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 1,387 records · Page 77Linked to original sources

arXiv2026

ProtModernBERT: Compute-Efficient Protein Language Modeling

Large protein language models provide a powerful framework for learning sequence-function relationships and guiding experimental design. However, architectural improvements are underexplored, especially in low-compute regimes of relevance for resource-constrained research laboratories. We show that the training cost for encoder-only transformer architectures can be reduced by leveraging modern architectural refinements, offering an encouraging performance / compute-efficiency trade-off. We introduce ProtModernBERT, a modernized BERT-style model pre-trained on amino acid sequences, and compare it with ProtBert, Amplify, and ESM2 across a set of supervised structural tasks -encompassing secondary structure prediction, disorder prediction, and subcellular localization- and ProteinGym's zero-shot functional benchmark. We find that ProtModernBERT matches or closely tracks the performance of Amplify-120M and ProtBert with up to 80-90\% reduction in pre-training FLOPs, and narrows the gap to ESM2 when controlling for pre-training. Our analysis using a cluster-based validation protocol for hyperparameter selection and randomly initialized models disentangles the contributions of architecture and data, showing that modern components provide stronger inductive biases that are more consequential than parameter count in low-compute regimes.

cs.LG↗

arXiv2024

Embedded Heterogeneous Attention Transformer for Cross-lingual Image Captioning

Cross-lingual image captioning is a challenging task that requires addressing both cross-lingual and cross-modal obstacles in multimedia analysis. The crucial issue in this task is to model the global and the local matching between the image and different languages. Existing cross-modal embedding methods based on the transformer architecture oversee the local matching between the image region and monolingual words, especially when dealing with diverse languages. To overcome these limitations, we propose an Embedded Heterogeneous Attention Transformer (EHAT) to establish cross-domain relationships and local correspondences between images and different languages by using a heterogeneous network. EHAT comprises Masked Heterogeneous Cross-attention (MHCA), Heterogeneous Attention Reasoning Network (HARN), and Heterogeneous Co-attention (HCA). The HARN serves as the core network and it captures cross-domain relationships by leveraging visual bounding box representation features to connect word features from two languages and to learn heterogeneous maps. MHCA and HCA facilitate cross-domain integration in the encoder through specialized heterogeneous attention mechanisms, enabling a single model to generate captions in two languages. We evaluate our approach on the MSCOCO dataset to generate captions in English and Chinese, two languages that exhibit significant differences in their language families. The experimental results demonstrate the superior performance of our method compared to existing advanced monolingual methods. Our proposed EHAT framework effectively addresses the challenges of cross-lingual image captioning, paving the way for improved multilingual image analysis and understanding.

cs.CV↗

arXiv2024

MDT-A2G: Exploring Masked Diffusion Transformers for Co-Speech Gesture Generation

Recent advancements in the field of Diffusion Transformers have substantially improved the generation of high-quality 2D images, 3D videos, and 3D shapes. However, the effectiveness of the Transformer architecture in the domain of co-speech gesture generation remains relatively unexplored, as prior methodologies have predominantly employed the Convolutional Neural Network (CNNs) or simple a few transformer layers. In an attempt to bridge this research gap, we introduce a novel Masked Diffusion Transformer for co-speech gesture generation, referred to as MDT-A2G, which directly implements the denoising process on gesture sequences. To enhance the contextual reasoning capability of temporally aligned speech-driven gestures, we incorporate a novel Masked Diffusion Transformer. This model employs a mask modeling scheme specifically designed to strengthen temporal relation learning among sequence gestures, thereby expediting the learning process and leading to coherent and realistic motions. Apart from audio, Our MDT-A2G model also integrates multi-modal information, encompassing text, emotion, and identity. Furthermore, we propose an efficient inference strategy that diminishes the denoising computation by leveraging previously calculated results, thereby achieving a speedup with negligible performance degradation. Experimental results demonstrate that MDT-A2G excels in gesture generation, boasting a learning speed that is over 6$\times$ faster than traditional diffusion transformers and an inference speed that is 5.7$\times$ than the standard diffusion model.

cs.CV↗

arXiv2025

Propositional Logic for Probing Generalization in Neural Networks

The extent to which neural networks are able to acquire and represent symbolic rules remains a key topic of research and debate. Much current work focuses on the impressive capabilities of large language models, as well as their often ill-understood failures on a wide range of reasoning tasks. In this paper, in contrast, we investigate the generalization behavior of three key neural architectures (Transformers, Graph Convolution Networks and LSTMs) in a controlled task rooted in propositional logic. The task requires models to generate satisfying assignments for logical formulas, making it a structured and interpretable setting for studying compositionality. We introduce a balanced extension of an existing dataset to eliminate superficial patterns and enable testing on unseen operator combinations. Using this dataset, we evaluate the ability of the three architectures to generalize beyond the training distribution. While all models perform well in-distribution, we find that generalization to unseen patterns, particularly those involving negation, remains a significant challenge. Transformers fail to apply negation compositionally, unless structural biases are introduced. Our findings highlight persistent limitations in the ability of standard architectures to learn systematic representations of logical operators, suggesting the need for stronger inductive biases to support robust rule-based reasoning.

cs.LG↗

arXiv2022

Restormer: Efficient Transformer for High-Resolution Image Restoration

Since convolutional neural networks (CNNs) perform well at learning generalizable image priors from large-scale data, these models have been extensively applied to image restoration and related tasks. Recently, another class of neural architectures, Transformers, have shown significant performance gains on natural language and high-level vision tasks. While the Transformer model mitigates the shortcomings of CNNs (i.e., limited receptive field and inadaptability to input content), its computational complexity grows quadratically with the spatial resolution, therefore making it infeasible to apply to most image restoration tasks involving high-resolution images. In this work, we propose an efficient Transformer model by making several key designs in the building blocks (multi-head attention and feed-forward network) such that it can capture long-range pixel interactions, while still remaining applicable to large images. Our model, named Restoration Transformer (Restormer), achieves state-of-the-art results on several image restoration tasks, including image deraining, single-image motion deblurring, defocus deblurring (single-image and dual-pixel data), and image denoising (Gaussian grayscale/color denoising, and real image denoising). The source code and pre-trained models are available at https://github.com/swz30/Restormer.

cs.CV↗

arXiv2020

Transformer Query-Target Knowledge Discovery (TEND): Drug Discovery from CORD-19

Previous work established skip-gram word2vec models could be used to mine knowledge in the materials science literature for the discovery of thermoelectrics. Recent transformer architectures have shown great progress in language modeling and associated fine-tuned tasks, but they have yet to be adapted for drug discovery. We present a RoBERTa transformer-based method that extends the masked language token prediction using query-target conditioning to treat the specificity challenge. The transformer discovery method entails several benefits over the word2vec method including domain-specific (antiviral) analogy performance, negation handling, and flexible query analysis (specific) and is demonstrated on influenza drug discovery. To stimulate COVID-19 research, we release an influenza clinical trials and antiviral analogies dataset used in conjunction with the COVID-19 Open Research Dataset Challenge (CORD-19) literature dataset in the study. We examine k-shot fine-tuning to improve the downstream analogies performance as well as to mine analogies for model explainability. Further, the query-target analysis is verified in a forward chaining analysis against the influenza drug clinical trials dataset, before adapted for COVID-19 drugs (combinations and side-effects) and on-going clinical trials. In consideration of the present topic, we release the model, dataset, and code.

cs.CL↗

arXiv2023

Continuous Spatiotemporal Transformers

Modeling spatiotemporal dynamical systems is a fundamental challenge in machine learning. Transformer models have been very successful in NLP and computer vision where they provide interpretable representations of data. However, a limitation of transformers in modeling continuous dynamical systems is that they are fundamentally discrete time and space models and thus have no guarantees regarding continuous sampling. To address this challenge, we present the Continuous Spatiotemporal Transformer (CST), a new transformer architecture that is designed for the modeling of continuous systems. This new framework guarantees a continuous and smooth output via optimization in Sobolev space. We benchmark CST against traditional transformers as well as other spatiotemporal dynamics modeling methods and achieve superior performance in a number of tasks on synthetic and real systems, including learning brain dynamics from calcium imaging data.

cs.LG↗

arXiv2023

When Do Transformers Shine in RL? Decoupling Memory from Credit Assignment

Reinforcement learning (RL) algorithms face two distinct challenges: learning effective representations of past and present observations, and determining how actions influence future returns. Both challenges involve modeling long-term dependencies. The Transformer architecture has been very successful to solve problems that involve long-term dependencies, including in the RL domain. However, the underlying reason for the strong performance of Transformer-based RL methods remains unclear: is it because they learn effective memory, or because they perform effective credit assignment? After introducing formal definitions of memory length and credit assignment length, we design simple configurable tasks to measure these distinct quantities. Our empirical results reveal that Transformers can enhance the memory capability of RL algorithms, scaling up to tasks that require memorizing observations $1500$ steps ago. However, Transformers do not improve long-term credit assignment. In summary, our results provide an explanation for the success of Transformers in RL, while also highlighting an important area for future research and benchmark design. Our code is open-sourced at https://github.com/twni2016/Memory-RL

cs.LG↗

arXiv2023

Heterogeneous graphs model spatial relationships between biological entities for breast cancer diagnosis

The heterogeneity of breast cancer presents considerable challenges for its early detection, prognosis, and treatment selection. Convolutional neural networks often neglect the spatial relationships within histopathological images, which can limit their accuracy. Graph neural networks (GNNs) offer a promising solution by coding the spatial relationships within images. Prior studies have investigated the modeling of histopathological images as cell and tissue graphs, but they have not fully tapped into the potential of extracting interrelationships between these biological entities. In this paper, we present a novel approach using a heterogeneous GNN that captures the spatial and hierarchical relations between cell and tissue graphs to enhance the extraction of useful information from histopathological images. We also compare the performance of a cross-attention-based network and a transformer architecture for modeling the intricate relationships within tissue and cell graphs. Our model demonstrates superior efficiency in terms of parameter count and achieves higher accuracy compared to the transformer-based state-of-the-art approach on three publicly available breast cancer datasets -- BRIGHT, BreakHis, and BACH.

cs.CV↗

arXiv2024

Unlocking the Power of Patch: Patch-Based MLP for Long-Term Time Series Forecasting

Recent studies have attempted to refine the Transformer architecture to demonstrate its effectiveness in Long-Term Time Series Forecasting (LTSF) tasks. Despite surpassing many linear forecasting models with ever-improving performance, we remain skeptical of Transformers as a solution for LTSF. We attribute the effectiveness of these models largely to the adopted Patch mechanism, which enhances sequence locality to an extent yet fails to fully address the loss of temporal information inherent to the permutation-invariant self-attention mechanism. Further investigation suggests that simple linear layers augmented with the Patch mechanism may outperform complex Transformer-based LTSF models. Moreover, diverging from models that use channel independence, our research underscores the importance of cross-variable interactions in enhancing the performance of multivariate time series forecasting. The interaction information between variables is highly valuable but has been misapplied in past studies, leading to suboptimal cross-variable models. Based on these insights, we propose a novel and simple Patch-based MLP (PatchMLP) for LTSF tasks. Specifically, we employ simple moving averages to extract smooth components and noise-containing residuals from time series data, engaging in semantic information interchange through channel mixing and specializing in random noise with channel independence processing. The PatchMLP model consistently achieves state-of-the-art results on several real-world datasets. We hope this surprising finding will spur new research directions in the LTSF field and pave the way for more efficient and concise solutions.

cs.LG↗

arXiv2024

Unveiling Induction Heads: Provable Training Dynamics and Feature Learning in Transformers

In-context learning (ICL) is a cornerstone of large language model (LLM) functionality, yet its theoretical foundations remain elusive due to the complexity of transformer architectures. In particular, most existing work only theoretically explains how the attention mechanism facilitates ICL under certain data models. It remains unclear how the other building blocks of the transformer contribute to ICL. To address this question, we study how a two-attention-layer transformer is trained to perform ICL on $n$-gram Markov chain data, where each token in the Markov chain statistically depends on the previous $n$ tokens. We analyze a sophisticated transformer model featuring relative positional embedding, multi-head softmax attention, and a feed-forward layer with normalization. We prove that the gradient flow with respect to a cross-entropy ICL loss converges to a limiting model that performs a generalized version of the induction head mechanism with a learned feature, resulting from the congruous contribution of all the building blocks. In the limiting model, the first attention layer acts as a $\mathit{copier}$, copying past tokens within a given window to each position, and the feed-forward network with normalization acts as a $\mathit{selector}$ that generates a feature vector by only looking at informationally relevant parents from the window. Finally, the second attention layer is a $\mathit{classifier}$ that compares these features with the feature at the output position, and uses the resulting similarity scores to generate the desired output. Our theory is further validated by experiments.

cs.LG↗

arXiv2025

xEdgeFace: Efficient Cross-Spectral Face Recognition for Edge Devices

Heterogeneous Face Recognition (HFR) addresses the challenge of matching face images across different sensing modalities, such as thermal to visible or near-infrared to visible, expanding the applicability of face recognition systems in real-world, unconstrained environments. While recent HFR methods have shown promising results, many rely on computation-intensive architectures, limiting their practicality for deployment on resource-constrained edge devices. In this work, we present a lightweight yet effective HFR framework by adapting a hybrid CNN-Transformer architecture originally designed for face recognition. Our approach enables efficient end-to-end training with minimal paired heterogeneous data while preserving strong performance on standard RGB face recognition tasks. This makes it a compelling solution for both homogeneous and heterogeneous scenarios. Extensive experiments across multiple challenging HFR and face recognition benchmarks demonstrate that our method consistently outperforms state-of-the-art approaches while maintaining a low computational overhead.

cs.CV↗

arXiv2026

SATFormer: Learning When to Use Early Value Residuals in Transformers

Recent Transformer architectures expose later layers to representations computed in earlier layers, motivated by the observation that low-level features can become harder to recover as the residual stream is transformed through depth. One such method adds static value residuals: learned mixing coefficients that expose the first-layer value projection V_1 uniformly across tokens and heads. However, we theorize that the usefulness of $V_1$ is unlikely to be constant across tokens, heads, and contexts, and therefore treat early-representation reuse as a retrieval problem. Here, we introduce Selective Access Transformer (SATFormer), which preserves the first-layer value pathway while controlling access with a context-dependent gate. Across models from 130M to 1.3B parameters, SATFormer improves validation loss over the static value-residual and Transformer baselines at every evaluated scale, and improves average zero-shot accuracy over both at all four scales. Its strongest gains appear on retrieval-intensive benchmarks, where it improves over static value residuals by approximately 1.5 average points, while maintaining near baseline throughput and memory. Mechanistic analyses suggest sparse, depth-dependent, head-specific, and category-sensitive access patterns, supporting the interpretation that SATFormer learns selective reuse of early representations rather than uniform residual copying. Our code is available at https://github.com/skyegunasekaran/satformer.

cs.LG↗

arXiv2026

ThinkingVLA: Interleaved Vision and Language Reasoning for Robotic Manipulation

Most Vision-Language-Action (VLA) models map observations directly to actions without explicit reasoning, limiting their capacity for reasoning-intensive long-horizon tasks. To address this, existing approaches adopt Chain-of-Thought (CoT) reasoning to enable subgoal decomposition and spatial anticipation. However, those methods lack a unified architecture for effective cross-modal reasoning and fail to explicitly include inverse reasoning ability based on the target state. We argue that manipulation planning naturally decomposes into prediction, anticipating the next visual state, and inverse dynamics, inferring the actions to reach it. Bridging both requires a unified autoregressive architecture that interleaves textual and visual reasoning in a single generation process. We propose \textbf{ThinkingVLA}, a generative model that realizes this decomposition within a unified Mixture-of-Transformers architecture. ThinkingVLA consists of a forward CoT that identifies the immediate subgoal and guides the visual forecasting; the predicted image then serves as the target state, grounding an inverse CoT that reasons about spatial relationships and action intent based on the predicted image; and the final action is generated conditioned on this full reasoning context. Extensive experiments on simulation and real-world benchmarks demonstrate that ThinkingVLA consistently outperforms state-of-the-art baselines, with particularly large gains on long-horizon manipulation tasks.

cs.RO↗

arXiv2022

Transkimmer: Transformer Learns to Layer-wise Skim

Transformer architecture has become the de-facto model for many machine learning tasks from natural language processing and computer vision. As such, improving its computational efficiency becomes paramount. One of the major computational inefficiency of Transformer-based models is that they spend the identical amount of computation throughout all layers. Prior works have proposed to augment the Transformer model with the capability of skimming tokens to improve its computational efficiency. However, they suffer from not having effectual and end-to-end optimization of the discrete skimming predictor. To address the above limitations, we propose the Transkimmer architecture, which learns to identify hidden state tokens that are not required by each layer. The skimmed tokens are then forwarded directly to the final output, thus reducing the computation of the successive layers. The key idea in Transkimmer is to add a parameterized predictor before each layer that learns to make the skimming decision. We also propose to adopt reparameterization trick and add skim loss for the end-to-end training of Transkimmer. Transkimmer achieves 10.97x average speedup on GLUE benchmark compared with vanilla BERT-base baseline with less than 1% accuracy degradation.

cs.CL↗

arXiv2022

Efficient Joint Detection and Multiple Object Tracking with Spatially Aware Transformer

We propose a light-weight and highly efficient Joint Detection and Tracking pipeline for the task of Multi-Object Tracking using a fully-transformer architecture. It is a modified version of TransTrack, which overcomes the computational bottleneck associated with its design, and at the same time, achieves state-of-the-art MOTA score of 73.20%. The model design is driven by a transformer based backbone instead of CNN, which is highly scalable with the input resolution. We also propose a drop-in replacement for Feed Forward Network of transformer encoder layer, by using Butterfly Transform Operation to perform channel fusion and depth-wise convolution to learn spatial context within the feature maps, otherwise missing within the attention maps of the transformer. As a result of our modifications, we reduce the overall model size of TransTrack by 58.73% and the complexity by 78.72%. Therefore, we expect our design to provide novel perspectives for architecture optimization in future research related to multi-object tracking.

cs.CV↗

arXiv2022

Patch Slimming for Efficient Vision Transformers

This paper studies the efficiency problem for visual transformers by excavating redundant calculation in given networks. The recent transformer architecture has demonstrated its effectiveness for achieving excellent performance on a series of computer vision tasks. However, similar to that of convolutional neural networks, the huge computational cost of vision transformers is still a severe issue. Considering that the attention mechanism aggregates different patches layer-by-layer, we present a novel patch slimming approach that discards useless patches in a top-down paradigm. We first identify the effective patches in the last layer and then use them to guide the patch selection process of previous layers. For each layer, the impact of a patch on the final output feature is approximated and patches with less impact will be removed. Experimental results on benchmark datasets demonstrate that the proposed method can significantly reduce the computational costs of vision transformers without affecting their performances. For example, over 45% FLOPs of the ViT-Ti model can be reduced with only 0.2% top-1 accuracy drop on the ImageNet dataset.

cs.CV↗

arXiv2022

ViSTA: Vision and Scene Text Aggregation for Cross-Modal Retrieval

Visual appearance is considered to be the most important cue to understand images for cross-modal retrieval, while sometimes the scene text appearing in images can provide valuable information to understand the visual semantics. Most of existing cross-modal retrieval approaches ignore the usage of scene text information and directly adding this information may lead to performance degradation in scene text free scenarios. To address this issue, we propose a full transformer architecture to unify these cross-modal retrieval scenarios in a single $\textbf{Vi}$sion and $\textbf{S}$cene $\textbf{T}$ext $\textbf{A}$ggregation framework (ViSTA). Specifically, ViSTA utilizes transformer blocks to directly encode image patches and fuse scene text embedding to learn an aggregated visual representation for cross-modal retrieval. To tackle the modality missing problem of scene text, we propose a novel fusion token based transformer aggregation approach to exchange the necessary scene text information only through the fusion token and concentrate on the most important features in each modality. To further strengthen the visual modality, we develop dual contrastive learning losses to embed both image-text pairs and fusion-text pairs into a common cross-modal space. Compared to existing methods, ViSTA enables to aggregate relevant scene text semantics with visual appearance, and hence improve results under both scene text free and scene text aware scenarios. Experimental results show that ViSTA outperforms other methods by at least $\bf{8.4}\%$ at Recall@1 for scene text aware retrieval task. Compared with state-of-the-art scene text free retrieval methods, ViSTA can achieve better accuracy on Flicker30K and MSCOCO while running at least three times faster during the inference stage, which validates the effectiveness of the proposed framework.

cs.CV↗