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 1,675 records · Page 93Linked to original sources

arXiv2022

Fourier Image Transformer

Transformer architectures show spectacular performance on NLP tasks and have recently also been used for tasks such as image completion or image classification. Here we propose to use a sequential image representation, where each prefix of the complete sequence describes the whole image at reduced resolution. Using such Fourier Domain Encodings (FDEs), an auto-regressive image completion task is equivalent to predicting a higher resolution output given a low-resolution input. Additionally, we show that an encoder-decoder setup can be used to query arbitrary Fourier coefficients given a set of Fourier domain observations. We demonstrate the practicality of this approach in the context of computed tomography (CT) image reconstruction. In summary, we show that Fourier Image Transformer (FIT) can be used to solve relevant image analysis tasks in Fourier space, a domain inherently inaccessible to convolutional architectures.

cs.CV↗

arXiv2022

Structure-Aware Transformer for Graph Representation Learning

The Transformer architecture has gained growing attention in graph representation learning recently, as it naturally overcomes several limitations of graph neural networks (GNNs) by avoiding their strict structural inductive biases and instead only encoding the graph structure via positional encoding. Here, we show that the node representations generated by the Transformer with positional encoding do not necessarily capture structural similarity between them. To address this issue, we propose the Structure-Aware Transformer, a class of simple and flexible graph Transformers built upon a new self-attention mechanism. This new self-attention incorporates structural information into the original self-attention by extracting a subgraph representation rooted at each node before computing the attention. We propose several methods for automatically generating the subgraph representation and show theoretically that the resulting representations are at least as expressive as the subgraph representations. Empirically, our method achieves state-of-the-art performance on five graph prediction benchmarks. Our structure-aware framework can leverage any existing GNN to extract the subgraph representation, and we show that it systematically improves performance relative to the base GNN model, successfully combining the advantages of GNNs and Transformers. Our code is available at https://github.com/BorgwardtLab/SAT.

stat.ML↗

arXiv2022

A Deep Learning Approach Using Masked Image Modeling for Reconstruction of Undersampled K-spaces

Magnetic Resonance Imaging (MRI) scans are time consuming and precarious, since the patients remain still in a confined space for extended periods of time. To reduce scanning time, some experts have experimented with undersampled k spaces, trying to use deep learning to predict the fully sampled result. These studies report that as many as 20 to 30 minutes could be saved off a scan that takes an hour or more. However, none of these studies have explored the possibility of using masked image modeling (MIM) to predict the missing parts of MRI k spaces. This study makes use of 11161 reconstructed MRI and k spaces of knee MRI images from Facebook's fastmri dataset. This tests a modified version of an existing model using baseline shifted window (Swin) and vision transformer architectures that makes use of MIM on undersampled k spaces to predict the full k space and consequently the full MRI image. Modifications were made using pytorch and numpy libraries, and were published to a github repository. After the model reconstructed the k space images, the basic Fourier transform was applied to determine the actual MRI image. Once the model reached a steady state, experimentation with hyperparameters helped to achieve pinpoint accuracy for the reconstructed images. The model was evaluated through L1 loss, gradient normalization, and structural similarity values. The model produced reconstructed images with L1 loss values averaging to <0.01 and gradient normalization values <0.1 after training finished. The reconstructed k spaces yielded structural similarity values of over 99% for both training and validation with the fully sampled k spaces, while validation loss continually decreased under 0.01. These data strongly support the idea that the algorithm works for MRI reconstruction, as they indicate the model's reconstructed image aligns extremely well with the original, fully sampled k space.

eess.IV↗

arXiv2023

Partial Tensorized Transformers for Natural Language Processing

The transformer architecture has revolutionized Natural Language Processing (NLP) and other machine-learning tasks, due to its unprecedented accuracy. However, their extensive memory and parameter requirements often hinder their practical applications. In this work, we study the effect of tensor-train decomposition to improve the accuracy and compress transformer vision-language neural networks, namely BERT and ViT. We focus both on embedding-layer compression and partial tensorization of neural networks (PTNN) through an algorithmic approach. Our novel PTNN approach significantly improves the accuracy of existing models by up to 5%, all without the need for post-training adjustments, breaking new ground in the field of tensor decomposition.

cs.CL↗

arXiv2024

Powerformer: A Section-adaptive Transformer for Power Flow Adjustment

In this paper, we present a novel transformer architecture tailored for learning robust power system state representations, which strives to optimize power dispatch for the power flow adjustment across different transmission sections. Specifically, our proposed approach, named Powerformer, develops a dedicated section-adaptive attention mechanism, separating itself from the self-attention used in conventional transformers. This mechanism effectively integrates power system states with transmission section information, which facilitates the development of robust state representations. Furthermore, by considering the graph topology of power system and the electrical attributes of bus nodes, we introduce two customized strategies to further enhance the expressiveness: graph neural network propagation and multi-factor attention mechanism. Extensive evaluations are conducted on three power system scenarios, including the IEEE 118-bus system, a realistic 300-bus system in China, and a large-scale European system with 9241 buses, where Powerformer demonstrates its superior performance over several baseline methods.

cs.LG↗

arXiv2024

Breaking Symmetry When Training Transformers

As we show in this paper, the prediction for output token $n+1$ of Transformer architectures without one of the mechanisms of positional encodings and causal attention is invariant to permutations of input tokens $1, 2, ..., n-1$. Usually, both mechanisms are employed and the symmetry with respect to the input tokens is broken. Recently, it has been shown that one can train Transformers without positional encodings. This must be enabled by the causal attention mechanism. In this paper, we elaborate on the argument that the causal connection mechanism must be responsible for the fact that Transformers are able to model input sequences where the order is important. Vertical "slices" of Transformers are all encouraged to represent the same location $k$ in the input sequence. We hypothesize that residual connections contribute to this phenomenon, and demonstrate evidence for this.

cs.LG↗

arXiv2024

Preventing Catastrophic Forgetting through Memory Networks in Continuous Detection

Modern pre-trained architectures struggle to retain previous information while undergoing continuous fine-tuning on new tasks. Despite notable progress in continual classification, systems designed for complex vision tasks such as detection or segmentation still struggle to attain satisfactory performance. In this work, we introduce a memory-based detection transformer architecture to adapt a pre-trained DETR-style detector to new tasks while preserving knowledge from previous tasks. We propose a novel localized query function for efficient information retrieval from memory units, aiming to minimize forgetting. Furthermore, we identify a fundamental challenge in continual detection referred to as background relegation. This arises when object categories from earlier tasks reappear in future tasks, potentially without labels, leading them to be implicitly treated as background. This is an inevitable issue in continual detection or segmentation. The introduced continual optimization technique effectively tackles this challenge. Finally, we assess the performance of our proposed system on continual detection benchmarks and demonstrate that our approach surpasses the performance of existing state-of-the-art resulting in 5-7% improvements on MS-COCO and PASCAL-VOC on the task of continual detection.

cs.CV↗

arXiv2024

Graph as Point Set

Graph is a fundamental data structure to model interconnections between entities. Set, on the contrary, stores independent elements. To learn graph representations, current Graph Neural Networks (GNNs) primarily use message passing to encode the interconnections. In contrast, this paper introduces a novel graph-to-set conversion method that bijectively transforms interconnected nodes into a set of independent points and then uses a set encoder to learn the graph representation. This conversion method holds dual significance. Firstly, it enables using set encoders to learn from graphs, thereby significantly expanding the design space of GNNs. Secondly, for Transformer, a specific set encoder, we provide a novel and principled approach to inject graph information losslessly, different from all the heuristic structural/positional encoding methods adopted in previous graph transformers. To demonstrate the effectiveness of our approach, we introduce Point Set Transformer (PST), a transformer architecture that accepts a point set converted from a graph as input. Theoretically, PST exhibits superior expressivity for both short-range substructure counting and long-range shortest path distance tasks compared to existing GNNs. Extensive experiments further validate PST's outstanding real-world performance. Besides Transformer, we also devise a Deepset-based set encoder, which achieves performance comparable to representative GNNs, affirming the versatility of our graph-to-set method.

cs.LG↗

arXiv2025

DistrAttention: An Efficient and Flexible Self-Attention Mechanism on Modern GPUs

The Transformer architecture has revolutionized deep learning, delivering the state-of-the-art performance in areas such as natural language processing, computer vision, and time series prediction. However, its core component, self-attention, has the quadratic time complexity relative to input sequence length, which hinders the scalability of Transformers. The exsiting approaches on optimizing self-attention either discard full-contextual information or lack of flexibility. In this work, we design DistrAttention, an effcient and flexible self-attention mechanism with the full context. DistrAttention achieves this by grouping data on the embedding dimensionality, usually referred to as $d$. We realize DistrAttention with a lightweight sampling and fusion method that exploits locality-sensitive hashing to group similar data. A block-wise grouping framework is further designed to limit the errors introduced by locality sensitive hashing. By optimizing the selection of block sizes, DistrAttention could be easily integrated with FlashAttention-2, gaining high-performance on modern GPUs. We evaluate DistrAttention with extensive experiments. The results show that our method is 37% faster than FlashAttention-2 on calculating self-attention. In ViT inference, DistrAttention is the fastest and the most accurate among approximate self-attention mechanisms. In Llama3-1B, DistrAttention still achieves the lowest inference time with only 1% accuray loss.

cs.LG↗

arXiv2025

GPU Memory Requirement Prediction for Deep Learning Task Based on Bidirectional Gated Recurrent Unit Optimization Transformer

In response to the increasingly critical demand for accurate prediction of GPU memory resources in deep learning tasks, this paper deeply analyzes the current research status and innovatively proposes a deep learning model that integrates bidirectional gated recurrent units (BiGRU) to optimize the Transformer architecture, aiming to improve the accuracy of memory demand prediction. To verify the effectiveness of the model, a carefully designed comparative experiment was conducted, selecting four representative basic machine learning models: decision tree, random forest, Adaboost, and XGBoost as benchmarks. The detailed experimental results show that the BiGRU Transformer optimization model proposed in this paper exhibits significant advantages in key evaluation indicators: in terms of mean square error (MSE) and root mean square error (RMSE), the model achieves the lowest value among all comparison models, and its predicted results have the smallest deviation from the actual values; In terms of mean absolute error (MAE) and coefficient of determination (R2) indicators, the model also performs well and the results are balanced and stable, with comprehensive predictive performance far exceeding the benchmark machine learning methods compared. In summary, the Transformer model based on bidirectional gated recurrent unit optimization successfully constructed in this study can efficiently and accurately complete GPU memory demand prediction tasks in deep learning tasks, and its prediction accuracy has been significantly improved compared to traditional machine learning methods. This research provides strong technical support and reliable theoretical basis for optimizing resource scheduling and management of deep learning tasks, and improving the utilization efficiency of computing clusters.

cs.LG↗

arXiv2026

Transformers in the Dark: Navigating Unknown Search Spaces via Bandit Feedback

Effective problem solving with Large Language Models (LLMs) can be enhanced when they are paired with external search algorithms. By viewing the space of diverse ideas and their follow-up possibilities as a tree structure, the search algorithm can navigate such a search space and guide the LLM toward better solutions more efficiently. While the search algorithm enables an effective balance between exploitation and exploration of a tree-structured space, the need for an external component can complicate the overall problem-solving process. We therefore pose the following question: Can LLMs or their underlying Transformer architectures approximate a search algorithm? To answer this question, we first introduce a simplified framework in which tree extensions and feedback signals are externally specified, allowing for controlled evaluation of search capabilities. We call this setting unknown tree search with bandit feedback. Within this setting, we show that Transformers are theoretically expressive enough to implement distinct search strategies and can be trained from scratch to approximate those strategies. Our Transformer models exhibit the possibility of generalizing to unseen conditions such as longer horizons or deeper trees. Furthermore, we demonstrate that continued task-focused training unlocks the complete capabilities of a pretrained LLM, by fine-tuning the LLM on search trajectories.

cs.LG↗

arXiv2026

Deep Learning-based Bug Triage System

Effective bug triage is crucial for streamlining the software development lifecycle by accurately categorizing and assigning reported software defects. In this paper, we propose an automated bug triage system built upon the pre-trained RoBERTa-base transformer architecture. By leveraging deep contextual representations, our approach efficiently classifies incoming bug reports to optimize assignment. Experimental evaluation demonstrates that the proposed system achieves a strong bug identification accuracy of 0.90 within just five training epochs. These findings highlight the efficiency and high performance of fine-tuned transformer models for practical software engineering automation.

cs.SE↗

arXiv2026

OASIS: On-the-fly ASynchronous Image Stream classification -- Transformer-based models for the early detection of strongly lensed supernovae

Time-domain surveys often produce asynchronous image sequences across irregular observation times and photometric bands. In many applications, predictions are needed after each new observation, particularly when transient sources must be identified early to trigger follow-up. Strongly lensed supernovae (LSNe), for which follow-up is particularly time-critical, provide a useful example. We formulate this setting as On-the-fly ASynchronous Image Stream (OASIS) classification and explore reusable transformer architectures for causal, on-the-fly classification. We demonstrate these architectures on the lensed-transient classification problem studied by Bag et al. (2026a,b), distinguishing LSNe Ia from unlensed astrophysical transients. We compare them with the recurrent baseline adopted by Bag et al. (2026b), which combines a ConvLSTM branch for image processing with a separate LSTM branch for encoding observation times. The baseline maps asynchronous observations onto a fixed band-by-observation grid, filling unobserved bands with zeros. Using a common training setup, we compare two transformer-based approaches to modeling interactions among spatio-temporal tokens: factorized attention, which separates spatial and temporal interactions, and joint attention, which models them together. On the demonstration dataset, factorized architectures achieve mean losses of $0.095$--$0.100$, compared with $0.128$ for the recurrent baseline and $0.107$ or higher for joint architectures. A selective readout further reduces the best loss to $0.089$ for a factorized model. These results suggest that factorized attention suits this dataset better, without implying universal architectural superiority. We release the architectures as reusable foundations for time-domain astronomy and other applications involving evolving spatial features that require on-the-fly predictions.

astro-ph.IM↗

arXiv2026

Quantum-Enhanced Learning Framework for Intelligent and AI-Native 6G Wireless Networks

The recent convergence of 6G wireless systems and Tiny Machine Learning (TinyML) has driven the need for on-device intelligence in edge networks, where ultra-low latency, stringent energy budgets, and tight compute constraints demand novel architectures. Lightweight deep models efficiently extract local patterns but fail to capture global dependencies, while attention mechanisms do so at the expense of energy and computational cost. To bridge this gap, we introduce Quantumer, a hybrid TinyML--quantum framework that integrates multi-scale dilated convolutions and scaled dot-product attention within a lightweight transformer architecture, employing a two-stage transfer learning pipeline from Quantum Pre-Training (Quantumer-Q) to Classical Fine-Tuning (Quantumer-C). We also present QuantiblentLayer, a four-qubit variational circuit that maps compact traffic representations into measurement-based Hilbert-space features using trainable rotations and cyclic entangling operations. The circuit is used only during offline pre-training as a nonlinear embedding teacher and is removed before Quantumer-C deployment, leaving a fully classical inference model without runtime quantum execution. By transferring these quantum-assisted embeddings into an energy-efficient, lightweight transformer, Quantumer achieves strong detection performance with minimal compute and memory overhead on resource-constrained edge devices. The intrusion detection system (IDS) is used as a case study and evaluated on the Edge-IIoTset, TON IoT, and WUSTL-IIoT-2021 datasets. Quantumer-Q achieves competitive compact-model performance with 105.86K parameters, 0.4038 MB memory usage, 0.5525 MB model size, and 5.5646 MFLOPs; the INT8 Raspberry Pi 4 deployment obtains 16.8413 ms latency with a 0.6493 MB footprint. These results support training-time quantum-assisted representation learning for compact edge-deployable IDS.

quant-ph↗

arXiv2022

Adaptive Sparse and Monotonic Attention for Transformer-based Automatic Speech Recognition

The Transformer architecture model, based on self-attention and multi-head attention, has achieved remarkable success in offline end-to-end Automatic Speech Recognition (ASR). However, self-attention and multi-head attention cannot be easily applied for streaming or online ASR. For self-attention in Transformer ASR, the softmax normalization function-based attention mechanism makes it impossible to highlight important speech information. For multi-head attention in Transformer ASR, it is not easy to model monotonic alignments in different heads. To overcome these two limits, we integrate sparse attention and monotonic attention into Transformer-based ASR. The sparse mechanism introduces a learned sparsity scheme to enable each self-attention structure to fit the corresponding head better. The monotonic attention deploys regularization to prune redundant heads for the multi-head attention structure. The experiments show that our method can effectively improve the attention mechanism on widely used benchmarks of speech recognition.

cs.CL↗

arXiv2022

Evade the Trap of Mediocrity: Promoting Diversity and Novelty in Text Generation via Concentrating Attention

Recently, powerful Transformer architectures have proven superior in generating high-quality sentences. Nevertheless, these models tend to produce dull high-frequency phrases, severely hurting the diversity and novelty of generated text. In this work, we dig into the intrinsic mechanism of this problem and found that sparser attention values in Transformer could improve diversity. To understand such a phenomenon, we first conduct both empirical and theoretical analysis and then attribute it to representation degeneration caused by the attentive mixture of the hidden states during training. We term this process the Trap of Mediocrity. To escape from such a trap, we introduce a novel attention regularization loss to control the sharpness of the attention distribution, which is transparent to model structures and can be easily implemented within 20 lines of python code. We prove that this method could be mathematically regarded as learning a Bayesian approximation of posterior attention. Experiments show that our method improved the diversity and novelty of the generated text while maintaining comparable quality on a variety of conditional and unconditional generation tasks.

cs.CL↗

arXiv2025

Attention Mechanisms Don't Learn Additive Models: Rethinking Feature Importance for Transformers

We address the critical challenge of applying feature attribution methods to the transformer architecture, which dominates current applications in natural language processing and beyond. Traditional attribution methods to explainable AI (XAI) explicitly or implicitly rely on linear or additive surrogate models to quantify the impact of input features on a model's output. In this work, we formally prove an alarming incompatibility: transformers are structurally incapable of representing linear or additive surrogate models used for feature attribution, undermining the grounding of these conventional explanation methodologies. To address this discrepancy, we introduce the Softmax-Linked Additive Log Odds Model (SLALOM), a novel surrogate model specifically designed to align with the transformer framework. SLALOM demonstrates the capacity to deliver a range of insightful explanations with both synthetic and real-world datasets. We highlight SLALOM's unique efficiency-quality curve by showing that SLALOM can produce explanations with substantially higher fidelity than competing surrogate models or provide explanations of comparable quality at a fraction of their computational costs. We release code for SLALOM as an open-source project online at https://github.com/tleemann/slalom_explanations.

cs.LG↗

arXiv2024

Grokked Transformers are Implicit Reasoners: A Mechanistic Journey to the Edge of Generalization

We study whether transformers can learn to implicitly reason over parametric knowledge, a skill that even the most capable language models struggle with. Focusing on two representative reasoning types, composition and comparison, we consistently find that transformers can learn implicit reasoning, but only through grokking, i.e., extended training far beyond overfitting. The levels of generalization also vary across reasoning types: when faced with out-of-distribution examples, transformers fail to systematically generalize for composition but succeed for comparison. We delve into the model's internals throughout training, conducting analytical experiments that reveal: 1) the mechanism behind grokking, such as the formation of the generalizing circuit and its relation to the relative efficiency of generalizing and memorizing circuits, and 2) the connection between systematicity and the configuration of the generalizing circuit. Our findings guide data and training setup to better induce implicit reasoning and suggest potential improvements to the transformer architecture, such as encouraging cross-layer knowledge sharing. Furthermore, we demonstrate that for a challenging reasoning task with a large search space, GPT-4-Turbo and Gemini-1.5-Pro based on non-parametric memory fail badly regardless of prompting styles or retrieval augmentation, while a fully grokked transformer can achieve near-perfect accuracy, showcasing the power of parametric memory for complex reasoning.

cs.CL↗