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 595 records · Page 33Linked to original sources

arXiv2025

Distributed Cross-Channel Hierarchical Aggregation for Foundation Models

Vision-based scientific foundation models hold significant promise for advancing scientific discovery and innovation. This potential stems from their ability to aggregate images from diverse sources such as varying physical groundings or data acquisition systems and to learn spatio-temporal correlations using transformer architectures. However, tokenizing and aggregating images can be compute-intensive, a challenge not fully addressed by current distributed methods. In this work, we introduce the Distributed Cross-Channel Hierarchical Aggregation (D-CHAG) approach designed for datasets with a large number of channels across image modalities. Our method is compatible with any model-parallel strategy and any type of vision transformer architecture, significantly improving computational efficiency. We evaluated D-CHAG on hyperspectral imaging and weather forecasting tasks. When integrated with tensor parallelism and model sharding, our approach achieved up to a 75% reduction in memory usage and more than doubled sustained throughput on up to 1,024 AMD GPUs on the Frontier Supercomputer.

cs.LG↗

arXiv2021

Twins: Revisiting the Design of Spatial Attention in Vision Transformers

Very recently, a variety of vision transformer architectures for dense prediction tasks have been proposed and they show that the design of spatial attention is critical to their success in these tasks. In this work, we revisit the design of the spatial attention and demonstrate that a carefully-devised yet simple spatial attention mechanism performs favourably against the state-of-the-art schemes. As a result, we propose two vision transformer architectures, namely, Twins-PCPVT and Twins-SVT. Our proposed architectures are highly-efficient and easy to implement, only involving matrix multiplications that are highly optimized in modern deep learning frameworks. More importantly, the proposed architectures achieve excellent performance on a wide range of visual tasks, including image level classification as well as dense detection and segmentation. The simplicity and strong performance suggest that our proposed architectures may serve as stronger backbones for many vision tasks. Our code is released at https://github.com/Meituan-AutoML/Twins .

cs.CV↗

arXiv2022

How Much Does Attention Actually Attend? Questioning the Importance of Attention in Pretrained Transformers

The attention mechanism is considered the backbone of the widely-used Transformer architecture. It contextualizes the input by computing input-specific attention matrices. We find that this mechanism, while powerful and elegant, is not as important as typically thought for pretrained language models. We introduce PAPA, a new probing method that replaces the input-dependent attention matrices with constant ones -- the average attention weights over multiple inputs. We use PAPA to analyze several established pretrained Transformers on six downstream tasks. We find that without any input-dependent attention, all models achieve competitive performance -- an average relative drop of only 8% from the probing baseline. Further, little or no performance drop is observed when replacing half of the input-dependent attention matrices with constant (input-independent) ones. Interestingly, we show that better-performing models lose more from applying our method than weaker models, suggesting that the utilization of the input-dependent attention mechanism might be a factor in their success. Our results motivate research on simpler alternatives to input-dependent attention, as well as on methods for better utilization of this mechanism in the Transformer architecture.

cs.CL↗

arXiv2025

Sparse Query Attention (SQA): A Computationally Efficient Attention Mechanism with Query Heads Reduction

The Transformer architecture, underpinned by the Multi-Head Attention (MHA) mechanism, has become the de facto standard for state-of-the-art models in artificial intelligence. However, the quadratic computational complexity of MHA with respect to sequence length presents a significant barrier to scaling, particularly for applications involving long contexts. Prevailing solutions, such as Multi-Query Attention (MQA) and Grouped-Query Attention (GQA), have effectively addressed the memory bandwidth bottleneck that dominates autoregressive inference latency by sharing Key and Value projections. While highly successful, these methods do not reduce the fundamental number of floating-point operations (FLOPs) required for the attention score computation, which remains a critical bottleneck for training and full-sequence processing. This paper introduces Sparse Query Attention (SQA), a novel attention architecture that pursues an alternative and complementary optimization path. Instead of reducing Key/Value heads, SQA reduces the number of Query heads. This architectural modification directly decreases the computational complexity of the attention mechanism by a factor proportional to the reduction in query heads, thereby lowering the overall FLOPs. This work presents the theoretical foundation of SQA, its mathematical formulation, and a family of architectural variants. Empirical benchmarks on long sequences (32k-200k tokens) demonstrate that SQA can achieve significant throughput improvements of up to 3x in computation-bound scenarios such as model pre-training, fine-tuning, and encoder-based tasks, with only a minimal impact on model quality in preliminary smallscale experiments. SQA was discovered serendipitously during the development of the upcoming Reactive Transformer architecture, suggesting its potential as a powerful tool for building more efficient and scalable models

cs.LG↗

arXiv2026

Rethinking Attention Locality in Spiking Transformers

Spiking Transformers provide a promising paradigm for efficient visual processing with spike-driven computation, yet their Softmax-free Spiking Self-Attention (SSA) struggles to establish spatially localized token interactions. Although existing locality-enhanced SSA methods improve accuracy, it remains unclear whether they consistently induce spatial locality across layers and different Spiking Transformer architectures. Through Mean Attention Distance (MAD) analysis, we reveal that computational locality does not necessarily translate into spatial locality and show that uniformly applying the same locality enhancement overlooks architecture-dependent deployment requirements. Motivated by these observations, we propose Spatially Contiguous Local Attention with Boundary Continuity Pathway (SCLA-BCP). SCLA computes attention within non-overlapping regions of spatially adjacent tokens, while BCP facilitates cross-boundary information exchange through a lightweight convolutional pathway. Furthermore, we develop a hierarchical locality deployment strategy to effectively apply SCLA-BCP across the two major Spiking Transformer architectures. Extensive experiments on seven static and neuromorphic datasets covering classification, detection, and segmentation demonstrate consistent improvements with limited parameter and energy overhead. Notably, our approach improves mAP@50 by up to 9.50% on COCO 2017 and mIoU by up to 3.42% on ADE20K. Visualizations, MAD analysis, and ablation studies further validate its effectiveness.

cs.CV↗

arXiv2026

Hybrid Linear Attention Done Right: Efficient Distillation and Effective Architectures for Extremely Long Contexts

Hybrid Transformer architectures, which combine softmax attention blocks and recurrent neural networks (RNNs), have shown a desirable performance-throughput tradeoff for long-context modeling, but their adoption and studies are hindered by the prohibitive cost of large-scale pre-training from scratch. Some recent studies have shown that pre-trained softmax attention blocks can be converted into RNN blocks through parameter transfer and knowledge distillation. However, these transfer methods require substantial amounts of training data (more than 10B tokens), and the resulting hybrid models also exhibit poor long-context performance, which is the scenario where hybrid models enjoy significant inference speedups over Transformer-based models. In this paper, we present HALO (Hybrid Attention via Layer Optimization), a pipeline for distilling Transformer models into RNN-attention hybrid models. We then present HypeNet, a hybrid architecture with superior length generalization enabled by a novel position encoding scheme (named HyPE) and various architectural modifications. We convert the Qwen3 series into HypeNet using HALO, achieving performance comparable to the original Transformer models while enjoying superior long-context performance and efficiency. The conversion requires just 2.3B tokens, less than 0.01% of their pre-training data

cs.CL↗

arXiv2025

Autoregressive Distillation of Diffusion Transformers

Diffusion models with transformer architectures have demonstrated promising capabilities in generating high-fidelity images and scalability for high resolution. However, iterative sampling process required for synthesis is very resource-intensive. A line of work has focused on distilling solutions to probability flow ODEs into few-step student models. Nevertheless, existing methods have been limited by their reliance on the most recent denoised samples as input, rendering them susceptible to exposure bias. To address this limitation, we propose AutoRegressive Distillation (ARD), a novel approach that leverages the historical trajectory of the ODE to predict future steps. ARD offers two key benefits: 1) it mitigates exposure bias by utilizing a predicted historical trajectory that is less susceptible to accumulated errors, and 2) it leverages the previous history of the ODE trajectory as a more effective source of coarse-grained information. ARD modifies the teacher transformer architecture by adding token-wise time embedding to mark each input from the trajectory history and employs a block-wise causal attention mask for training. Furthermore, incorporating historical inputs only in lower transformer layers enhances performance and efficiency. We validate the effectiveness of ARD in a class-conditioned generation on ImageNet and T2I synthesis. Our model achieves a $5\times$ reduction in FID degradation compared to the baseline methods while requiring only 1.1\% extra FLOPs on ImageNet-256. Moreover, ARD reaches FID of 1.84 on ImageNet-256 in merely 4 steps and outperforms the publicly available 1024p text-to-image distilled models in prompt adherence score with a minimal drop in FID compared to the teacher. Project page: https://github.com/alsdudrla10/ARD.

cs.CV↗

arXiv2020

E.T.: Entity-Transformers. Coreference augmented Neural Language Model for richer mention representations via Entity-Transformer blocks

In the last decade, the field of Neural Language Modelling has witnessed enormous changes, with the development of novel models through the use of Transformer architectures. However, even these models struggle to model long sequences due to memory constraints and increasing computational complexity. Coreference annotations over the training data can provide context far beyond the modelling limitations of such language models. In this paper we present an extension over the Transformer-block architecture used in neural language models, specifically in GPT2, in order to incorporate entity annotations during training. Our model, GPT2E, extends the Transformer layers architecture of GPT2 to Entity-Transformers, an architecture designed to handle coreference information when present. To that end, we achieve richer representations for entity mentions, with insignificant training cost. We show the comparative model performance between GPT2 and GPT2E in terms of Perplexity on the CoNLL 2012 and LAMBADA datasets as well as the key differences in the entity representations and their effects in downstream tasks such as Named Entity Recognition. Furthermore, our approach can be adopted by the majority of Transformer-based language models.

cs.CL↗

arXiv2025

Forecasting NCAA Basketball Outcomes with Deep Learning: A Comparative Study of LSTM and Transformer Models

In this research, I explore advanced deep learning methodologies to forecast the outcomes of the 2025 NCAA Division 1 Men's and Women's Basketball tournaments. Leveraging historical NCAA game data, I implement two sophisticated sequence-based models: Long Short-Term Memory (LSTM) and Transformer architectures. The predictive power of these models is augmented through comprehensive feature engineering, including team quality metrics derived from Generalized Linear Models (GLM), Elo ratings, seed differences, and aggregated box-score statistics. To evaluate the robustness and reliability of predictions, I train each model variant using both Binary Cross-Entropy (BCE) and Brier loss functions, providing insights into classification performance and probability calibration. My comparative analysis reveals that while the Transformer architecture optimized with BCE yields superior discriminative power (highest AUC of 0.8473), the LSTM model trained with Brier loss demonstrates superior probabilistic calibration (lowest Brier score of 0.1589). These findings underscore the importance of selecting appropriate model architectures and loss functions based on the specific requirements of forecasting tasks. The detailed analytical pipeline presented here serves as a reproducible framework for future predictive modeling tasks in sports analytics and beyond.

cs.LG↗

arXiv2026

AeroCast: Probabilistic 3D Trajectory Prediction for Non-Cooperative Aerial Obstacles via Transformer-MDN Architecture

Autonomous aerial vehicles operating in shared airspace must predict the future positions of non-cooperative obstacles to plan evasive maneuvers before a collision becomes unavoidable. Unlike cooperative systems that share intent, non-cooperative obstacles such as birds, uncontrolled drones, or debris exhibit multi-modal motion that deterministic predictors cannot adequately represent. Existing methods either rely on recurrent encoders that propagate temporal information sequentially, limiting their ability to capture long-range kinematic precursors of maneuver initiation, or produce point forecasts that provide no distributional information to downstream planners. This paper presents AeroCast, a probabilistic trajectory prediction framework that combines a Transformer encoder with a Mixture Density Network output head to predict per-timestep Gaussian mixture distributions over future three-dimensional displacements. A translation-invariant consecutive displacement encoding and a calibration-oriented training objective address the input design and mode-degeneracy challenges specific to mixture-based aerial trajectory prediction. On a hybrid real-and-synthetic quadrotor corpus spanning nine motion categories, AeroCast reduces Average Displacement Error and Final Displacement Error by approximately 50% relative to the baselines over a five-second horizon, and achieves the lowest negative log-likelihood and Continuous Ranked Probability Score among all compared methods. Ablation analysis identifies velocity input and model capacity as the primary contributors to prediction quality, and positional encoding as essential for long-horizon trajectory coherence. AeroCast inference completes in 0.1ms per sample, compatible with real-time onboard deployment at 100Hz.

cs.RO↗

arXiv2023

SwiFT: Swin 4D fMRI Transformer

Modeling spatiotemporal brain dynamics from high-dimensional data, such as functional Magnetic Resonance Imaging (fMRI), is a formidable task in neuroscience. Existing approaches for fMRI analysis utilize hand-crafted features, but the process of feature extraction risks losing essential information in fMRI scans. To address this challenge, we present SwiFT (Swin 4D fMRI Transformer), a Swin Transformer architecture that can learn brain dynamics directly from fMRI volumes in a memory and computation-efficient manner. SwiFT achieves this by implementing a 4D window multi-head self-attention mechanism and absolute positional embeddings. We evaluate SwiFT using multiple large-scale resting-state fMRI datasets, including the Human Connectome Project (HCP), Adolescent Brain Cognitive Development (ABCD), and UK Biobank (UKB) datasets, to predict sex, age, and cognitive intelligence. Our experimental outcomes reveal that SwiFT consistently outperforms recent state-of-the-art models. Furthermore, by leveraging its end-to-end learning capability, we show that contrastive loss-based self-supervised pre-training of SwiFT can enhance performance on downstream tasks. Additionally, we employ an explainable AI method to identify the brain regions associated with sex classification. To our knowledge, SwiFT is the first Swin Transformer architecture to process dimensional spatiotemporal brain functional data in an end-to-end fashion. Our work holds substantial potential in facilitating scalable learning of functional brain imaging in neuroscience research by reducing the hurdles associated with applying Transformer models to high-dimensional fMRI.

cs.CV↗

arXiv2024

Enhanced Structured State Space Models via Grouped FIR Filtering and Attention Sink Mechanisms

Structured State Space Models (SSMs) have emerged as compelling alternatives to Transformer architectures, offering linear-time complexity and superior performance in various sequence modeling tasks. Despite their advantages, SSMs like the original Mamba-2 face training difficulties due to the sensitivities introduced by the extended series of recurrent matrix multiplications. In this paper, we propose an advanced architecture that mitigates these challenges by decomposing A-multiplications into multiple groups and optimizing positional encoding through Grouped Finite Impulse Response (FIR) filtering. This new structure, denoted as Grouped FIR-enhanced SSM (GFSSM), employs semiseparable matrices for efficient computation. Furthermore, inspired by the "attention sink" phenomenon identified in streaming language models, we incorporate a similar mechanism to enhance the stability and performance of our model over extended sequences. Our approach further bridges the gap between SSMs and Transformer architectures, offering a viable path forward for scalable and high-performing sequence modeling.

cs.CL↗

arXiv2019

Predicting retrosynthetic pathways using a combined linguistic model and hyper-graph exploration strategy

We present an extension of our Molecular Transformer architecture combined with a hyper-graph exploration strategy for automatic retrosynthesis route planning without human intervention. The single-step retrosynthetic model sets a new state of the art for predicting reactants as well as reagents, solvents and catalysts for each retrosynthetic step. We introduce new metrics (coverage, class diversity, round-trip accuracy and Jensen-Shannon divergence) to evaluate the single-step retrosynthetic models, using the forward prediction and a reaction classification model always based on the transformer architecture. The hypergraph is constructed on the fly, and the nodes are filtered and further expanded based on a Bayesian-like probability. We critically assessed the end-to-end framework with several retrosynthesis examples from literature and academic exams. Overall, the frameworks has a very good performance with few weaknesses due to the bias induced during the training process. The use of the newly introduced metrics opens up the possibility to optimize entire retrosynthetic frameworks through focusing on the performance of the single-step model only.

cs.LG↗

arXiv2025

SWIFT:A Scalable lightWeight Infrastructure for Fine-Tuning

Recent development in Large Language Models (LLMs) and Multi-modal Large Language Models (MLLMs) have leverage Attention-based Transformer architectures and achieved superior performance and generalization capabilities. They have since covered extensive areas of traditional learning tasks. For instance, text-based tasks such as text-classification and sequence-labeling, as well as multi-modal tasks like Visual Question Answering (VQA) and Optical Character Recognition (OCR), which were previously addressed using different models, can now be tackled based on one foundation model. Consequently, the training and lightweight fine-tuning of LLMs and MLLMs, especially those based on Transformer architecture, has become particularly important. In recognition of these overwhelming needs, we develop SWIFT, a customizable one-stop infrastructure for large models. With support of over $300+$ LLMs and $50+$ MLLMs, SWIFT stands as the open-source framework that provide the most comprehensive support for fine-tuning large models. In particular, it is the first training framework that provides systematic support for MLLMs. In addition to the core functionalities of fine-tuning, SWIFT also integrates post-training processes such as inference, evaluation, and model quantization, to facilitate fast adoptions of large models in various application scenarios. With a systematic integration of various training techniques, SWIFT offers helpful utilities such as benchmark comparisons among different training techniques for large models. For fine-tuning models specialized in agent framework, we show that notable improvements on the ToolBench leader-board can be achieved by training with customized dataset on SWIFT, with an increase of 5.2%-21.8% in the Act.EM metric over various baseline models, a reduction in hallucination by 1.6%-14.1%, and an average performance improvement of 8%-17%.

cs.CL↗

arXiv2025

Boltzmann Attention Sampling for Image Analysis with Small Objects

Detecting and segmenting small objects, such as lung nodules and tumor lesions, remains a critical challenge in image analysis. These objects often occupy less than 0.1% of an image, making traditional transformer architectures inefficient and prone to performance degradation due to redundant attention computations on irrelevant regions. Existing sparse attention mechanisms rely on rigid hierarchical structures, which are poorly suited for detecting small, variable, and uncertain object locations. In this paper, we propose BoltzFormer, a novel transformer-based architecture designed to address these challenges through dynamic sparse attention. BoltzFormer identifies and focuses attention on relevant areas by modeling uncertainty using a Boltzmann distribution with an annealing schedule. Initially, a higher temperature allows broader area sampling in early layers, when object location uncertainty is greatest. As the temperature decreases in later layers, attention becomes more focused, enhancing efficiency and accuracy. BoltzFormer seamlessly integrates into existing transformer architectures via a modular Boltzmann attention sampling mechanism. Comprehensive evaluations on benchmark datasets demonstrate that BoltzFormer significantly improves segmentation performance for small objects while reducing attention computation by an order of magnitude compared to previous state-of-the-art methods.

cs.CV↗

arXiv2025

How Large Language Models are Designed to Hallucinate

Large language models (LLMs) achieve remarkable fluency across linguistic and reasoning tasks but remain systematically prone to hallucination. Prevailing accounts attribute hallucinations to data gaps, limited context, or optimization errors. We argue instead that hallucination is a structural outcome of the transformer architecture. As coherence engines, transformers are compelled to produce fluent continuations, with self-attention simulating the relational structure of meaning but lacking the existential grounding of temporality, mood, and care that stabilizes human understanding. On this basis, we distinguish ontological hallucination, arising when continuations require disclosure of beings in world, and residual reasoning hallucination, where models mimic inference by recycling traces of human reasoning in text. We illustrate these patterns through case studies aligned with Heideggerian categories and an experiment across twelve LLMs showing how simulated "self-preservation" emerges under extended prompts. Our contribution is threefold: (1) a comparative account showing why existing explanations are insufficient; (2) a predictive taxonomy of hallucination linked to existential structures with proposed benchmarks; and (3) design directions toward "truth-constrained" architectures capable of withholding or deferring when disclosure is absent. We conclude that hallucination is not an incidental defect but a defining limit of transformer-based models, an outcome scaffolding can mask but never resolve.

cs.CY↗

arXiv2026

GeoNorm: Unify Pre-Norm and Post-Norm with Geodesic Optimization

The placement of normalization layers, specifically Pre-Norm and Post-Norm, remains an open question in Transformer architecture design. In this work, we rethink these approaches through the lens of manifold optimization, interpreting the outputs of the Feed-Forward Network (FFN) and attention layers as update directions in optimization. Building on this perspective, we introduce GeoNorm, a novel method that replaces standard normalization with geodesic updates on the manifold. Furthermore, analogous to learning rate schedules, we propose a layer-wise update decay for the FFN and attention components. Comprehensive experiments demonstrate that GeoNorm consistently outperforms existing normalization methods in Transformer models. Crucially, GeoNorm can be seamlessly integrated into standard Transformer architectures, achieving performance improvements with negligible additional computational cost.

cs.LG↗

arXiv2023

What and How does In-Context Learning Learn? Bayesian Model Averaging, Parameterization, and Generalization

In this paper, we conduct a comprehensive study of In-Context Learning (ICL) by addressing several open questions: (a) What type of ICL estimator is learned by large language models? (b) What is a proper performance metric for ICL and what is the error rate? (c) How does the transformer architecture enable ICL? To answer these questions, we adopt a Bayesian view and formulate ICL as a problem of predicting the response corresponding to the current covariate, given a number of examples drawn from a latent variable model. To answer (a), we show that, without updating the neural network parameters, ICL implicitly implements the Bayesian model averaging algorithm, which is proven to be approximately parameterized by the attention mechanism. For (b), we analyze the ICL performance from an online learning perspective and establish a $\mathcal{O}(1/T)$ regret bound for perfectly pretrained ICL, where $T$ is the number of examples in the prompt. To answer (c), we show that, in addition to encoding Bayesian model averaging via attention, the transformer architecture also enables a fine-grained statistical analysis of pretraining under realistic assumptions. In particular, we prove that the error of pretrained model is bounded by a sum of an approximation error and a generalization error, where the former decays to zero exponentially as the depth grows, and the latter decays to zero sublinearly with the number of tokens in the pretraining dataset. Our results provide a unified understanding of the transformer and its ICL ability with bounds on ICL regret, approximation, and generalization, which deepens our knowledge of these essential aspects of modern language models.

stat.ML↗