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 1,477 records · Page 82Linked to original sources

arXiv2025

Polynomial Composition Activations: Unleashing the Dynamics of Large Language Models

Transformers have found extensive applications across various domains due to the powerful fitting capabilities. This success can be partially attributed to their inherent nonlinearity. Thus, in addition to the ReLU function employed in the original transformer architecture, researchers have explored alternative modules such as GeLU and SwishGLU to enhance nonlinearity and thereby augment representational capacity. In this paper, we propose a novel category of polynomial composition activations (PolyCom), designed to optimize the dynamics of transformers. Theoretically, we provide a comprehensive mathematical analysis of PolyCom, highlighting its enhanced expressivity and efficacy relative to other activation functions. Notably, we demonstrate that networks incorporating PolyCom achieve the $\textbf{optimal approximation rate}$, indicating that PolyCom networks require minimal parameters to approximate general smooth functions in Sobolev spaces. We conduct empirical experiments on the pre-training configurations of large language models (LLMs), including both dense and sparse architectures. By substituting conventional activation functions with PolyCom, we enable LLMs to capture higher-order interactions within the data, thus improving performance metrics in terms of accuracy and convergence rates. Extensive experimental results demonstrate the effectiveness of our method, showing substantial improvements over other activation functions. Code is available at https://github.com/BryceZhuo/PolyCom.

cs.CL↗

arXiv2021

Decoupled Spatial-Temporal Transformer for Video Inpainting

Video inpainting aims to fill the given spatiotemporal holes with realistic appearance but is still a challenging task even with prosperous deep learning approaches. Recent works introduce the promising Transformer architecture into deep video inpainting and achieve better performance. However, it still suffers from synthesizing blurry texture as well as huge computational cost. Towards this end, we propose a novel Decoupled Spatial-Temporal Transformer (DSTT) for improving video inpainting with exceptional efficiency. Our proposed DSTT disentangles the task of learning spatial-temporal attention into 2 sub-tasks: one is for attending temporal object movements on different frames at same spatial locations, which is achieved by temporally-decoupled Transformer block, and the other is for attending similar background textures on same frame of all spatial positions, which is achieved by spatially-decoupled Transformer block. The interweaving stack of such two blocks makes our proposed model attend background textures and moving objects more precisely, and thus the attended plausible and temporally-coherent appearance can be propagated to fill the holes. In addition, a hierarchical encoder is adopted before the stack of Transformer blocks, for learning robust and hierarchical features that maintain multi-level local spatial structure, resulting in the more representative token vectors. Seamless combination of these two novel designs forms a better spatial-temporal attention scheme and our proposed model achieves better performance than state-of-the-art video inpainting approaches with significant boosted efficiency.

cs.CV↗

arXiv2023

RoFormer: Enhanced Transformer with Rotary Position Embedding

Position encoding recently has shown effective in the transformer architecture. It enables valuable supervision for dependency modeling between elements at different positions of the sequence. In this paper, we first investigate various methods to integrate positional information into the learning process of transformer-based language models. Then, we propose a novel method named Rotary Position Embedding(RoPE) to effectively leverage the positional information. Specifically, the proposed RoPE encodes the absolute position with a rotation matrix and meanwhile incorporates the explicit relative position dependency in self-attention formulation. Notably, RoPE enables valuable properties, including the flexibility of sequence length, decaying inter-token dependency with increasing relative distances, and the capability of equipping the linear self-attention with relative position encoding. Finally, we evaluate the enhanced transformer with rotary position embedding, also called RoFormer, on various long text classification benchmark datasets. Our experiments show that it consistently overcomes its alternatives. Furthermore, we provide a theoretical analysis to explain some experimental results. RoFormer is already integrated into Huggingface: \url{https://huggingface.co/docs/transformers/model_doc/roformer}.

cs.CL↗

arXiv2023

LipsFormer: Introducing Lipschitz Continuity to Vision Transformers

We present a Lipschitz continuous Transformer, called LipsFormer, to pursue training stability both theoretically and empirically for Transformer-based models. In contrast to previous practical tricks that address training instability by learning rate warmup, layer normalization, attention formulation, and weight initialization, we show that Lipschitz continuity is a more essential property to ensure training stability. In LipsFormer, we replace unstable Transformer component modules with Lipschitz continuous counterparts: CenterNorm instead of LayerNorm, spectral initialization instead of Xavier initialization, scaled cosine similarity attention instead of dot-product attention, and weighted residual shortcut. We prove that these introduced modules are Lipschitz continuous and derive an upper bound on the Lipschitz constant of LipsFormer. Our experiments show that LipsFormer allows stable training of deep Transformer architectures without the need of careful learning rate tuning such as warmup, yielding a faster convergence and better generalization. As a result, on the ImageNet 1K dataset, LipsFormer-Swin-Tiny based on Swin Transformer training for 300 epochs can obtain 82.7\% without any learning rate warmup. Moreover, LipsFormer-CSwin-Tiny, based on CSwin, training for 300 epochs achieves a top-1 accuracy of 83.5\% with 4.7G FLOPs and 24M parameters. The code will be released at \url{https://github.com/IDEA-Research/LipsFormer}.

cs.CV↗

arXiv2024

A Machine Learning Approach Towards SKILL Code Autocompletion

As Moore's Law continues to increase the complexity of electronic systems, Electronic Design Automation (EDA) must advance to meet global demand. An important example of an EDA technology is SKILL, a scripting language used to customize and extend EDA software. Recently, code generation models using the transformer architecture have achieved impressive results in academic settings and have even been used in commercial developer tools to improve developer productivity. To the best of our knowledge, this study is the first to apply transformers to SKILL code autocompletion towards improving the productivity of hardware design engineers. In this study, a novel, data-efficient methodology for generating SKILL code is proposed and experimentally validated. More specifically, we propose a novel methodology for (i) creating a high-quality SKILL dataset with both unlabeled and labeled data, (ii) a training strategy where T5 models pre-trained on general programming language code are fine-tuned on our custom SKILL dataset using unsupervised and supervised learning, and (iii) evaluating synthesized SKILL code. We show that models trained using the proposed methodology outperform baselines in terms of human-judgment score and BLEU score. A major challenge faced was the extremely small amount of available SKILL code data that can be used to train a transformer model to generate SKILL code. Despite our validated improvements, the extremely small dataset available to us was still not enough to train a model that can reliably autocomplete SKILL code. We discuss this and other limitations as well as future work that could address these limitations.

cs.SE↗

arXiv2024

Transformers Provably Learn Sparse Token Selection While Fully-Connected Nets Cannot

The transformer architecture has prevailed in various deep learning settings due to its exceptional capabilities to select and compose structural information. Motivated by these capabilities, Sanford et al. proposed the sparse token selection task, in which transformers excel while fully-connected networks (FCNs) fail in the worst case. Building upon that, we strengthen the FCN lower bound to an average-case setting and establish an algorithmic separation of transformers over FCNs. Specifically, a one-layer transformer trained with gradient descent provably learns the sparse token selection task and, surprisingly, exhibits strong out-of-distribution length generalization. We provide empirical simulations to justify our theoretical findings.

stat.ML↗

arXiv2024

Skill matching at scale: freelancer-project alignment for efficient multilingual candidate retrieval

Finding the perfect match between a job proposal and a set of freelancers is not an easy task to perform at scale, especially in multiple languages. In this paper, we propose a novel neural retriever architecture that tackles this problem in a multilingual setting. Our method encodes project descriptions and freelancer profiles by leveraging pre-trained multilingual language models. The latter are used as backbone for a custom transformer architecture that aims to keep the structure of the profiles and project. This model is trained with a contrastive loss on historical data. Thanks to several experiments, we show that this approach effectively captures skill matching similarity and facilitates efficient matching, outperforming traditional methods.

cs.CL↗

arXiv2024

Optimizing Encoder-Only Transformers for Session-Based Recommendation Systems

Session-based recommendation is the task of predicting the next item a user will interact with, often without access to historical user data. In this work, we introduce Sequential Masked Modeling, a novel approach for encoder-only transformer architectures to tackle the challenges of single-session recommendation. Our method combines data augmentation through window sliding with a unique penultimate token masking strategy to capture sequential dependencies more effectively. By enhancing how transformers handle session data, Sequential Masked Modeling significantly improves next-item prediction performance. We evaluate our approach on three widely-used datasets, Yoochoose 1/64, Diginetica, and Tmall, comparing it to state-of-the-art single-session, cross-session, and multi-relation approaches. The results demonstrate that our Transformer-SMM models consistently outperform all models that rely on the same amount of information, while even rivaling methods that have access to more extensive user history. This study highlights the potential of encoder-only transformers in session-based recommendation and opens the door for further improvements.

cs.IR↗

arXiv2025

Balancing Accuracy, Calibration, and Efficiency in Active Learning with Vision Transformers Under Label Noise

Fine-tuning pre-trained convolutional neural networks on ImageNet for downstream tasks is well-established. Still, the impact of model size on the performance of vision transformers in similar scenarios, particularly under label noise, remains largely unexplored. Given the utility and versatility of transformer architectures, this study investigates their practicality under low-budget constraints and noisy labels. We explore how classification accuracy and calibration are affected by symmetric label noise in active learning settings, evaluating four vision transformer configurations (Base and Large with 16x16 and 32x32 patch sizes) and three Swin Transformer configurations (Tiny, Small, and Base) on CIFAR10 and CIFAR100 datasets, under varying label noise rates. Our findings show that larger ViT models (ViTl32 in particular) consistently outperform their smaller counterparts in both accuracy and calibration, even under moderate to high label noise, while Swin Transformers exhibit weaker robustness across all noise levels. We find that smaller patch sizes do not always lead to better performance, as ViTl16 performs consistently worse than ViTl32 while incurring a higher computational cost. We also find that information-based Active Learning strategies only provide meaningful accuracy improvements at moderate label noise rates, but they result in poorer calibration compared to models trained on randomly acquired labels, especially at high label noise rates. We hope these insights provide actionable guidance for practitioners looking to deploy vision transformers in resource-constrained environments, where balancing model complexity, label noise, and compute efficiency is critical in model fine-tuning or distillation.

cs.CV↗

arXiv2026

Bearing Syntactic Fruit with Stack-Augmented Neural Networks

When children learn language, they make syntactic generalizations based on hierarchical rules. A recent line of work has inquired as to whether common neural network architectures share this inductive bias for hierarchical syntax, finding that they do so only under special conditions: when augmented with ground-truth parse tree structures, when pre-trained on massive corpora, or when trained long past convergence. In this paper, we demonstrate, for the first time, neural network architectures that generalize in human-like fashion when trained only on surface forms: stack-augmented neural networks. We test three base architectures (transformer, simple RNN, LSTM) augmented with two styles of stack, one of which leverages nondeterminism. We find that transformers with nondeterministic stacks generalize best on multiple tasks designed to measure hierarchical inductive bias. This suggests that stack-augmented neural networks may be more accurate models of human syntax acquisition than standard architectures, serving as useful objects of psycholinguistic study. Our code is publicly available.

cs.CL↗

arXiv2026

Clash of the models: Comparing performance of BERT-based variants for generic news frame detection

Framing continues to remain one of the most extensively applied theories in political communication. Developments in computation, particularly with the introduction of transformer architecture and more so with large language models (LLMs), have naturally prompted scholars to explore various novel computational approaches, especially for deductive frame detection, in recent years. While many studies have shown that different transformer models outperform their preceding models that use bag-of-words features, the debate continues to evolve regarding how these models compare with each other on classification tasks. By placing itself at this juncture, this study makes three key contributions: First, it comparatively performs generic news frame detection and compares the performance of five BERT-based variants (BERT, RoBERTa, DeBERTa, DistilBERT and ALBERT) to add to the debate on best practices around employing computational text analysis for political communication studies. Second, it introduces various fine-tuned models capable of robustly performing generic news frame detection. Third, building upon numerous previous studies that work with US-centric data, this study provides the scholarly community with a labelled generic news frames dataset based on the Swiss electoral context that aids in testing the contextual robustness of these computational approaches to framing analysis.

cs.CL↗

arXiv2026

Dynamic sparsity in tree-structured feed-forward layers at scale

At typical context lengths, the feed-forward MLP block accounts for a large share of a transformer's compute budget, motivating sparse alternatives to dense MLP blocks. We study sparse, tree-structured feed-forward layers as drop-in replacements for MLP blocks in deep transformer architectures, enabling conditional computation via hard hierarchical routing without a separate router network. We demonstrate for the first time that this form of tree-structured conditional sparsity can be applied for autoregressive language modeling and downstream question answering, including zero- and few-shot settings, and its scalability beyond 1B parameters. Despite activating fewer than 5% of the feed-forward block's units per token, our models match dense baselines under controlled training and fine-tuning protocols. We further analyze training dynamics and identify an emergent auto-pruning effect: the interaction of hard routing with asymmetric nonlinearities progressively deactivates unused paths, yielding partial conversion of dynamic routing into static structural sparsity. We show that simple architectural choices can modulate this behavior and recover balanced trees without auxiliary losses. Overall, our work demonstrates that tree-structured feed-forward layers provide a scalable and controllable mechanism for sparsifying large transformer models.

cs.CL↗

arXiv2026

MEG-Mamba: A Scalable State-Space Foundation Model for Magnetoencephalography

Magnetoencephalography (MEG) is an imaging technique that offers a non-invasive, millisecond-resolution view of human brain activity. The increasing availability of MEG data presents an opportunity to take advantage of a recent advance in artificial intelligence, namely self-supervised foundation models. Existing foundation models for MEG (and electroencephalography) have been built on a transformer architecture. Here, we introduce MEG-Mamba: a generative foundation model for neural activity (source reconstructed, parcellated MEG) built on a Mamba architecture. MEG-Mamba is trained to predict the next token of a discretised MEG signal, conditioned on a brain region and recording session. MEG-Mamba surpasses the generative fidelity of a transformer-based alternative (MEG-GPT) while pre-training in less time (22 vs 400 GPU-hours) and modelling a longer context (4 vs 0.32 s). We evaluate MEG-Mamba by examining the spatio-spectral characteristics of the data it generates and by interpreting its learned embeddings. Furthermore, we demonstrate that lightweight stimulus conditioning (LoRA) can be used to generate realistic task-evoked responses that are not included in pre-training. Our results suggest that Mamba is a promising backbone for scaling neural foundation models.

q-bio.NC↗

arXiv2024

CSWin-UNet: Transformer UNet with Cross-Shaped Windows for Medical Image Segmentation

Deep learning, especially convolutional neural networks (CNNs) and Transformer architectures, have become the focus of extensive research in medical image segmentation, achieving impressive results. However, CNNs come with inductive biases that limit their effectiveness in more complex, varied segmentation scenarios. Conversely, while Transformer-based methods excel at capturing global and long-range semantic details, they suffer from high computational demands. In this study, we propose CSWin-UNet, a novel U-shaped segmentation method that incorporates the CSWin self-attention mechanism into the UNet to facilitate horizontal and vertical stripes self-attention. This method significantly enhances both computational efficiency and receptive field interactions. Additionally, our innovative decoder utilizes a content-aware reassembly operator that strategically reassembles features, guided by predicted kernels, for precise image resolution restoration. Our extensive empirical evaluations on diverse datasets, including synapse multi-organ CT, cardiac MRI, and skin lesions, demonstrate that CSWin-UNet maintains low model complexity while delivering high segmentation accuracy. Codes are available at https://github.com/eatbeanss/CSWin-UNet.

eess.IV↗

arXiv2026

SCAN: Visual Explanations with Self-Confidence and Analysis Networks

Explainable AI (XAI) has become essential in computer vision to make the decision-making processes of deep learning models transparent. However, current visual explanation (XAI) methods face a critical trade-off between the high fidelity of architecture-specific methods and the broad applicability of universal ones. This often results in abstract or fragmented explanations and makes it difficult to compare explanatory power across diverse model families, such as CNNs and Transformers. This paper introduces the Self-Confidence and Analysis Networks (SCAN), a novel universal framework that overcomes these limitations for both convolutional neural network and transformer architectures. SCAN utilizes an AutoEncoder-based approach to reconstruct features from a model's intermediate layers. Guided by the Information Bottleneck principle, it generates a high-resolution Self-Confidence Map that identifies information-rich regions. Extensive experiments on diverse architectures and datasets demonstrate that SCAN consistently achieves outstanding performance on various quantitative metrics such as AUC-D, Negative AUC, Drop%, and Win%. Qualitatively, it produces significantly clearer, object-focused explanations than existing methods. By providing a unified framework that is both architecturally universal and highly faithful, SCAN enhances model transparency and offers a more reliable tool for understanding the decision-making processes of complex neural networks.

cs.CV↗

arXiv2022

Vision Transformer Compression with Structured Pruning and Low Rank Approximation

Transformer architecture has gained popularity due to its ability to scale with large dataset. Consequently, there is a need to reduce the model size and latency, especially for on-device deployment. We focus on vision transformer proposed for image recognition task (Dosovitskiy et al., 2021), and explore the application of different compression techniques such as low rank approximation and pruning for this purpose. Specifically, we investigate a structured pruning method proposed recently in Zhu et al. (2021) and find that mostly feedforward blocks are pruned with this approach, that too, with severe degradation in accuracy. We propose a hybrid compression approach to mitigate this where we compress the attention blocks using low rank approximation and use the previously mentioned pruning with a lower rate for feedforward blocks in each transformer layer. Our technique results in 50% compression with 14% relative increase in classification error whereas we obtain 44% compression with 20% relative increase in error when only pruning is applied. We propose further enhancements to bridge the accuracy gap but leave it as a future work.

cs.CV↗

arXiv2022

XMorpher: Full Transformer for Deformable Medical Image Registration via Cross Attention

An effective backbone network is important to deep learning-based Deformable Medical Image Registration (DMIR), because it extracts and matches the features between two images to discover the mutual correspondence for fine registration. However, the existing deep networks focus on single image situation and are limited in registration task which is performed on paired images. Therefore, we advance a novel backbone network, XMorpher, for the effective corresponding feature representation in DMIR. 1) It proposes a novel full transformer architecture including dual parallel feature extraction networks which exchange information through cross attention, thus discovering multi-level semantic correspondence while extracting respective features gradually for final effective registration. 2) It advances the Cross Attention Transformer (CAT) blocks to establish the attention mechanism between images which is able to find the correspondence automatically and prompts the features to fuse efficiently in the network. 3) It constrains the attention computation between base windows and searching windows with different sizes, and thus focuses on the local transformation of deformable registration and enhances the computing efficiency at the same time. Without any bells and whistles, our XMorpher gives Voxelmorph 2.8% improvement on DSC , demonstrating its effective representation of the features from the paired images in DMIR. We believe that our XMorpher has great application potential in more paired medical images. Our XMorpher is open on https://github.com/Solemoon/XMorpher

cs.CV↗

arXiv2022

MTSMAE: Masked Autoencoders for Multivariate Time-Series Forecasting

Large-scale self-supervised pre-training Transformer architecture have significantly boosted the performance for various tasks in natural language processing (NLP) and computer vision (CV). However, there is a lack of researches on processing multivariate time-series by pre-trained Transformer, and especially, current study on masking time-series for self-supervised learning is still a gap. Different from language and image processing, the information density of time-series increases the difficulty of research. The challenge goes further with the invalidity of the previous patch embedding and mask methods. In this paper, according to the data characteristics of multivariate time-series, a patch embedding method is proposed, and we present an self-supervised pre-training approach based on Masked Autoencoders (MAE), called MTSMAE, which can improve the performance significantly over supervised learning without pre-training. Evaluating our method on several common multivariate time-series datasets from different fields and with different characteristics, experiment results demonstrate that the performance of our method is significantly better than the best method currently available.

cs.LG↗