SEARCH · Searcharxiv

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 901 records · Page 50Linked to original sources

arXiv2025

On the Influence of Shape, Texture and Color for Learning Semantic Segmentation

Recent research has investigated the shape and texture biases of pre-trained deep neural networks (DNNs) in image classification. Those works test how much a trained DNN relies on specific image cues like texture. The present study shifts the focus to understanding the cue influence during training, analyzing what DNNs can learn from shape, texture, and color cues in absence of the others; investigating their individual and combined influence on the learning success. We analyze these cue influences at multiple levels by decomposing datasets into cue-specific versions. Addressing semantic segmentation, we learn the given task from these reduced cue datasets, creating cue experts. Early fusion of cues is performed by constructing appropriate datasets. This is complemented by a late fusion of experts which allows us to study cue influence location-dependent on pixel level. Experiments on Cityscapes, PASCAL Context, and a synthetic CARLA dataset show that while no single cue dominates, the shape + color expert predominantly improves the prediction of small objects and border pixels. The cue performance order is consistent for the tested convolutional and transformer architecture, indicating similar cue extraction capabilities, although pre-trained transformers are said to be more biased towards shape than convolutional neural networks.

cs.CV↗

arXiv2026

Resonant Sparse Geometry Networks

We introduce Resonant Sparse Geometry Networks (RSGN), a brain-inspired architecture with self-organizing sparse hierarchical input-dependent connectivity. Unlike Transformer architectures that employ dense attention mechanisms with O(n^2) computational complexity, RSGN embeds computational nodes in learned hyperbolic space where connection strength decays with geodesic distance, achieving dynamic sparsity that adapts to each input. The architecture operates on two distinct timescales: fast differentiable activation propagation optimized through gradient descent, and slow Hebbian-inspired structural learning for connectivity adaptation through local correlation rules. We provide rigorous mathematical analysis demonstrating that RSGN achieves O(n*k) computational complexity, where k << n represents the average active neighborhood size. Experimental evaluation on hierarchical classification and long-range dependency tasks demonstrates that RSGN achieves 96.5% accuracy on long-range dependency tasks while using approximately 15x fewer parameters than standard Transformers. On challenging hierarchical classification with 20 classes, RSGN achieves 23.8% accuracy (compared to 5% random baseline) with only 41,672 parameters, nearly 10x fewer than the Transformer baselines which require 403,348 parameters to achieve 30.1% accuracy. Our ablation studies confirm the contribution of each architectural component, with Hebbian learning providing consistent improvements. These results suggest that brain-inspired principles of sparse, geometrically-organized computation offer a promising direction toward more efficient and biologically plausible neural architectures.

cs.LG↗

arXiv2021

Video Swin Transformer

The vision community is witnessing a modeling shift from CNNs to Transformers, where pure Transformer architectures have attained top accuracy on the major video recognition benchmarks. These video models are all built on Transformer layers that globally connect patches across the spatial and temporal dimensions. In this paper, we instead advocate an inductive bias of locality in video Transformers, which leads to a better speed-accuracy trade-off compared to previous approaches which compute self-attention globally even with spatial-temporal factorization. The locality of the proposed video architecture is realized by adapting the Swin Transformer designed for the image domain, while continuing to leverage the power of pre-trained image models. Our approach achieves state-of-the-art accuracy on a broad range of video recognition benchmarks, including on action recognition (84.9 top-1 accuracy on Kinetics-400 and 86.1 top-1 accuracy on Kinetics-600 with ~20x less pre-training data and ~3x smaller model size) and temporal modeling (69.6 top-1 accuracy on Something-Something v2). The code and models will be made publicly available at https://github.com/SwinTransformer/Video-Swin-Transformer.

cs.CV↗

arXiv2026

Llama-Mimi: Exploring the Limits of Flattened Speech Language Modeling

Speech Language Models (SpeechLMs) model tokenized speech to capture both semantic and acoustic information. When neural audio codecs based on Residual Vector Quantization (RVQ) are used as audio tokenizers, they produce multiple discrete tokens per time step, yielding inherently multi-level representations. To process these multi-level tokens together, prior work typically adopts hierarchical architectures to capture this structure. In contrast, recent progress in NLP has progressively reduced architectural inductive biases, moving toward simpler and more scalable single-Transformer architectures. In this work, we propose Llama-Mimi, which flattens multi-level RVQ tokens produced by the Mimi neural audio codec into a single sequence and models them autoregressively with a Transformer decoder. We show that Llama-Mimi outperforms a CSM-based hierarchical model on most tasks and achieves the best performance on acoustic consistency. Our models, code, and speech samples are publicly available.

cs.CL↗

arXiv2025

Benchmarking On-Device Machine Learning on Apple Silicon with MLX

The recent widespread adoption of Large Language Models (LLMs) and machine learning in general has sparked research interest in exploring the possibilities of deploying these models on smaller devices such as laptops and mobile phones. This creates a need for frameworks and approaches that are capable of taking advantage of on-device hardware. The MLX framework was created to address this need. It is a framework optimized for machine learning (ML) computations on Apple silicon devices, facilitating easier research, experimentation, and prototyping. This paper presents a performance evaluation of MLX, focusing on inference latency of transformer models. We compare the performance of different transformer architecture implementations in MLX with their Pytorch counterparts. For this research we create a framework called MLX-transformers which includes different transformer implementations in MLX and downloads the model checkpoints in pytorch and converts it to the MLX format. By leveraging the advanced architecture and capabilities of Apple Silicon, MLX-Transformers enables seamless execution of transformer models directly sourced from Hugging Face, eliminating the need for checkpoint conversion often required when porting models between frameworks. Our study benchmarks different transformer models on two Apple Silicon macbook devices against an NVIDIA CUDA GPU. Specifically, we compare the inference latency performance of models with the same parameter sizes and checkpoints. We evaluate the performance of BERT, RoBERTa, and XLM-RoBERTa models, with the intention of extending future work to include models of different modalities, thus providing a more comprehensive assessment of MLX's capabilities. The results highlight MLX's potential in enabling efficient and more accessible on-device ML applications within Apple's ecosystem.

cs.LG↗

arXiv2025

Speed Always Wins: A Survey on Efficient Architectures for Large Language Models

Large Language Models (LLMs) have delivered impressive results in language understanding, generation, reasoning, and pushes the ability boundary of multimodal models. Transformer models, as the foundation of modern LLMs, offer a strong baseline with excellent scaling properties. However, the traditional transformer architecture requires substantial computations and poses significant obstacles for large-scale training and practical deployment. In this survey, we offer a systematic examination of innovative LLM architectures that address the inherent limitations of transformers and boost the efficiency. Starting from language modeling, this survey covers the background and technical details of linear and sparse sequence modeling methods, efficient full attention variants, sparse mixture-of-experts, hybrid model architectures incorporating the above techniques, and emerging diffusion LLMs. Additionally, we discuss applications of these techniques to other modalities and consider their wider implications for developing scalable, resource-aware foundation models. By grouping recent studies into the above category, this survey presents a blueprint of modern efficient LLM architectures, and we hope this could help motivate future research toward more efficient, versatile AI systems.

cs.CL↗

arXiv2024

On the Power of Convolution Augmented Transformer

The transformer architecture has catalyzed revolutionary advances in language modeling. However, recent architectural recipes, such as state-space models, have bridged the performance gap. Motivated by this, we examine the benefits of Convolution-Augmented Transformer (CAT) for recall, copying, and length generalization tasks. CAT incorporates convolutional filters in the K/Q/V embeddings of an attention layer. Through CAT, we show that the locality of the convolution synergizes with the global view of the attention. Unlike comparable architectures, such as Mamba or transformer, CAT can provably solve the associative recall (AR) and copying tasks using a single layer while also enjoying guaranteed length generalization. We also establish computational tradeoffs between convolution and attention by characterizing how convolution can mitigate the need for full attention by summarizing the context window and creating salient summary tokens to attend. Evaluations on real datasets corroborate our findings and demonstrate that CAT and its variations indeed enhance the language modeling performance.

cs.LG↗

arXiv2026

Echo State Transformer: Attention Over Finite Memories

While Large Language Models and their underlying Transformer architecture are remarkably efficient, they do not reflect how our brain processes and learns a diversity of cognitive tasks such as language, nor how it leverages working memory. Furthermore, Transformers encounters a computational limitation: quadratic complexity growth with sequence length. Motivated by these limitations, we aim to design architectures that leverage efficient working memory dynamics to overcome standard computational barriers. We introduce Echo State Transformers (EST), a hybrid architecture that resolves this challenge while demonstrating state of the art performance in classification and detection tasks. EST integrates the Transformer attention mechanisms with nodes from Reservoir Computing to create a fixed-size memory system. Drawing inspiration from Echo State Networks, our approach leverages several reservoirs (random recurrent networks) in parallel as a lightweight and efficient working memory. These independent units possess distinct and learned internal dynamics with an adaptive leak rate, enabling them to dynamically adjust their own temporality. By applying attention on those fixed number of units instead of input tokens, EST achieves linear complexity for the whole sequence, effectively breaking the quadratic scaling problem of standard Transformers. We evaluate ESTs on a recent timeseries benchmark: the Time Series Library, which comprises 69 tasks across five categories. Results show that ESTs ranks first overall in two of five categories, outperforming strong state-of-the-art baselines on classification and anomaly detection tasks, while remaining competitive on short-term forecasting. These results demonstrate that by shifting the attention mechanism from the entire input sequence to a fixed set of evolving memory units, it is possible to maintains high sensitivity to temporal events while achieving constant computational complexity per step.

cs.LG↗

arXiv2023

InCrowdFormer: On-Ground Pedestrian World Model From Egocentric Views

We introduce an on-ground Pedestrian World Model, a computational model that can predict how pedestrians move around an observer in the crowd on the ground plane, but from just the egocentric-views of the observer. Our model, InCrowdFormer, fully leverages the Transformer architecture by modeling pedestrian interaction and egocentric to top-down view transformation with attention, and autoregressively predicts on-ground positions of a variable number of people with an encoder-decoder architecture. We encode the uncertainties arising from unknown pedestrian heights with latent codes to predict the posterior distributions of pedestrian positions. We validate the effectiveness of InCrowdFormer on a novel prediction benchmark of real movements. The results show that InCrowdFormer accurately predicts the future coordination of pedestrians. To the best of our knowledge, InCrowdFormer is the first-of-its-kind pedestrian world model which we believe will benefit a wide range of egocentric-view applications including crowd navigation, tracking, and synthesis.

cs.CV↗

arXiv2024

Adaptive Multi Scale Document Binarisation Using Vision Mamba

Enhancing and preserving the readability of document images, particularly historical ones, is crucial for effective document image analysis. Numerous models have been proposed for this task, including convolutional-based, transformer-based, and hybrid convolutional-transformer architectures. While hybrid models address the limitations of purely convolutional or transformer-based methods, they often suffer from issues like quadratic time complexity. In this work, we propose a Mamba-based architecture for document binarisation, which efficiently handles long sequences by scaling linearly and optimizing memory usage. Additionally, we introduce novel modifications to the skip connections by incorporating Difference of Gaussians (DoG) features, inspired by conventional signal processing techniques. These multiscale high-frequency features enable the model to produce high-quality, detailed outputs.

cs.CV↗

arXiv2024

Survey and Evaluation of Converging Architecture in LLMs based on Footsteps of Operations

The advent of the Attention mechanism and Transformer architecture enables contextually natural text generation and compresses the burden of processing entire source information into singular vectors. Based on these two main ideas, model sizes gradually increases to accommodate more precise and comprehensive information, leading to the current state-of-the-art LLMs being very large, with parameters around 70 billion. As the model sizes are growing, the demand for substantial storage and computational capacity increases. This leads to the development of high-bandwidth memory and accelerators, as well as a variety of model architectures designed to meet these requirements. We note that LLM architectures have increasingly converged. This paper analyzes how these converged architectures perform in terms of layer configurations, operational mechanisms, and model sizes, considering various hyperparameter settings. In this paper, we conduct a concise survey of the history of LLMs by tracing the evolution of their operational improvements. Furthermore, we summarize the performance trends of LLMs under various hyperparameter settings using the RTX 6000, which features the state-of-the-art Ada Lovelace architecture. We conclude that even the same model can exhibit different behaviors depending on the hyperparameters or whether it is deployed in server or edge environments.

cs.LG↗

arXiv2025

Cross-Attention Message-Passing Transformers for Code-Agnostic Decoding in 6G Networks

Channel coding for 6G networks is expected to support a wide range of requirements arising from heterogeneous communication scenarios. These demands challenge traditional code-specific decoders, which lack the flexibility and scalability required for next-generation systems. To tackle this problem, we propose an AI-native foundation model for unified and code-agnostic decoding based on the transformer architecture. We first introduce a cross-attention message-passing transformer (CrossMPT). CrossMPT employs two masked cross-attention blocks that iteratively update two distinct input representations-magnitude and syndrome vectors-allowing the model to effectively learn the decoding problem. Notably, our CrossMPT has achieved state-of-the-art decoding performance among single neural decoders. Building on this, we develop foundation CrossMPT (FCrossMPT) by making the architecture invariant to code length, rate, and class, allowing a single trained model to decode a broad range of codes without retraining. To further enhance decoding performance, particularly for short blocklength codes, we propose CrossMPT ensemble decoder (CrossED), an ensemble decoder composed of multiple parallel CrossMPT blocks employing different parity-check matrices. This architecture can also serve as a foundation model, showing strong generalization across diverse code types. Overall, the proposed AI-native code-agnostic decoder offers flexibility, scalability, and high performance, presenting a promising direction to channel coding for 6G networks.

cs.IT↗

arXiv2022

ByT5: Towards a token-free future with pre-trained byte-to-byte models

Most widely-used pre-trained language models operate on sequences of tokens corresponding to word or subword units. By comparison, token-free models that operate directly on raw text (bytes or characters) have many benefits: they can process text in any language out of the box, they are more robust to noise, and they minimize technical debt by removing complex and error-prone text preprocessing pipelines. Since byte or character sequences are longer than token sequences, past work on token-free models has often introduced new model architectures designed to amortize the cost of operating directly on raw text. In this paper, we show that a standard Transformer architecture can be used with minimal modifications to process byte sequences. We characterize the trade-offs in terms of parameter count, training FLOPs, and inference speed, and show that byte-level models are competitive with their token-level counterparts. We also demonstrate that byte-level models are significantly more robust to noise and perform better on tasks that are sensitive to spelling and pronunciation. As part of our contribution, we release a new set of pre-trained byte-level Transformer models based on the T5 architecture, as well as all code and data used in our experiments.

cs.CL↗

arXiv2024

Trackastra: Transformer-based cell tracking for live-cell microscopy

Cell tracking is a ubiquitous image analysis task in live-cell microscopy. Unlike multiple object tracking (MOT) for natural images, cell tracking typically involves hundreds of similar-looking objects that can divide in each frame, making it a particularly challenging problem. Current state-of-the-art approaches follow the tracking-by-detection paradigm, i.e. first all cells are detected per frame and successively linked in a second step to form biologically consistent cell tracks. Linking is commonly solved via discrete optimization methods, which require manual tuning of hyperparameters for each dataset and are therefore cumbersome to use in practice. Here we propose Trackastra, a general purpose cell tracking approach that uses a simple transformer architecture to directly learn pairwise associations of cells within a temporal window from annotated data. Importantly, unlike existing transformer-based MOT pipelines, our learning architecture also accounts for dividing objects such as cells and allows for accurate tracking even with simple greedy linking, thus making strides towards removing the requirement for a complex linking step. The proposed architecture operates on the full spatio-temporal context of detections within a time window by avoiding the computational burden of processing dense images. We show that our tracking approach performs on par with or better than highly tuned state-of-the-art cell tracking algorithms for various biological datasets, such as bacteria, cell cultures and fluorescent particles. We provide code at https://github.com/weigertlab/trackastra.

cs.CV↗

arXiv2024

Mamba for Scalable and Efficient Personalized Recommendations

In this effort, we propose using the Mamba for handling tabular data in personalized recommendation systems. We present the \textit{FT-Mamba} (Feature Tokenizer\,$+$\,Mamba), a novel hybrid model that replaces Transformer layers with Mamba layers within the FT-Transformer architecture, for handling tabular data in personalized recommendation systems. The \textit{Mamba model} offers an efficient alternative to Transformers, reducing computational complexity from quadratic to linear by enhancing the capabilities of State Space Models (SSMs). FT-Mamba is designed to improve the scalability and efficiency of recommendation systems while maintaining performance. We evaluate FT-Mamba in comparison to a traditional Transformer-based model within a Two-Tower architecture on three datasets: Spotify music recommendation, H\&M fashion recommendation, and vaccine messaging recommendation. Each model is trained on 160,000 user-action pairs, and performance is measured using precision (P), recall (R), Mean Reciprocal Rank (MRR), and Hit Ratio (HR) at several truncation values. Our results demonstrate that FT-Mamba outperforms the Transformer-based model in terms of computational efficiency while maintaining or exceeding performance across key recommendation metrics. By leveraging Mamba layers, FT-Mamba provides a scalable and effective solution for large-scale personalized recommendation systems, showcasing the potential of the Mamba architecture to enhance both efficiency and accuracy.

cs.IR↗

arXiv2026

Relational Attention for Data-Efficient Language Modeling

We present Relational BabyLM, a system submission to the BabyLM 2026 challenge that combines two cognitively motivated inductive biases in a single decoder-only Transformer. Architecturally, we replace standard self-attention with a Dual Attention Transformer (DAT), which separates the routing of object-level ("sensory") lexical features from structural/relational information (Altabaa and Lafferty, 2025; Altabaa et al., 2024; Webb et al., 2024; Kerg et al., 2022; Webb et al., 2021). Relational attention (RA) disentangled from self-attention greatly increases data efficiency and out-of-training-sample generalization on purely relational tasks, but language modeling requires object-level and relational information to be integrated as well as disentangled, and RA-based LMs have remained largely unexplored. BabyLM's data-constrained training and comprehensive evaluation is an ideal testing ground for whether that data efficiency transfers. As a training intervention, we add a Next-Latent Prediction (NextLat; Teoh et al. 2026) objective that encourages hidden states to compress history incrementally into a dense belief state. Architecture is the dominant factor for structural linguistic generalization; the objective is secondary but still significant. DAT's three relational attention types (full RA vs. the simpler RCA and DisRCA variants) are largely interchangeable at 10M words; full RA pulls ahead at 100M. We also introduce a novel symbol-retrieval mechanism (RoPE-based, as opposed to learned, relative symbols) that matches learned symbol libraries while adding no parameters. On the strict (100M-word) track, our best model ranks 6th of 55 overall and 3rd of 55 on the leaderboard's NLP-task subset at the time of writing; our two strongest models outperform the GPT-2 baseline on most benchmarks, with one attaining the highest EWoK score among strict-track entries.

cs.CL↗

arXiv2026

On-Policy Attention Linearization

Hybrid transformer architectures that replace most softmax attention layers with linear attention offer transformer-level quality at a fraction of the memory cost. Rather than pretraining such models, a growing body of work distills them from already trained full-attention transformers. However, these distilled models often collapse on long-context retrieval and reasoning tasks, particularly when operating in thinking mode, where the efficiency gains of hybrid architectures matter most. Since linear attention layers must compress context into a fixed-size state, their errors compound over long sequences. As off-policy distillation never teaches the student model to recover from this drift, tasks that necessitate longer sequence lengths become especially challenging. We introduce On-Policy Attention Linearization (OPAL) in which the hybrid attention student samples its own long-context trajectories and receives dense supervision from the frozen full-attention teacher. Applying OPAL to Qwen3-4B and MiMo-7B-RL-0530, we recover $87$--$94\%$ of full-attention performance on commonsense reasoning, $100\%$ on needle-in-a-haystack (NIAH) retrieval, and $83$--$93\%$ on mathematical reasoning with only 3B training tokens. We achieve these results without supervised fine-tuning (SFT) or reinforcement learning with verifiable rewards (RLVR). Compared with the strongest prior linearization method, which recovers $68\%$ of its teacher's retrieval performance and $21.6\%$ absolute average mathematical reasoning accuracy, OPAL fully recovers retrieval and achieves $67.6$--$72.2\%$ on math reasoning.

cs.LG↗

arXiv2025

A Survey of RWKV

The Receptance Weighted Key Value (RWKV) model offers a novel alternative to the Transformer architecture, merging the benefits of recurrent and attention-based systems. Unlike conventional Transformers, which depend heavily on self-attention, RWKV adeptly captures long-range dependencies with minimal computational demands. By utilizing a recurrent framework, RWKV addresses some computational inefficiencies found in Transformers, particularly in tasks with long sequences. RWKV has recently drawn considerable attention for its robust performance across multiple domains. Despite its growing popularity, no systematic review of the RWKV model exists. This paper seeks to fill this gap as the first comprehensive review of the RWKV architecture, its core principles, and its varied applications, such as natural language generation, natural language understanding, and computer vision. We assess how RWKV compares to traditional Transformer models, highlighting its capability to manage long sequences efficiently and lower computational costs. Furthermore, we explore the challenges RWKV encounters and propose potential directions for future research and advancement. We consistently maintain the related open-source materials at: https://github.com/MLGroupJLU/RWKV-Survey.

cs.CL↗