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 487 records · Page 27Linked to original sources

arXiv2025

DeepKoopFormer: A Koopman Enhanced Transformer Based Architecture for Time Series Forecasting

Time series forecasting plays a vital role across scientific, industrial, and environmental domains, especially when dealing with high-dimensional and nonlinear systems. While Transformer-based models have recently achieved state-of-the-art performance in long-range forecasting, they often suffer from interpretability issues and instability in the presence of noise or dynamical uncertainty. In this work, we propose DeepKoopFormer, a principled forecasting framework that combines the representational power of Transformers with the theoretical rigor of Koopman operator theory. Our model features a modular encoder-propagator-decoder structure, where temporal dynamics are learned via a spectrally constrained, linear Koopman operator in a latent space. We impose structural guarantees-such as bounded spectral radius, Lyapunov based energy regularization, and orthogonal parameterization to ensure stability and interpretability. Comprehensive evaluations are conducted on both synthetic dynamical systems, real-world climate dataset (wind speed and surface pressure), financial time series (cryptocurrency), and electricity generation dataset using the Python package that is prepared for this purpose. Across all experiments, DeepKoopFormer consistently outperforms standard LSTM and baseline Transformer models in terms of accuracy, robustness to noise, and long-term forecasting stability. These results establish DeepKoopFormer as a flexible, interpretable, and robust framework for forecasting in high dimensional and dynamical settings.

cs.LG↗

arXiv2025

Kolmogorov GAM Networks are all you need!

Kolmogorov GAM (K-GAM) networks are shown to be an efficient architecture for training and inference. They are an additive model with an embedding that is independent of the function of interest. They provide an alternative to the transformer architecture. They are the machine learning version of Kolmogorov's Superposition Theorem (KST) which provides an efficient representations of a multivariate function. Such representations have use in machine learning for encoding dictionaries (a.k.a. "look-up" tables). KST theory also provides a representation based on translates of the Köppen function. The goal of our paper is to interpret this representation in a machine learning context for applications in Artificial Intelligence (AI). Our architecture is equivalent to a topological embedding which is independent of the function together with an additive layer that uses a Generalized Additive Model (GAM). This provides a class of learning procedures with far fewer parameters than current deep learning algorithms. Implementation can be parallelizable which makes our algorithms computationally attractive. To illustrate our methodology, we use the Iris data from statistical learning. We also show that our additive model with non-linear embedding provides an alternative to transformer architectures which from a statistical viewpoint are kernel smoothers. Additive KAN models therefore provide a natural alternative to transformers. Finally, we conclude with directions for future research.

cs.LG↗

arXiv2022

A Systematic Review and Replicability Study of BERT4Rec for Sequential Recommendation

BERT4Rec is an effective model for sequential recommendation based on the Transformer architecture. In the original publication, BERT4Rec claimed superiority over other available sequential recommendation approaches (e.g. SASRec), and it is now frequently being used as a state-of-the art baseline for sequential recommendations. However, not all subsequent publications confirmed this result and proposed other models that were shown to outperform BERT4Rec in effectiveness. In this paper we systematically review all publications that compare BERT4Rec with another popular Transformer-based model, namely SASRec, and show that BERT4Rec results are not consistent within these publications. To understand the reasons behind this inconsistency, we analyse the available implementations of BERT4Rec and show that we fail to reproduce results of the original BERT4Rec publication when using their default configuration parameters. However, we are able to replicate the reported results with the original code if training for a much longer amount of time (up to 30x) compared to the default configuration. We also propose our own implementation of BERT4Rec based on the Hugging Face Transformers library, which we demonstrate replicates the originally reported results on 3 out 4 datasets, while requiring up to 95% less training time to converge. Overall, from our systematic review and detailed experiments, we conclude that BERT4Rec does indeed exhibit state-of-the-art effectiveness for sequential recommendation, but only when trained for a sufficient amount of time. Additionally, we show that our implementation can further benefit from adapting other Transformer architectures that are available in the Hugging Face Transformers library (e.g. using disentangled attention, as provided by DeBERTa, or larger hidden layer size cf. ALBERT).

cs.IR↗

arXiv2024

LightM-UNet: Mamba Assists in Lightweight UNet for Medical Image Segmentation

UNet and its variants have been widely used in medical image segmentation. However, these models, especially those based on Transformer architectures, pose challenges due to their large number of parameters and computational loads, making them unsuitable for mobile health applications. Recently, State Space Models (SSMs), exemplified by Mamba, have emerged as competitive alternatives to CNN and Transformer architectures. Building upon this, we employ Mamba as a lightweight substitute for CNN and Transformer within UNet, aiming at tackling challenges stemming from computational resource limitations in real medical settings. To this end, we introduce the Lightweight Mamba UNet (LightM-UNet) that integrates Mamba and UNet in a lightweight framework. Specifically, LightM-UNet leverages the Residual Vision Mamba Layer in a pure Mamba fashion to extract deep semantic features and model long-range spatial dependencies, with linear computational complexity. Extensive experiments conducted on two real-world 2D/3D datasets demonstrate that LightM-UNet surpasses existing state-of-the-art literature. Notably, when compared to the renowned nnU-Net, LightM-UNet achieves superior segmentation performance while drastically reducing parameter and computation costs by 116x and 21x, respectively. This highlights the potential of Mamba in facilitating model lightweighting. Our code implementation is publicly available at https://github.com/MrBlankness/LightM-UNet.

eess.IV↗

arXiv2024

Inference Optimization of Foundation Models on AI Accelerators

Powerful foundation models, including large language models (LLMs), with Transformer architectures have ushered in a new era of Generative AI across various industries. Industry and research community have witnessed a large number of new applications, based on those foundation models. Such applications include question and answer, customer services, image and video generation, and code completions, among others. However, as the number of model parameters reaches to hundreds of billions, their deployment incurs prohibitive inference costs and high latency in real-world scenarios. As a result, the demand for cost-effective and fast inference using AI accelerators is ever more higher. To this end, our tutorial offers a comprehensive discussion on complementary inference optimization techniques using AI accelerators. Beginning with an overview of basic Transformer architectures and deep learning system frameworks, we deep dive into system optimization techniques for fast and memory-efficient attention computations and discuss how they can be implemented efficiently on AI accelerators. Next, we describe architectural elements that are key for fast transformer inference. Finally, we examine various model compression and fast decoding strategies in the same context.

cs.AI↗

arXiv2019

LakhNES: Improving multi-instrumental music generation with cross-domain pre-training

We are interested in the task of generating multi-instrumental music scores. The Transformer architecture has recently shown great promise for the task of piano score generation; here we adapt it to the multi-instrumental setting. Transformers are complex, high-dimensional language models which are capable of capturing long-term structure in sequence data, but require large amounts of data to fit. Their success on piano score generation is partially explained by the large volumes of symbolic data readily available for that domain. We leverage the recently-introduced NES-MDB dataset of four-instrument scores from an early video game sound synthesis chip (the NES), which we find to be well-suited to training with the Transformer architecture. To further improve the performance of our model, we propose a pre-training technique to leverage the information in a large collection of heterogeneous music, namely the Lakh MIDI dataset. Despite differences between the two corpora, we find that this transfer learning procedure improves both quantitative and qualitative performance for our primary task.

cs.SD↗

arXiv2024

Can Transformers Predict Vibrations?

Highly accurate time-series vibration prediction is an important research issue for electric vehicles (EVs). EVs often experience vibrations when driving on rough terrains, known as torsional resonance. This resonance, caused by the interaction between motor and tire vibrations, puts excessive loads on the vehicle's drive shaft. However, current damping technologies only detect resonance after the vibration amplitude of the drive shaft torque reaches a certain threshold, leading to significant loads on the shaft at the time of detection. In this study, we propose a novel approach to address this issue by introducing Resoformer, a transformer-based model for predicting torsional resonance. Resoformer utilizes time-series of the motor rotation speed as input and predicts the amplitude of torsional vibration at a specified quantile occurring in the shaft after the input series. By calculating the attention between recursive and convolutional features extracted from the measured data points, Resoformer improves the accuracy of vibration forecasting. To evaluate the model, we use a vibration dataset called VIBES (Dataset for Forecasting Vibration Transition in EVs), consisting of 2,600 simulator-generated vibration sequences. Our experiments, conducted on strong baselines built on the VIBES dataset, demonstrate that Resoformer achieves state-of-the-art results. In conclusion, our study answers the question "Can Transformers Forecast Vibrations?" While traditional transformer architectures show low performance in forecasting torsional resonance waves, our findings indicate that combining recurrent neural network and temporal convolutional network using the transformer architecture improves the accuracy of long-term vibration forecasting.

cs.LG↗

arXiv2026

AnyGS2Mesh: Feed-Forward Mesh Reconstruction from 3D Gaussian Splatting with Arbitrary-Resolution Views

Existing 3D mesh reconstruction methods from Gaussian scene representations predominantly rely on iterative optimization, resulting in slow inference and limited scalability to high-resolution inputs. In this paper, we present AnyGS2Mesh, the first feed-forward framework for directly reconstructing 3D meshes from 3D Gaussian Splatting representations with support for arbitrary input image resolutions. Our approach incorporates a Gaussian-Guided Transformer architecture that exploits explicit 3D geometric priors for efficient mesh generation. We introduce three key components: (1) a Gaussian-Guided Spatial Reasoning Transformer represents Gaussian primitives as structured 3D tokens and jointly reasons over Gaussian and image features; (2) a Streaming and Patchwise Geometry Encoder processes native-resolution views sequentially and aggregates information across variable-length view sets; (3) a Scale-Aligned Hybrid Depth Refiner uses a PatchFusion-style encoder--decoder to fuse RGB-conditioned predicted depth with Gaussian-rendered metric depth, combining fine local structures with globally consistent metric scale. The refined depth maps are integrated through TSDF fusion, followed by Marching Cubes for deterministic mesh extraction. Extensive experiments show that AnyGS2Mesh achieves state-of-the-art reconstruction quality while significantly reducing inference time compared with optimization-based baselines, enabling near-real-time, high-quality mesh reconstruction. Our results demonstrate the potential of combining Gaussian representations and feed-forward Transformer architectures for scalable 3D geometry reconstruction. The code will be made publicly available upon acceptance.

cs.CG↗

arXiv2026

Retargeting Motions to Diverse Skeletons via Learnable Flattening

Cross-structural motion retargeting aims to transfer motion between different skeletal topologies. Despite recent progress, existing state-of-the-art models struggle with reliability in zero-shot settings, i.e. skeletons with different topologies which were unseen during training, and recent Transformer-based attempts have failed to outperform specialized geometric methods. We bridge this gap with a Transformer Autoencoder that learns a topology- and translation-invariant latent space. Our core contribution is a learnable flattening of skeletal graphs that captures both local dependencies and global structure. Unlike the standard transformer architecture, which adds positional information to token content, we integrate graph-based positional encodings multiplicatively, a design choice that follows directly from our flattening formulation. The resulting model handles diverse skeletal topologies within a single unified architecture and trains in a fully unsupervised manner, requiring no paired retargeting data. Ablation studies show, that the graph encodings, multiplicative formulation, and Transformer backbone is critical for the performance. In zero-shot evaluations, our method reduces global joint position error by $43-47\%$ over current benchmarks. A user study ($n = 37$), including expert animators, further ranks our approach highest in motion alignment and physical plausibility ($p < 0.05$). These results demonstrate that our model design is key to making transformer architectures effective for motion retargeting, outperforming existing approaches.

cs.CV↗

arXiv2026

Formally Verified Autoresearch for Analyzing the Expressivity of Multi-Head Attention

Multi-head attention is a central component of the transformer architecture, yet how its expressivity depends on the number of heads remains poorly understood. We study a one-layer, attention-only transformer with additive positional embeddings and define the head complexity $H^\ast(f)$ of a Boolean function $f$ as the minimum number of attention heads required to represent it. We derive an exact linear-fractional normal form for head complexity, prove a general threshold-degree lower bound, and give an explicit ten-bit function for which this bound is strict. For symmetric Boolean functions, we independently show that head complexity equals the number of times the function changes value between consecutive Hamming-weight levels, even when positional embeddings are allowed. This agrees with the concurrent characterization of Hsu and Xu (arXiv:2608.04243) in a model without positional embeddings. The symmetric lower bound combines our general threshold-degree bound with the classical result of Aspnes et al. (1994). These results were discovered and formalized by a frontier model through conjecture generation, theorem proving, and machine verification, while human researchers asked the high-level questions and reviewed the generated theorems. We make the Lean formalization and research traces publicly available to support agent-assisted theorem proving and formal verification in mechanistic interpretability. This work provides a proof of concept for using the capabilities of frontier models, grounded by formal verification, to develop an understanding of transformer architectures.

cs.CC↗

arXiv2024

A Meta-Learning Perspective on Transformers for Causal Language Modeling

The Transformer architecture has become prominent in developing large causal language models. However, mechanisms to explain its capabilities are not well understood. Focused on the training process, here we establish a meta-learning view of the Transformer architecture when trained for the causal language modeling task, by explicating an inner optimization process within the Transformer. Further, within the inner optimization, we discover and theoretically analyze a special characteristic of the norms of learned token representations within Transformer-based causal language models. Our analysis is supported by experiments in various settings.

cs.LG↗

arXiv2025

GVTNet: Graph Vision Transformer For Face Super-Resolution

Recent advances in face super-resolution research have utilized the Transformer architecture. This method processes the input image into a series of small patches. However, because of the strong correlation between different facial components in facial images. When it comes to super-resolution of low-resolution images, existing algorithms cannot handle the relationships between patches well, resulting in distorted facial components in the super-resolution results. To solve the problem, we propose a transformer architecture based on graph neural networks called graph vision transformer network. We treat each patch as a graph node and establish an adjacency matrix based on the information between patches. In this way, the patch only interacts between neighboring patches, further processing the relationship of facial components. Quantitative and visualization experiments have underscored the superiority of our algorithm over state-of-the-art techniques. Through detailed comparisons, we have demonstrated that our algorithm possesses more advanced super-resolution capabilities, particularly in enhancing facial components. The PyTorch code is available at https://github.com/continueyang/GVTNet

cs.CV↗

arXiv2024

Multi-State-Action Tokenisation in Decision Transformers for Multi-Discrete Action Spaces

Decision Transformers, in their vanilla form, struggle to perform on image-based environments with multi-discrete action spaces. Although enhanced Decision Transformer architectures have been developed to improve performance, these methods have not specifically addressed this problem of multi-discrete action spaces which hampers existing Decision Transformer architectures from learning good representations. To mitigate this, we propose Multi-State Action Tokenisation (M-SAT), an approach for tokenising actions in multi-discrete action spaces that enhances the model's performance in such environments. Our approach involves two key changes: disentangling actions to the individual action level and tokenising the actions with auxiliary state information. These two key changes also improve individual action level interpretability and visibility within the attention layers. We demonstrate the performance gains of M-SAT on challenging ViZDoom environments with multi-discrete action spaces and image-based state spaces, including the Deadly Corridor and My Way Home scenarios, where M-SAT outperforms the baseline Decision Transformer without any additional data or heavy computational overheads. Additionally, we find that removing positional encoding does not adversely affect M-SAT's performance and, in some cases, even improves it.

cs.LG↗

arXiv2022

IRISformer: Dense Vision Transformers for Single-Image Inverse Rendering in Indoor Scenes

Indoor scenes exhibit significant appearance variations due to myriad interactions between arbitrarily diverse object shapes, spatially-changing materials, and complex lighting. Shadows, highlights, and inter-reflections caused by visible and invisible light sources require reasoning about long-range interactions for inverse rendering, which seeks to recover the components of image formation, namely, shape, material, and lighting. In this work, our intuition is that the long-range attention learned by transformer architectures is ideally suited to solve longstanding challenges in single-image inverse rendering. We demonstrate with a specific instantiation of a dense vision transformer, IRISformer, that excels at both single-task and multi-task reasoning required for inverse rendering. Specifically, we propose a transformer architecture to simultaneously estimate depths, normals, spatially-varying albedo, roughness and lighting from a single image of an indoor scene. Our extensive evaluations on benchmark datasets demonstrate state-of-the-art results on each of the above tasks, enabling applications like object insertion and material editing in a single unconstrained real image, with greater photorealism than prior works. Code and data are publicly released at https://github.com/ViLab-UCSD/IRISformer.

cs.CV↗

arXiv2023

To Transformers and Beyond: Large Language Models for the Genome

In the rapidly evolving landscape of genomics, deep learning has emerged as a useful tool for tackling complex computational challenges. This review focuses on the transformative role of Large Language Models (LLMs), which are mostly based on the transformer architecture, in genomics. Building on the foundation of traditional convolutional neural networks and recurrent neural networks, we explore both the strengths and limitations of transformers and other LLMs for genomics. Additionally, we contemplate the future of genomic modeling beyond the transformer architecture based on current trends in research. The paper aims to serve as a guide for computational biologists and computer scientists interested in LLMs for genomic data. We hope the paper can also serve as an educational introduction and discussion for biologists to a fundamental shift in how we will be analyzing genomic data in the future.

q-bio.GN↗

arXiv2024

Symmetric Dot-Product Attention for Efficient Training of BERT Language Models

Initially introduced as a machine translation model, the Transformer architecture has now become the foundation for modern deep learning architecture, with applications in a wide range of fields, from computer vision to natural language processing. Nowadays, to tackle increasingly more complex tasks, Transformer-based models are stretched to enormous sizes, requiring increasingly larger training datasets, and unsustainable amount of compute resources. The ubiquitous nature of the Transformer and its core component, the attention mechanism, are thus prime targets for efficiency research. In this work, we propose an alternative compatibility function for the self-attention mechanism introduced by the Transformer architecture. This compatibility function exploits an overlap in the learned representation of the traditional scaled dot-product attention, leading to a symmetric with pairwise coefficient dot-product attention. When applied to the pre-training of BERT-like models, this new symmetric attention mechanism reaches a score of 79.36 on the GLUE benchmark against 78.74 for the traditional implementation, leads to a reduction of 6% in the number of trainable parameters, and reduces the number of training steps required before convergence by half.

cs.CL↗

arXiv2024

Shifted Window Fourier Transform And Retention For Image Captioning

Image Captioning is an important Language and Vision task that finds application in a variety of contexts, ranging from healthcare to autonomous vehicles. As many real-world applications rely on devices with limited resources, much effort in the field was put into the development of lighter and faster models. However, much of the current optimizations focus on the Transformer architecture in contrast to the existence of more efficient methods. In this work, we introduce SwiFTeR, an architecture almost entirely based on Fourier Transform and Retention, to tackle the main efficiency bottlenecks of current light image captioning models, being the visual backbone's onerosity, and the decoder's quadratic cost. SwiFTeR is made of only 20M parameters, and requires 3.1 GFLOPs for a single forward pass. Additionally, it showcases superior scalability to the caption length and its small memory requirements enable more images to be processed in parallel, compared to the traditional transformer-based architectures. For instance, it can generate 400 captions in one second. Although, for the time being, the caption quality is lower (110.2 CIDEr-D), most of the decrease is not attributed to the architecture but rather an incomplete training practice which currently leaves much room for improvements. Overall, SwiFTeR points toward a promising direction to new efficient architectural design. The implementation code will be released in the future.

cs.CV↗

arXiv2026

Translational Gaps in Graph Transformers for Longitudinal EHR Prediction: A Critical Appraisal of GT-BEHRT

Transformer-based models have improved predictive modeling on longitudinal electronic health records through large-scale self-supervised pretraining. However, most EHR transformer architectures treat each clinical encounter as an unordered collection of codes, which limits their ability to capture meaningful relationships within a visit. Graph-transformer approaches aim to address this limitation by modeling visit-level structure while retaining the ability to learn long-term temporal patterns. This paper provides a critical review of GT-BEHRT, a graph-transformer architecture evaluated on MIMIC-IV intensive care outcomes and heart failure prediction in the All of Us Research Program. We examine whether the reported performance gains reflect genuine architectural benefits and whether the evaluation methodology supports claims of robustness and clinical relevance. We analyze GT-BEHRT across seven dimensions relevant to modern machine learning systems, including representation design, pretraining strategy, cohort construction transparency, evaluation beyond discrimination, fairness assessment, reproducibility, and deployment feasibility. GT-BEHRT reports strong discrimination for heart failure prediction within 365 days, with AUROC 94.37 +/- 0.20, AUPRC 73.96 +/- 0.83, and F1 64.70 +/- 0.85. Despite these results, we identify several important gaps, including the lack of calibration analysis, incomplete fairness evaluation, sensitivity to cohort selection, limited analysis across phenotypes and prediction horizons, and limited discussion of practical deployment considerations. Overall, GT-BEHRT represents a meaningful architectural advance in EHR representation learning, but more rigorous evaluation focused on calibration, fairness, and deployment is needed before such models can reliably support clinical decision-making.

cs.LG↗