically links a vision transformer with a text transformer to better differentiate medical images with similarities, augmented by an adaptive weight operator to catch and amplify image labels with minor similarities. Furthermore, its Triple-LSTM module refines generated sentences using targeted image objects. Extensive evaluations over three public datasets have demonstrated that CA-TriNet outperforms state-of-the-art models in terms of comprehensive ability, even pre-trained large language models on some metrics.

cs.CV↗

arXiv2026

Dimensional Collapse in Transformer Attention Outputs: A Challenge for Sparse Dictionary Learning

Transformer architectures, and their attention mechanisms in particular, form the foundation of modern large language models. While transformer models are widely believed to operate in high-dimensional hidden spaces, we show that attention outputs are in fact confined to a surprisingly low-dimensional subspace, with an effective dimensionality of only about $60\%$ of the full space. In contrast, MLP outputs and residual streams remain much closer to full-rank, exhibiting effective ranks around $90\%$. This striking dimensional discrepancy is consistently observed across diverse model families and datasets, and is strongly shaped by the attention output projection matrix. Critically, we find this low-rank structure as a key factor of the prevalent dead feature problem in sparse dictionary learning, where it creates a mismatch between randomly initialized features and the intrinsic geometry of the activation space. Building on this insight, we propose a subspace-constrained training method for sparse autoencoders (SAEs), initializing feature directions into the active subspace of activations. Our approach reduces dead features from 87\% to below 1\% in Attention Output SAEs with 1M features, and can further extend to other sparse dictionary learning methods. Our findings provide both new insights into the geometry of attention and practical tools for improving sparse dictionary learning in large language models.

cs.LG↗