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 1,621 records · Page 90Linked to original sources

arXiv2025

DarSwin-Unet: Distortion Aware Encoder-Decoder Architecture

Wide-angle fisheye images are becoming increasingly common for perception tasks in applications such as robotics, security, and mobility (e.g. drones, avionics). However, current models often either ignore the distortions in wide-angle images or are not suitable to perform pixel-level tasks. In this paper, we present an encoder-decoder model based on a radial transformer architecture that adapts to distortions in wide-angle lenses by leveraging the physical characteristics defined by the radial distortion profile. In contrast to the original model, which only performs classification tasks, we introduce a U-Net architecture, DarSwin-Unet, designed for pixel level tasks. Furthermore, we propose a novel strategy that minimizes sparsity when sampling the image for creating its input tokens. Our approach enhances the model capability to handle pixel-level tasks in wide-angle fisheye images, making it more effective for real-world applications. Compared to other baselines, DarSwin-Unet achieves the best results across different datasets, with significant gains when trained on bounded levels of distortions (very low, low, medium, and high) and tested on all, including out-of-distribution distortions. We demonstrate its performance on depth estimation and show through extensive experiments that DarSwin-Unet can perform zero-shot adaptation to unseen distortions of different wide-angle lenses.

cs.CV↗

arXiv2026

On the Necessity of Attention-FFN Split in Vision Transformers

The standard Transformer architecture relies on a rigid pattern that alternates Attention and Feed-Forward Network (FFN) layers. Despite its widespread adoption, the inductive bias imposed by this strict separation has not been systematically examined. In this work, we investigate the necessity of the Attention-FFN dichotomy in Vision Transformers (ViTs). To facilitate this analysis, we introduce the AttenFeed module, a unified component that integrates the functional properties of both Attention and FFN. Based on this module, we devise the unified Vision Transformer (uViT), which replaces the conventional alternating Attention-FFN structure with a sequence of AttenFeed modules. We then use uViT as a control group that relaxes the Attention-FFN dichotomy of the standard ViT and systematically compare the two models across multiple datasets and model scales. Our experiments reveal that the Attention-FFN dichotomy can hinder performance at smaller model scales due to the rigid parameter allocation of ViTs. The AttenFeed module and uViT serve as new analytical tools for understanding the Attention-FFN structure and offer theoretical insights into the heuristically designed architecture of conventional ViTs.

cs.CV↗

arXiv2022

Memory Augmented Lookup Dictionary based Language Modeling for Automatic Speech Recognition

Recent studies have shown that using an external Language Model (LM) benefits the end-to-end Automatic Speech Recognition (ASR). However, predicting tokens that appear less frequently in the training set is still quite challenging. The long-tail prediction problems have been widely studied in many applications, but only been addressed by a few studies for ASR and LMs. In this paper, we propose a new memory augmented lookup dictionary based Transformer architecture for LM. The newly introduced lookup dictionary incorporates rich contextual information in training set, which is vital to correctly predict long-tail tokens. With intensive experiments on Chinese and English data sets, our proposed method is proved to outperform the baseline Transformer LM by a great margin on both word/character error rate and tail tokens error rate. This is achieved without impact on the decoding efficiency. Overall, we demonstrate the effectiveness of our proposed method in boosting the ASR decoding performance, especially for long-tail tokens.

cs.CL↗

arXiv2023

Priors are Powerful: Improving a Transformer for Multi-camera 3D Detection with 2D Priors

Transfomer-based approaches advance the recent development of multi-camera 3D detection both in academia and industry. In a vanilla transformer architecture, queries are randomly initialised and optimised for the whole dataset, without considering the differences among input frames. In this work, we propose to leverage the predictions from an image backbone, which is often highly optimised for 2D tasks, as priors to the transformer part of a 3D detection network. The method works by (1). augmenting image feature maps with 2D priors, (2). sampling query locations via ray-casting along 2D box centroids, as well as (3). initialising query features with object-level image features. Experimental results shows that 2D priors not only help the model converge faster, but also largely improve the baseline approach by up to 12% in terms of average precision.

cs.CV↗

arXiv2024

Transformers Learn Nonlinear Features In Context: Nonconvex Mean-field Dynamics on the Attention Landscape

Large language models based on the Transformer architecture have demonstrated impressive capabilities to learn in context. However, existing theoretical studies on how this phenomenon arises are limited to the dynamics of a single layer of attention trained on linear regression tasks. In this paper, we study the optimization of a Transformer consisting of a fully connected layer followed by a linear attention layer. The MLP acts as a common nonlinear representation or feature map, greatly enhancing the power of in-context learning. We prove in the mean-field and two-timescale limit that the infinite-dimensional loss landscape for the distribution of parameters, while highly nonconvex, becomes quite benign. We also analyze the second-order stability of mean-field dynamics and show that Wasserstein gradient flow almost always avoids saddle points. Furthermore, we establish novel methods for obtaining concrete improvement rates both away from and near critical points. This represents the first saddle point analysis of mean-field dynamics in general and the techniques are of independent interest.

stat.ML↗

arXiv2024

Enhancing Maritime Trajectory Forecasting via H3 Index and Causal Language Modelling (CLM)

The prediction of ship trajectories is a growing field of study in artificial intelligence. Traditional methods rely on the use of LSTM, GRU networks, and even Transformer architectures for the prediction of spatio-temporal series. This study proposes a viable alternative for predicting these trajectories using only GNSS positions. It considers this spatio-temporal problem as a natural language processing problem. The latitude/longitude coordinates of AIS messages are transformed into cell identifiers using the H3 index. Thanks to the pseudo-octal representation, it becomes easier for language models to learn the spatial hierarchy of the H3 index. The method is compared with a classical Kalman filter, widely used in the maritime domain, and introduces the Fréchet distance as the main evaluation metric. We show that it is possible to predict ship trajectories quite precisely up to 8 hours ahead with 30 minutes of context, using solely GNSS positions, without relying on any additional information such as speed, course, or external conditions - unlike many traditional methods. We demonstrate that this alternative works well enough to predict trajectories worldwide.

cs.LG↗

arXiv2024

Enhancing Lie Detection Accuracy: A Comparative Study of Classic ML, CNN, and GCN Models using Audio-Visual Features

Inaccuracies in polygraph tests often lead to wrongful convictions, false information, and bias, all of which have significant consequences for both legal and political systems. Recently, analyzing facial micro-expressions has emerged as a method for detecting deception; however, current models have not reached high accuracy and generalizability. The purpose of this study is to aid in remedying these problems. The unique multimodal transformer architecture used in this study improves upon previous approaches by using auditory inputs, visual facial micro-expressions, and manually transcribed gesture annotations, moving closer to a reliable non-invasive lie detection model. Visual and auditory features were extracted using the Vision Transformer and OpenSmile models respectively, which were then concatenated with the transcriptions of participants micro-expressions and gestures. Various models were trained for the classification of lies and truths using these processed and concatenated features. The CNN Conv1D multimodal model achieved an average accuracy of 95.4%. However, further research is still required to create higher-quality datasets and even more generalized models for more diverse applications.

cs.MM↗

arXiv2025

EquAct: An SE(3)-Equivariant Multi-Task Transformer for Open-Loop Robotic Manipulation

Transformer architectures can effectively learn language-conditioned, multi-task 3D open-loop manipulation policies from demonstrations by jointly processing natural language instructions and 3D observations. However, although both the robot policy and language instructions inherently encode rich 3D geometric structures, standard transformers lack built-in guarantees of geometric consistency, often resulting in unpredictable behavior under SE(3) transformations of the scene. In this paper, we leverage SE(3) equivariance as a key structural property shared by both policy and language, and propose EquAct-a novel SE(3)-equivariant multi-task transformer. EquAct is theoretically guaranteed to be SE(3) equivariant and consists of two key components: (1) an efficient SE(3)-equivariant point cloud-based U-net with spherical Fourier features for policy reasoning, and (2) SE(3)-invariant Feature-wise Linear Modulation (iFiLM) layers for language conditioning. To evaluate its spatial generalization ability, we benchmark EquAct on 18 RLBench simulation tasks with both SE(3) and SE(2) scene perturbations, and on 4 physical tasks. EquAct performs state-of-the-art across these simulation and physical tasks.

cs.RO↗

arXiv2026

Transformers Can Implement Preconditioned Richardson Iteration for In-Context Gaussian Kernel Regression

In this paper, we study in-context kernel ridge regression (KRR) with Gaussian kernels and show, both theoretically and empirically, that a standard softmax-attention transformer can approximate the KRR predictor during its forward pass. Under bounded-data assumptions, we construct a single-head transformer whose forward pass approximately implements \textit{preconditioned Richardson iteration} on the associated kernel system. The construction uses $O(\log(1/ε))$ blocks and MLP width $O(\sqrt{N/ε})$ to achieve $ε$-accurate prediction for prompts of length $N$. Our construction reveals a functional decomposition within the transformer architecture: softmax attention produces a row-normalized Gaussian-kernel operator needed for \emph{cross-token} interactions, while MLP layers act locally to approximate the \emph{intra-token} scalar arithmetic required by the update. Empirically, we train GPT-2-style transformers on Gaussian-process regression tasks and observe that they progressively align with the exact Gaussian KRR estimator across depth in terms of both \emph{prediction error} and \emph{induced weights}, with ablations further supporting this trend. Comparisons with classical KRR solvers also show that deeper layers align with later solver iterates. Together, we empirically demonstrate that the pretrained transformers exhibit progressive refinement toward exact Gaussian KRR across depth, and theoretically establish inexact preconditioned Richardson iteration as a concrete mechanism for approximating this predictor within an explicitly constructed softmax-attention transformer.

cs.LG↗

arXiv2026

LEOSTP: A Spatio-Temporal Traffic Prediction Framework for LEO Satellite Networks

With the evolution of next-generation mobile communication networks and the commercial boom of Low Earth Orbit (LEO) satellites, globally covered satellite networks are gradually becoming a crucial infrastructure for massive user access and seamless connectivity. Accurate traffic prediction is crucial for maintaining the quality of service (QoS) and resource allocation efficiency in satellite networks. However, existing methods struggle to effectively address the three major challenges of LEO networks: highly complex temporal dynamics caused by satellite cross-regional movement, multivariate dependencies in multi-satellite collaboration, and strong spatial heterogeneity driven by user distribution, human activity intensity, and local geographic environments. In this article, we propose a LEO Satellite Traffic Predictor (LEOSTP) framework, a diffusion model-based end-to-end model that forecasts future satellite traffic by jointly leveraging historical traffic patterns and contextual characteristics of the corresponding service regions. The framework consists of two core modules: 1) The general traffic feature extractor module combines the diffusion process with a Transformer architecture to model the multi-scale temporal features of the traffic itself. 2) The external condition encoder module integrates geographic semantic information such as population distribution, point-of-interest (POI) distribution, and local time into the prediction process through a Transformer-based encoder. In this way, the model captures the deep correlation between the external environment and traffic dynamics. Experimental results based on large-scale simulated constellation data show that LEOSTP significantly outperforms traditional statistical models such as ARIMA and SVR, and classical sequence models including LSTM and Transformer, in prediction accuracy.

cs.IT↗

arXiv2024

Extracting thin film structures of energy materials using transformers

Neutron-Transformer Reflectometry and Advanced Computation Engine (N-TRACE ), a neural network model using transformer architecture, is introduced for neutron reflectometry data analysis. It offers fast, accurate initial parameter estimations and efficient refinements, improving efficiency and precision for real-time data analysis of lithium-mediated nitrogen reduction for electrochemical ammonia synthesis, with relevance to other chemical transformations and batteries. Despite limitations in generalizing across systems, it shows promises for the use of transformers as the basis for models that could replace trial-and-error approaches to modeling reflectometry data.

physics.comp-ph↗

arXiv2019

Voice Transformer Network: Sequence-to-Sequence Voice Conversion Using Transformer with Text-to-Speech Pretraining

We introduce a novel sequence-to-sequence (seq2seq) voice conversion (VC) model based on the Transformer architecture with text-to-speech (TTS) pretraining. Seq2seq VC models are attractive owing to their ability to convert prosody. While seq2seq models based on recurrent neural networks (RNNs) and convolutional neural networks (CNNs) have been successfully applied to VC, the use of the Transformer network, which has shown promising results in various speech processing tasks, has not yet been investigated. Nonetheless, their data-hungry property and the mispronunciation of converted speech make seq2seq models far from practical. To this end, we propose a simple yet effective pretraining technique to transfer knowledge from learned TTS models, which benefit from large-scale, easily accessible TTS corpora. VC models initialized with such pretrained model parameters are able to generate effective hidden representations for high-fidelity, highly intelligible converted speech. Experimental results show that such a pretraining scheme can facilitate data-efficient training and outperform an RNN-based seq2seq VC model in terms of intelligibility, naturalness, and similarity.

eess.AS↗

arXiv2023

Encoding Sentence Position in Context-Aware Neural Machine Translation with Concatenation

Context-aware translation can be achieved by processing a concatenation of consecutive sentences with the standard Transformer architecture. This paper investigates the intuitive idea of providing the model with explicit information about the position of the sentences contained in the concatenation window. We compare various methods to encode sentence positions into token representations, including novel methods. Our results show that the Transformer benefits from certain sentence position encoding methods on English to Russian translation if trained with a context-discounted loss (Lupo et al., 2022). However, the same benefits are not observed in English to German. Further empirical efforts are necessary to define the conditions under which the proposed approach is beneficial.

cs.CL↗

arXiv2023

Skill Transformer: A Monolithic Policy for Mobile Manipulation

We present Skill Transformer, an approach for solving long-horizon robotic tasks by combining conditional sequence modeling and skill modularity. Conditioned on egocentric and proprioceptive observations of a robot, Skill Transformer is trained end-to-end to predict both a high-level skill (e.g., navigation, picking, placing), and a whole-body low-level action (e.g., base and arm motion), using a transformer architecture and demonstration trajectories that solve the full task. It retains the composability and modularity of the overall task through a skill predictor module while reasoning about low-level actions and avoiding hand-off errors, common in modular approaches. We test Skill Transformer on an embodied rearrangement benchmark and find it performs robust task planning and low-level control in new scenarios, achieving a 2.5x higher success rate than baselines in hard rearrangement problems.

cs.LG↗

arXiv2024

VideoPoet: A Large Language Model for Zero-Shot Video Generation

We present VideoPoet, a language model capable of synthesizing high-quality video, with matching audio, from a large variety of conditioning signals. VideoPoet employs a decoder-only transformer architecture that processes multimodal inputs -- including images, videos, text, and audio. The training protocol follows that of Large Language Models (LLMs), consisting of two stages: pretraining and task-specific adaptation. During pretraining, VideoPoet incorporates a mixture of multimodal generative objectives within an autoregressive Transformer framework. The pretrained LLM serves as a foundation that can be adapted for a range of video generation tasks. We present empirical results demonstrating the model's state-of-the-art capabilities in zero-shot video generation, specifically highlighting VideoPoet's ability to generate high-fidelity motions. Project page: http://sites.research.google/videopoet/

cs.CV↗

arXiv2024

Dynamic Universal Approximation Theory: The Basic Theory for Transformer-based Large Language Models

Language models have emerged as a critical area of focus in artificial intelligence, particularly with the introduction of groundbreaking innovations like ChatGPT. Large-scale Transformer networks have quickly become the leading approach for advancing natural language processing algorithms. Built on the Transformer architecture, these models enable interactions that closely mimic human communication and, equipped with extensive knowledge, can even assist in guiding human tasks. Despite their impressive capabilities and growing complexity, a key question remains-the theoretical foundations of large language models (LLMs). What makes Transformer so effective for powering intelligent language applications, such as translation and coding? What underlies LLMs' ability for In-Context Learning (ICL)? How does the LoRA scheme enhance the fine-tuning of LLMs? And what supports the practicality of pruning LLMs? To address these critical questions and explore the technological strategies within LLMs, we leverage the Universal Approximation Theory (UAT) to offer a theoretical backdrop, shedding light on the mechanisms that underpin these advancements.

cs.AI↗

arXiv2026

Universal Approximation of Operators with Transformers and Neural Integral Operators

We study the universal approximation properties of transformers and neural integral operators for operators in Banach spaces. In particular, we show that the transformer architecture is a universal approximator of integral operators between Hölder spaces. Moreover, we show that a generalized version of neural integral operators, based on the Gavurin integral, are universal approximators of arbitrary operators between Banach spaces. Lastly, we show that a modified version of transformer, which uses Leray-Schauder mappings, is a universal approximator of operators between arbitrary Banach spaces.

cs.LG↗

arXiv2025

Fundamental Limitations on Subquadratic Alternatives to Transformers

The Transformer architecture is widely deployed in many popular and impactful Large Language Models. At its core is the attention mechanism for calculating correlations between pairs of tokens. Performing an attention computation takes quadratic time in the input size, and had become the time bottleneck for transformer operations. In order to circumvent this, researchers have used a variety of approaches, including designing heuristic algorithms for performing attention computations faster, and proposing alternatives to the attention mechanism which can be computed more quickly. For instance, state space models such as Mamba were designed to replace attention with an almost linear time alternative. In this paper, we prove that any such approach cannot perform important tasks that Transformer is able to perform (assuming a popular conjecture from fine-grained complexity theory). We focus on document similarity tasks, where one is given as input many documents and would like to find a pair which is (approximately) the most similar. We prove that Transformer is able to perform this task, and we prove that this task cannot be performed in truly subquadratic time by any algorithm. Thus, any model which can be evaluated in subquadratic time - whether because of subquadratic-time heuristics for attention, faster attention replacements like Mamba, or any other reason - cannot perform this task. In other words, in order to perform tasks that (implicitly or explicitly) involve document similarity, one may as well use Transformer and cannot avoid its quadratic running time.

cs.LG↗