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 1,639 records · Page 91Linked to original sources

arXiv2025

Your Attention Matters: to Improve Model Robustness to Noise and Spurious Correlations

Self-attention mechanisms are foundational to Transformer architectures, supporting their impressive success in a wide range of tasks. While there are many self-attention variants, their robustness to noise and spurious correlations has not been well studied. This study evaluates Softmax, Sigmoid, Linear, Doubly Stochastic, and Cosine attention within Vision Transformers under different data corruption scenarios. Through testing across the CIFAR-10, CIFAR-100, and Imagenette datasets, we show that Doubly Stochastic attention is the most robust. It consistently outperformed the next best mechanism by $0.1\%-5.1\%$ when training data, or both training and testing data, were corrupted. Our findings inform self-attention selection in contexts with imperfect data. The code used is available at https://github.com/ctamayor/NeurIPS-Robustness-ViT.

cs.LG↗

arXiv2026

The Deterministic Horizon: Impossibility Results as Design Specifications for Trustworthy AI Systems

Large language models now write software, draft legal documents, and produce clinical notes, yet fundamental limits, from Turing and Arrow to the No Free Lunch theorems, shape what computation can do. This thesis turns such impossibility results from curiosities into design rules. Its flagship result proves an accuracy ceiling set by architecture alone: past a critical reasoning depth, no amount of training moves it, at any adapter rank, sample size, or loss function. Computable before deployment from layer count and embedding width, this Deterministic Horizon is measured between nineteen and thirty-one across twelve transformer architectures, and fine-tuning on optimal-length traces recovers under four percentage points. The mechanism is a capacity invariant of the residual stream, and an information-theoretic conversion yields super-exponential accuracy decay past the horizon. An unconditional circuit-complexity lower bound for modular exponentiation against constant-depth prime-modulus circuits complements this result. The same argument recasts across subfields: preference learning under any misspecified model jumps discontinuously in sample complexity; multi-stage retrieval pipelines require at least as many independent metrics as stages; standard truthful auctions fail for agents with prompt-dependent valuations; and zero-knowledge verification of neural inference pays a measured overhead of one hundred ten to one hundred ninety times per non-linear activation. Together these form a catalogue of sixteen specifications, each pairing a computable boundary, a quantified violation cost, and a constructive design rule: two compositions are proved, one pairing is an honest obstruction, and four remain open. The impossibility-specification methodology is offered for the generative research programme that trustworthy AI may need. Every fundamental limit of AI is also a design rule.

cs.AI↗

arXiv2026

ConsisFormer: Compute-Efficient Transformer for Wireless Foundation Models Based on Channel Consistency

Wireless foundation models (WFMs) have recently emerged as a promising paradigm for AI-native 6G networks, enabling universal channel representations adaptable to diverse communication and sensing tasks. Existing WFMs are predominantly built upon the Transformer architecture, which delivers superior performance but incurs computational complexity proportional to the square of the input sequence length, posing a significant barrier to their deployment under stringent inference latency constraints. To address this issue, in this paper, we propose ConsisFormer, a compute-efficient Transformer design based on short-term consistency of wireless channels, as a WFM backbone. By utilizing the observation that adjacent time or frequency instances share similar clusters of scatterers and thus exhibit similar channel characteristics, we develop an adaptive token aggregation (ATA) module to dynamically merge neighboring channel state information (CSI) tokens, thereby reducing the length of the token sequence involved in self-attention calculations to lower the computational cost. Furthermore, we propose a feature sequence interpolation (FSI) method to recover the full CSI representation based on the sparse feature sequence outputted from the Transformer blocks, thus keeping the performance unaffected while ensuring low complexity. Moreover, we propose an aggregated auto-encoder (AAE) pre-training paradigm for WFMs, enabling robust channel representation learning from sparsified CSI tokens via compression and recovery. Simulation results show that the proposed design reduces the computational complexity of WFM by over $83\%$ with negligible performance loss on various tasks including channel prediction, LoS/NLOS classification, beam prediction, and localization.

eess.SP↗

arXiv2026

XRFormer: Multiscale Tokenization for XRF Representation Learning

X-ray fluorescence (XRF) spectroscopy is a key modality for material analysis in cultural heritage. However, automated learning from XRF spectra remains challenging: XRF spectra are complex one-dimensional signals composed of sharp elemental peaks, broader structures, and background variations that are not taken into account by existing learning-based models. This paper introduces XRFormer, a transformer architecture tailored to XRF spectra through a multiscale convolutional tokenizer that injects locality and multi-resolution inductive biases before global self-attention. The tokenizer progressively reduces spectral resolution while increasing embedding dimensionality, and the resulting token sequence is processed by a standard transformer encoder. We further investigate self-supervised pretraining for XRF representation learning using Masked Spectral modeling (MSM) and a physics-informed Peak Presence Prediction (PPP) objective. Experiments on the Pigments Checker STANDARD v.5 dataset for pigment identification and unmixing show that XRFormer consistently outperforms ViT, SpectralFormer (with and without CAF), and a 1D-CNN baseline for pigment identification. For pigment unmixing, XRFormer achieves robust abundance estimation while maintaining significantly higher parameter efficiency than SpectralFormer, operating at a lower token resolution (128 vs. 512 tokens) and with less than half the number of parameters (1.5M vs. 3.37M). MSM yields consistent gains across both tasks, while PPP further enhances performance for both identification and unmixing when tuned with an appropriate peak prominence. These results highlight multiscale, modality-aware tokenization as an effective and parameter efficient foundation for transformer-based XRF modeling under data-limited conditions. A GitHub repository is provided at https://github.com/sofiane1010/XRFormer.

cs.CV↗

arXiv2023

Legal-HNet: Mixing Legal Long-Context Tokens with Hartley Transform

Since its introduction, the transformers architecture has seen great adoption in NLP applications, but it also has limitations. Although the self-attention mechanism allows for generating very rich representations of the input text, its effectiveness may be limited in specialized domains such as legal, where, for example, language models often have to process very long texts. In this paper, we explore alternatives to replace the attention-based layers with simpler token-mixing mechanisms: Hartley and Fourier transforms. Using these non-parametric techniques, we train models with long input documents from scratch in the legal domain setting. We also introduce a new hybrid Seq2Seq architecture, a no-attention-based encoder connected with an attention-based decoder, which performs quite well on existing summarization tasks with much less compute and memory requirements. We believe that similar, if not better performance, as in the case of long correlations of abstractive text summarization tasks, can be achieved by adopting these simpler infrastructures. This not only makes training models from scratch accessible to more people, but also contributes to the reduction of the carbon footprint during training.

cs.CL↗

arXiv2021

Blending Anti-Aliasing into Vision Transformer

The transformer architectures, based on self-attention mechanism and convolution-free design, recently found superior performance and booming applications in computer vision. However, the discontinuous patch-wise tokenization process implicitly introduces jagged artifacts into attention maps, arising the traditional problem of aliasing for vision transformers. Aliasing effect occurs when discrete patterns are used to produce high frequency or continuous information, resulting in the indistinguishable distortions. Recent researches have found that modern convolution networks still suffer from this phenomenon. In this work, we analyze the uncharted problem of aliasing in vision transformer and explore to incorporate anti-aliasing properties. Specifically, we propose a plug-and-play Aliasing-Reduction Module(ARM) to alleviate the aforementioned issue. We investigate the effectiveness and generalization of the proposed method across multiple tasks and various vision transformer families. This lightweight design consistently attains a clear boost over several famous structures. Furthermore, our module also improves data efficiency and robustness of vision transformers.

cs.CV↗

arXiv2024

FLuRKA: Fast and accurate unified Low-Rank & Kernel Attention

Many efficient $\textit{approximate}$ self-attention techniques have become prevalent since the inception of the transformer architecture. Two popular classes of these techniques are low-rank and kernel methods. Each of these methods has its strengths. We observe these strengths synergistically complement each other and exploit them to fuse low-rank and kernel methods, producing a new class of transformers: FLuRKA ($\textbf{F}$ast $\textbf{L}$ow-$\textbf{R}$ank & $\textbf{K}$ernel$ \textbf{A}$ttention). FLuRKA are highly $\textit{training-efficient}$ with faster model speeds $\textit{and}$ similar model qualities compared to constituent low-rank and kernel methods. We theoretically and empirically evaluate the speed and quality of FLuRKA. Our model speed analysis posits a variety of parameter configurations where FLuRKA exhibit speedups over low-rank and kernel approximations and our model quality analysis bounds the error of FLuRKA with respect to full-attention. Empirically, we instantiate three FLuRKA variants which experience speedups of up to 3.3x and 1.7x over low-rank and kernel methods respectively. This translates to speedups of up to 20x over models with flash-attention. Across a diverse set of tasks spanning language modeling, language understanding, long sequence modeling, machine translation, and image classification, FLuRKA achieve comparable accuracy with underlying low-rank and kernel approximations, occasionally surpassing both.

cs.LG↗

arXiv2024

Functional Interpolation for Relative Positions Improves Long Context Transformers

Preventing the performance decay of Transformers on inputs longer than those used for training has been an important challenge in extending the context length of these models. Though the Transformer architecture has fundamentally no limits on the input sequence lengths it can process, the choice of position encoding used during training can limit the performance of these models on longer inputs. We propose a novel functional relative position encoding with progressive interpolation, FIRE, to improve Transformer generalization to longer contexts. We theoretically prove that this can represent some of the popular relative position encodings, such as T5's RPE, Alibi, and Kerple. We next empirically show that FIRE models have better generalization to longer contexts on both zero-shot language modeling and long text benchmarks.

cs.LG↗

arXiv2023

Towards Few-Annotation Learning in Computer Vision: Application to Image Classification and Object Detection tasks

In this thesis, we develop theoretical, algorithmic and experimental contributions for Machine Learning with limited labels, and more specifically for the tasks of Image Classification and Object Detection in Computer Vision. In a first contribution, we are interested in bridging the gap between theory and practice for popular Meta-Learning algorithms used in Few-Shot Classification. We make connections to Multi-Task Representation Learning, which benefits from solid theoretical foundations, to verify the best conditions for a more efficient meta-learning. Then, to leverage unlabeled data when training object detectors based on the Transformer architecture, we propose both an unsupervised pretraining and a semi-supervised learning method in two other separate contributions. For pretraining, we improve Contrastive Learning for object detectors by introducing the localization information. Finally, our semi-supervised method is the first tailored to transformer-based detectors.

cs.CV↗

arXiv2024

Recurrence over Video Frames (RoVF) for the Re-identification of Meerkats

Deep learning approaches for animal re-identification have had a major impact on conservation, significantly reducing the time required for many downstream tasks, such as well-being monitoring. We propose a method called Recurrence over Video Frames (RoVF), which uses a recurrent head based on the Perceiver architecture to iteratively construct an embedding from a video clip. RoVF is trained using triplet loss based on the co-occurrence of individuals in the video frames, where the individual IDs are unavailable. We tested this method and various models based on the DINOv2 transformer architecture on a dataset of meerkats collected at the Wellington Zoo. Our method achieves a top-1 re-identification accuracy of $49\%$, which is higher than that of the best DINOv2 model ($42\%$). We found that the model can match observations of individuals where humans cannot, and our model (RoVF) performs better than the comparisons with minimal fine-tuning. In future work, we plan to improve these models by using pre-text tasks, apply them to animal behaviour classification, and perform a hyperparameter search to optimise the models further.

cs.CV↗

arXiv2024

Adaptive Step-size Perception Unfolding Network with Non-local Hybrid Attention for Hyperspectral Image Reconstruction

Deep unfolding methods and transformer architecture have recently shown promising results in hyperspectral image (HSI) reconstruction. However, there still exist two issues: (1) in the data subproblem, most methods represents the stepsize utilizing a learnable parameter. Nevertheless, for different spectral channel, error between features and ground truth is unequal. (2) Transformer struggles to balance receptive field size with pixel-wise detail information. To overcome the aforementioned drawbacks, We proposed an adaptive step-size perception unfolding network (ASPUN), a deep unfolding network based on FISTA algorithm, which uses an adaptive step-size perception module to estimate the update step-size of each spectral channel. In addition, we design a Non-local Hybrid Attention Transformer(NHAT) module for fully leveraging the receptive field advantage of transformer. By plugging the NLHA into the Non-local Information Aggregation (NLIA) module, the unfolding network can achieve better reconstruction results. Experimental results show that our ASPUN is superior to the existing SOTA algorithms and achieves the best performance.

cs.CV↗

arXiv2024

A Review of Transformer-Based Models for Computer Vision Tasks: Capturing Global Context and Spatial Relationships

Transformer-based models have transformed the landscape of natural language processing (NLP) and are increasingly applied to computer vision tasks with remarkable success. These models, renowned for their ability to capture long-range dependencies and contextual information, offer a promising alternative to traditional convolutional neural networks (CNNs) in computer vision. In this review paper, we provide an extensive overview of various transformer architectures adapted for computer vision tasks. We delve into how these models capture global context and spatial relationships in images, empowering them to excel in tasks such as image classification, object detection, and segmentation. Analyzing the key components, training methodologies, and performance metrics of transformer-based models, we highlight their strengths, limitations, and recent advancements. Additionally, we discuss potential research directions and applications of transformer-based models in computer vision, offering insights into their implications for future advancements in the field.

cs.CV↗

arXiv2025

In-Context Learning of Linear Systems: Generalization Theory and Applications to Operator Learning

We study theoretical guarantees for solving linear systems in-context using a linear transformer architecture. For in-domain generalization, we provide neural scaling laws that bound the generalization error in terms of the number of tasks and sizes of samples used in training and inference. For out-of-domain generalization, we find that the behavior of trained transformers under task distribution shifts depends crucially on the distribution of the tasks seen during training. We introduce a novel notion of task diversity and show that it defines a necessary and sufficient condition for pre-trained transformers generalize under task distribution shifts. We also explore applications of learning linear systems in-context, such as to in-context operator learning for PDEs. Finally, we provide some numerical experiments to validate the established theory.

cs.LG↗

arXiv2025

Diversity-Guided MLP Reduction for Efficient Large Vision Transformers

Transformer models achieve excellent scaling property, where the performance is improved with the increment of model capacity. However, large-scale model parameters lead to an unaffordable cost of computing and memory. We analyze popular transformer architectures and find that multilayer perceptron (MLP) modules take up the majority of model parameters. To this end, we focus on the recoverability of the compressed models and propose a Diversity-Guided MLP Reduction (DGMR) method to significantly reduce the parameters of large vision transformers with only negligible performance degradation. Specifically, we conduct a Gram-Schmidt weight pruning strategy to eliminate redundant neurons of MLP hidden layer, while preserving weight diversity for better performance recover during distillation. Compared to the model trained from scratch, our pruned model only requires 0.06\% data of LAION-2B (for the training of large vision transformers) without labels (ImageNet-1K) to recover the original performance. Experimental results on several state-of-the-art large vision transformers demonstrate that our method achieves a more than 57.0\% parameter and FLOPs reduction in a near lossless manner. Notably, for EVA-CLIP-E (4.4B), our method accomplishes a 71.5\% parameter and FLOPs reduction without performance degradation. The source code and trained weights are available at https://github.com/visresearch/DGMR.

cs.CV↗

arXiv2025

Double-Constraint Diffusion Model with Nuclear Regularization for Ultra-low-dose PET Reconstruction

Ultra-low-dose positron emission tomography (PET) reconstruction holds significant potential for reducing patient radiation exposure and shortening examination times. However, it may also lead to increased noise and reduced imaging detail, which could decrease the image quality. In this study, we present a Double-Constraint Diffusion Model (DCDM), which freezes the weights of a pre-trained diffusion model and injects a trainable double-constraint controller into the encoding architecture, greatly reducing the number of trainable parameters for ultra-low-dose PET reconstruction. Unlike full fine-tuning models, DCDM can adapt to different dose levels without retraining all model parameters, thereby improving reconstruction flexibility. Specifically, the two constraint modules, named the Nuclear Transformer Constraint (NTC) and the Encoding Nexus Constraint (ENC), serve to refine the pre-trained diffusion model. The NTC leverages the nuclear norm as an approximation for matrix rank minimization, integrates the low-rank property into the Transformer architecture, and enables efficient information extraction from low-dose images and conversion into compressed feature representations in the latent space. Subsequently, the ENC utilizes these compressed feature representations to encode and control the pre-trained diffusion model, ultimately obtaining reconstructed PET images in the pixel space. In clinical reconstruction, the compressed feature representations from NTC help select the most suitable ENC for efficient unknown low-dose PET reconstruction. Experiments conducted on the UDPET public dataset and the Clinical dataset demonstrated that DCDM outperforms state-of-the-art methods on known dose reduction factors (DRF) and generalizes well to unknown DRF scenarios, proving valuable even at ultra-low dose levels, such as 1% of the full dose.

cs.CV↗

arXiv2026

Patch Hierarchical Attention Transformer for Efficient Particle Jet Tagging

Real-time jet tagging is critical for identifying short-lived particle decays in the high-throughput detectors of the Large Hadron Collider, where real-time trigger systems responsible for deciding which collision events to store impose strict latency and accuracy constraints. While transformer architectures achieve the highest jet tagging accuracy when compute is unconstrained, their quadratic self-attention cost makes inference restrictive on trigger budget. Existing efficient variants reduce the computational cost, but hinder the classification performance. To address this limitation, we introduce the Patch Hierarchical Attention Transformer (PHAT-JeT), which combines two mechanisms: a physics-inspired geometric message-passing module that encodes local detector-plane structure, and a hierarchical patch-based attention scheme that computes exact attention within small particle groups while preserving global context through lightweight patch-token communication. Within a restricted budget, PHAT-JeT achieves state-of-the-art accuracy and background rejection among all resource-constrained jet tagging models on four benchmarks (\textsc{hls4ml}, JetClass, Top Tagging, and Quark--Gluon). Our code is available at https://github.com/aaronw5/PHAT-JeT.

hep-ex↗

arXiv2026

Query Expansion and Key Specialization in Transformer Attention Geometry

The projection of queries and keys are central to the attention mechanism in Transformer architectures. While they are mathematically symmetric, they play different roles in attention mechanisms. The question of whether there is an effect from their functional distinction on their geometric development in training remains unanswered. We investigate the problem through the training of small GPT-like Transformers on character-level WikiText-103 for three different depths (4, 6, and 8 layers), three types of initialization for queries and keys, and four random seeds, resulting in 36 runs and 54 trajectories of average layers across seeds. We track the effective dimensionality of those layers using participation ratios and discover that effective dimension of queries expand while keys shrink, and that $PR_Q - PR_K$ is positive in all trajectories studied. In connection to attention, the shrinking of keys leads to a narrower spectrum of $QK^\top$ and more peaked attention weights. In order to determine if this connection is causal or coincidental, we directly control the spectrum of keys during training across five seeds: restricting it to make it shrink sharpens the attention with high directional confidence, while keeping it constant to the level of initial dispersion makes attention softer. Additional token-level checkpoint analyses show that the monotonic paired-contrast trend is not universal across pretrained families, but survives as an early-training regime that later decays over a full pretraining run, and the link between interaction-rank geometry and attention entropy remains visible in several models.

cs.AI↗

arXiv2024

Vision Transformer Computation and Resilience for Dynamic Inference

State-of-the-art deep learning models for computer vision tasks are based on the transformer architecture and often deployed in real-time applications. In this scenario, the resources available for every inference can vary, so it is useful to be able to dynamically adapt execution to trade accuracy for efficiency. To create dynamic models, we leverage the resilience of vision transformers to pruning and switch between different scaled versions of a model. Surprisingly, we find that most FLOPs are generated by convolutions, not attention. These relative FLOP counts are not a good predictor of GPU performance since GPUs have special optimizations for convolutions. Some models are fairly resilient and their model execution can be adapted without retraining, while all models achieve better accuracy with retraining alternative execution paths. These insights mean that we can leverage CNN accelerators and these alternative execution paths to enable efficient and dynamic vision transformer inference. Our analysis shows that leveraging this type of dynamic execution can lead to saving 28\% of energy with a 1.4\% accuracy drop for SegFormer (63 GFLOPs), with no additional training, and 53\% of energy for ResNet-50 (4 GFLOPs) with a 3.3\% accuracy drop by switching between pretrained Once-For-All models.

cs.CV↗