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 1,009 records · Page 56Linked to original sources

arXiv2025

Sonnet: Spectral Operator Neural Network for Multivariable Time Series Forecasting

Multivariable time series forecasting methods can integrate information from exogenous variables, leading to significant prediction accuracy gains. The transformer architecture has been widely applied in various time series forecasting models due to its ability to capture long-range sequential dependencies. However, a naïve application of transformers often struggles to effectively model complex relationships among variables over time. To mitigate against this, we propose a novel architecture, termed Spectral Operator Neural Network (Sonnet). Sonnet applies learnable wavelet transformations to the input and incorporates spectral analysis using the Koopman operator. Its predictive skill relies on the Multivariable Coherence Attention (MVCA), an operation that leverages spectral coherence to model variable dependencies. Our empirical analysis shows that Sonnet yields the best performance on $34$ out of $47$ forecasting tasks with an average mean absolute error (MAE) reduction of $2.2\%$ against the most competitive baseline. We further show that MVCA can remedy the deficiencies of naïve attention in various deep learning models, reducing MAE by $10.7\%$ on average in the most challenging forecasting tasks.

cs.LG↗

arXiv2025

Bridging the Dimensional Chasm: Uncover Layer-wise Dimensional Reduction in Transformers through Token Correlation

The geometric evolution of token representations in large language models (LLMs) presents a fundamental paradox: while human language inherently organizes semantic information in low-dimensional spaces ($\sim 10^1$ dimensions), modern LLMs employ high-dimensional embeddings ($\sim 10^3$ dimensions) processed through Transformer architectures. To resolve this paradox, this work bridges this conceptual gap by developing a geometric framework that tracks token dynamics across Transformers layers. Through layer-wise analysis of intrinsic dimensions across multiple architectures, we reveal an expansion-contraction pattern where tokens diffuse to a "working space" and then progressively project onto lower-dimensional submanifolds. Our finding implies a negative correlation between the working space dimension and parameter-sensitive performance of the LLMs, and indicates that effective models tend to compress tokens into approximately 10-dimensional submanifolds, closely resembling human semantic spaces. This work not only advances LLM interpretability by reframing Transformers layers as projectors that mediate between high-dimensional computation and low-dimensional semantics, but also provides practical tools for model diagnostics that do not rely on task-specific evaluations.

cs.CL↗

arXiv2026

Towards Understanding What State Space Models Learn About Code

State Space Models (SSMs) have emerged as an efficient alternative to the Transformer architecture. Prior work shows that, when trained under comparable conditions, SSMs can match or surpass Transformers on code understanding tasks. However, their internal mechanisms remain a black box. We present the first systematic analysis of what SSM-based code models learn along with the direct comparison between SSM and Transformer models in this domain. Our analysis shows that SSMs capture syntactic and semantic structure more effectively than Transformers during pretraining but forgets certain relations during fine-tuning on some tasks. To investigate this behavior, we introduce SSM-Interpret, a frequency-domain framework that exposes a spectral shift toward short-range dependencies during fine-tuning. Guided by these findings, we propose architectural modifications that significantly improve the performance of SSM-based code model by upto +6 MRR on NLCodeSearch. This demonstrates that our analysis not only explains model behavior but also leads directly to better designs.

cs.AI↗

arXiv2019

Character-based NMT with Transformer

Character-based translation has several appealing advantages, but its performance is in general worse than a carefully tuned BPE baseline. In this paper we study the impact of character-based input and output with the Transformer architecture. In particular, our experiments on EN-DE show that character-based Transformer models are more robust than their BPE counterpart, both when translating noisy text, and when translating text from a different domain. To obtain comparable BLEU scores in clean, in-domain data and close the gap with BPE-based models we use known techniques to train deeper Transformer models.

cs.CL↗

arXiv2020

The Chess Transformer: Mastering Play using Generative Language Models

This work demonstrates that natural language transformers can support more generic strategic modeling, particularly for text-archived games. In addition to learning natural language skills, the abstract transformer architecture can generate meaningful moves on a chessboard. With further fine-tuning, the transformer learns complex gameplay by training on 2.8 million chess games in Portable Game Notation. After 30,000 training steps, OpenAI's Generative Pre-trained Transformer (GPT-2) optimizes weights for 774 million parameters. This fine-tuned Chess Transformer generates plausible strategies and displays game formations identifiable as classic openings, such as English or the Slav Exchange. Finally, in live play, the novel model demonstrates a human-to-transformer interface that correctly filters illegal moves and provides a novel method to challenge the transformer's chess strategies. We anticipate future work will build on this transformer's promise, particularly in other strategy games where features can capture the underlying complex rule syntax from simple but expressive player annotations.

cs.AI↗

arXiv2022

SwinTrack: A Simple and Strong Baseline for Transformer Tracking

Recently Transformer has been largely explored in tracking and shown state-of-the-art (SOTA) performance. However, existing efforts mainly focus on fusing and enhancing features generated by convolutional neural networks (CNNs). The potential of Transformer in representation learning remains under-explored. In this paper, we aim to further unleash the power of Transformer by proposing a simple yet efficient fully-attentional tracker, dubbed SwinTrack, within classic Siamese framework. In particular, both representation learning and feature fusion in SwinTrack leverage the Transformer architecture, enabling better feature interactions for tracking than pure CNN or hybrid CNN-Transformer frameworks. Besides, to further enhance robustness, we present a novel motion token that embeds historical target trajectory to improve tracking by providing temporal context. Our motion token is lightweight with negligible computation but brings clear gains. In our thorough experiments, SwinTrack exceeds existing approaches on multiple benchmarks. Particularly, on the challenging LaSOT, SwinTrack sets a new record with 0.713 SUC score. It also achieves SOTA results on other benchmarks. We expect SwinTrack to serve as a solid baseline for Transformer tracking and facilitate future research. Our codes and results are released at https://github.com/LitingLin/SwinTrack.

cs.CV↗

arXiv2022

VAQF: Fully Automatic Software-Hardware Co-Design Framework for Low-Bit Vision Transformer

The transformer architectures with attention mechanisms have obtained success in Nature Language Processing (NLP), and Vision Transformers (ViTs) have recently extended the application domains to various vision tasks. While achieving high performance, ViTs suffer from large model size and high computation complexity that hinders the deployment of them on edge devices. To achieve high throughput on hardware and preserve the model accuracy simultaneously, we propose VAQF, a framework that builds inference accelerators on FPGA platforms for quantized ViTs with binary weights and low-precision activations. Given the model structure and the desired frame rate, VAQF will automatically output the required quantization precision for activations as well as the optimized parameter settings of the accelerator that fulfill the hardware requirements. The implementations are developed with Vivado High-Level Synthesis (HLS) on the Xilinx ZCU102 FPGA board, and the evaluation results with the DeiT-base model indicate that a frame rate requirement of 24 frames per second (FPS) is satisfied with 8-bit activation quantization, and a target of 30 FPS is met with 6-bit activation quantization. To the best of our knowledge, this is the first time quantization has been incorporated into ViT acceleration on FPGAs with the help of a fully automatic framework to guide the quantization strategy on the software side and the accelerator implementations on the hardware side given the target frame rate. Very small compilation time cost is incurred compared with quantization training, and the generated accelerators show the capability of achieving real-time execution for state-of-the-art ViT models on FPGAs.

cs.LG↗

arXiv2022

A Stronger Baseline For Automatic Pfirrmann Grading Of Lumbar Spine MRI Using Deep Learning

This paper addresses the challenge of grading visual features in lumbar spine MRI using Deep Learning. Such a method is essential for the automatic quantification of structural changes in the spine, which is valuable for understanding low back pain. Multiple recent studies investigated different architecture designs, and the most recent success has been attributed to the use of transformer architectures. In this work, we argue that with a well-tuned three-stage pipeline comprising semantic segmentation, localization, and classification, convolutional networks outperform the state-of-the-art approaches. We conducted an ablation study of the existing methods in a population cohort, and report performance generalization across various subgroups. Our code is publicly available to advance research on disc degeneration and low back pain.

eess.IV↗

arXiv2023

From Characters to Words: Hierarchical Pre-trained Language Model for Open-vocabulary Language Understanding

Current state-of-the-art models for natural language understanding require a preprocessing step to convert raw text into discrete tokens. This process known as tokenization relies on a pre-built vocabulary of words or sub-word morphemes. This fixed vocabulary limits the model's robustness to spelling errors and its capacity to adapt to new domains. In this work, we introduce a novel open-vocabulary language model that adopts a hierarchical two-level approach: one at the word level and another at the sequence level. Concretely, we design an intra-word module that uses a shallow Transformer architecture to learn word representations from their characters, and a deep inter-word Transformer module that contextualizes each word representation by attending to the entire word sequence. Our model thus directly operates on character sequences with explicit awareness of word boundaries, but without biased sub-word or word-level vocabulary. Experiments on various downstream tasks show that our method outperforms strong baselines. We also demonstrate that our hierarchical model is robust to textual corruption and domain shift.

cs.CL↗

arXiv2023

Cross-Spatial Pixel Integration and Cross-Stage Feature Fusion Based Transformer Network for Remote Sensing Image Super-Resolution

Remote sensing image super-resolution (RSISR) plays a vital role in enhancing spatial detials and improving the quality of satellite imagery. Recently, Transformer-based models have shown competitive performance in RSISR. To mitigate the quadratic computational complexity resulting from global self-attention, various methods constrain attention to a local window, enhancing its efficiency. Consequently, the receptive fields in a single attention layer are inadequate, leading to insufficient context modeling. Furthermore, while most transform-based approaches reuse shallow features through skip connections, relying solely on these connections treats shallow and deep features equally, impeding the model's ability to characterize them. To address these issues, we propose a novel transformer architecture called Cross-Spatial Pixel Integration and Cross-Stage Feature Fusion Based Transformer Network (SPIFFNet) for RSISR. Our proposed model effectively enhances global cognition and understanding of the entire image, facilitating efficient integration of features cross-stages. The model incorporates cross-spatial pixel integration attention (CSPIA) to introduce contextual information into a local window, while cross-stage feature fusion attention (CSFFA) adaptively fuses features from the previous stage to improve feature expression in line with the requirements of the current stage. We conducted comprehensive experiments on multiple benchmark datasets, demonstrating the superior performance of our proposed SPIFFNet in terms of both quantitative metrics and visual quality when compared to state-of-the-art methods.

cs.CV↗

arXiv2024

AM-SORT: Adaptable Motion Predictor with Historical Trajectory Embedding for Multi-Object Tracking

Many multi-object tracking (MOT) approaches, which employ the Kalman Filter as a motion predictor, assume constant velocity and Gaussian-distributed filtering noises. These assumptions render the Kalman Filter-based trackers effective in linear motion scenarios. However, these linear assumptions serve as a key limitation when estimating future object locations within scenarios involving non-linear motion and occlusions. To address this issue, we propose a motion-based MOT approach with an adaptable motion predictor, called AM-SORT, which adapts to estimate non-linear uncertainties. AM-SORT is a novel extension of the SORT-series trackers that supersedes the Kalman Filter with the transformer architecture as a motion predictor. We introduce a historical trajectory embedding that empowers the transformer to extract spatio-temporal features from a sequence of bounding boxes. AM-SORT achieves competitive performance compared to state-of-the-art trackers on DanceTrack, with 56.3 IDF1 and 55.6 HOTA. We conduct extensive experiments to demonstrate the effectiveness of our method in predicting non-linear movement under occlusions.

cs.CV↗

arXiv2024

DenseMamba: State Space Models with Dense Hidden Connection for Efficient Large Language Models

Large language models (LLMs) face a daunting challenge due to the excessive computational and memory requirements of the commonly used Transformer architecture. While state space model (SSM) is a new type of foundational network architecture offering lower computational complexity, their performance has yet to fully rival that of Transformers. This paper introduces DenseSSM, a novel approach to enhance the flow of hidden information between layers in SSMs. By selectively integrating shallowlayer hidden states into deeper layers, DenseSSM retains fine-grained information crucial for the final output. Dense connections enhanced DenseSSM still maintains the training parallelizability and inference efficiency. The proposed method can be widely applicable to various SSM types like RetNet and Mamba. With similar model size, DenseSSM achieves significant improvements, exemplified by DenseRetNet outperforming the original RetNet with up to 5% accuracy improvement on public benchmarks. code is avalaible at https://github.com/WailordHe/DenseSSM

cs.CL↗

arXiv2025

Unsupervised Meta-Learning via In-Context Learning

Unsupervised meta-learning aims to learn feature representations from unsupervised datasets that can transfer to downstream tasks with limited labeled data. In this paper, we propose a novel approach to unsupervised meta-learning that leverages the generalization abilities of in-context learning observed in transformer architectures. Our method reframes meta-learning as a sequence modeling problem, enabling the transformer encoder to learn task context from support images and utilize it to predict query images. At the core of our approach lies the creation of diverse tasks generated using a combination of data augmentations and a mixing strategy that challenges the model during training while fostering generalization to unseen tasks at test time. Experimental results on benchmark datasets showcase the superiority of our approach over existing unsupervised meta-learning baselines, establishing it as the new state-of-the-art. Remarkably, our method achieves competitive results with supervised and self-supervised approaches, underscoring its efficacy in leveraging generalization over memorization.

cs.LG↗

arXiv2024

HGTDP-DTA: Hybrid Graph-Transformer with Dynamic Prompt for Drug-Target Binding Affinity Prediction

Drug target binding affinity (DTA) is a key criterion for drug screening. Existing experimental methods are time-consuming and rely on limited structural and domain information. While learning-based methods can model sequence and structural information, they struggle to integrate contextual data and often lack comprehensive modeling of drug-target interactions. In this study, we propose a novel DTA prediction method, termed HGTDP-DTA, which utilizes dynamic prompts within a hybrid Graph-Transformer framework. Our method generates context-specific prompts for each drug-target pair, enhancing the model's ability to capture unique interactions. The introduction of prompt tuning further optimizes the prediction process by filtering out irrelevant noise and emphasizing task-relevant information, dynamically adjusting the input features of the molecular graph. The proposed hybrid Graph-Transformer architecture combines structural information from Graph Convolutional Networks (GCNs) with sequence information captured by Transformers, facilitating the interaction between global and local information. Additionally, we adopted the multi-view feature fusion method to project molecular graph views and affinity subgraph views into a common feature space, effectively combining structural and contextual information. Experiments on two widely used public datasets, Davis and KIBA, show that HGTDP-DTA outperforms state-of-the-art DTA prediction methods in both prediction performance and generalization ability.

cs.LG↗

arXiv2024

PASTA: Controllable Part-Aware Shape Generation with Autoregressive Transformers

The increased demand for tools that automate the 3D content creation process led to tremendous progress in deep generative models that can generate diverse 3D objects of high fidelity. In this paper, we present PASTA, an autoregressive transformer architecture for generating high quality 3D shapes. PASTA comprises two main components: An autoregressive transformer that generates objects as a sequence of cuboidal primitives and a blending network, implemented with a transformer decoder that composes the sequences of cuboids and synthesizes high quality meshes for each object. Our model is trained in two stages: First we train our autoregressive generative model using only annotated cuboidal parts as supervision and next, we train our blending network using explicit 3D supervision, in the form of watertight meshes. Evaluations on various ShapeNet objects showcase the ability of our model to perform shape generation from diverse inputs \eg from scratch, from a partial object, from text and images, as well size-guided generation, by explicitly conditioning on a bounding box that defines the object's boundaries. Moreover, as our model considers the underlying part-based structure of a 3D object, we are able to select a specific part and produce shapes with meaningful variations of this part. As evidenced by our experiments, our model generates 3D shapes that are both more realistic and diverse than existing part-based and non part-based methods, while at the same time is simpler to implement and train.

cs.CV↗

arXiv2025

Evaluating and Explaining Earthquake-Induced Liquefaction Potential through Multi-Modal Transformers

This study presents an explainable parallel transformer architecture for soil liquefaction prediction that integrates three distinct data streams: spectral seismic encoding, soil stratigraphy tokenization, and site-specific features. The architecture processes data from 165 case histories across 11 major earthquakes, employing Fast Fourier Transform for seismic waveform encoding and principles from large language models for soil layer tokenization. Interpretability is achieved through SHapley Additive exPlanations (SHAP), which decompose predictions into individual contributions from seismic characteristics, soil properties, and site conditions. The model achieves 93.75% prediction accuracy on cross-regional validation sets and demonstrates robust performance through sensitivity analysis of ground motion intensity and soil resistance parameters. Notably, validation against previously unseen ground motion data from the 2024 Noto Peninsula earthquake confirms the model's generalization capabilities and practical utility. Implementation as a publicly accessible web application enables rapid assessment of multiple sites simultaneously. This approach establishes a new framework in geotechnical deep learning where sophisticated multi-modal analysis meets practical engineering requirements through quantitative interpretation and accessible deployment.

cs.LG↗

arXiv2026

Interpretable Vision Transformers in Monocular Depth Estimation via SVDA

Monocular depth estimation is a central problem in computer vision with applications in robotics, AR, and autonomous driving, yet the self-attention mechanisms that drive modern Transformer architectures remain opaque. We introduce SVD-Inspired Attention (SVDA) into the Dense Prediction Transformer (DPT), providing the first spectrally structured formulation of attention for dense prediction tasks. SVDA decouples directional alignment from spectral modulation by embedding a learnable diagonal matrix into normalized query-key interactions, enabling attention maps that are intrinsically interpretable rather than post-hoc approximations. Experiments on KITTI and NYU-v2 show that SVDA preserves or slightly improves predictive accuracy while adding only minor computational overhead. More importantly, SVDA unlocks six spectral indicators that quantify entropy, rank, sparsity, alignment, selectivity, and robustness. These reveal consistent cross-dataset and depth-wise patterns in how attention organizes during training, insights that remain inaccessible in standard Transformers. By shifting the role of attention from opaque mechanism to quantifiable descriptor, SVDA redefines interpretability in monocular depth estimation and opens a principled avenue toward transparent dense prediction models.

cs.CV↗

arXiv2026

ChWDTA: Channel-wise Wavelet-Domain Transformer Attention and Entropy Modeling for Learned Image Compression

State-of-the-art learned image compression (LIC) schemes are increasingly based on hybrid CNN-transformer architectures. To further improve rate-distortion performance, we introduce channel-wise wavelet transforms into both the transformer and entropy-coding components. First, we propose a channel-wise wavelet-domain transformer attention (ChWDTA) mechanism. ChWDTA keeps the efficient windowed spatial self-attention used in modern LIC backbones, but computes the Q/K/V projections on channel-wise wavelet-transformed features before mapping the attention output back with the inverse transform. The resulting Channel-wise Wavelet-Domain Transformer Block (ChWDTB) therefore preserves the spatial tokenization pattern of windowed attention while sparsifying the channel covariance seen by the attention projections. Second, in the entropy-coding stage, we introduce a channel-wise wavelet packet (ChWP) decomposition that produces four equal-sized subbands, which better fit channel-wise slice-based autoregressive entropy modeling. When each channel-wise subband is divided into two slices, we use eight slices for entropy coding. With this configuration, the proposed scheme obtains BD-rate reductions of -17.82%, -19.15%, and -22.56% on the Kodak, CLIC Professional Validation, and Tecnick test sets, respectively. Even when each channel-wise subband is coded as a single slice, the scheme still retains most of the coding gains with lower complexity. The results confirm the advantage of introducing wavelet transform in CNN-transformer-based LIC schemes.

eess.IV↗