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 991 records · Page 55Linked to original sources

arXiv2026

Full-bandwidth transformer

Autoregressive transformers compute along two axes: horizontally across generated tokens, and vertically through model depth. Dense attention gives each token broad horizontal access to the past, but the vertical feedback channel between decoding steps remains narrow: only the sampled token returns to the bottom of the stack, while the top-layer hidden state is discarded. We introduce the full-bandwidth transformer, which widens this channel with latent feedback: at each decoding step, the previous top-layer hidden state is fused with the sampled token embedding through a gated linear unit and fed back as the next input. Latent feedback lets non-verbalized computation re-enter the stack with a renewed depth budget, while preserving the standard transformer architecture, KV cache, and language-modeling objective. To train full-bandwidth transformers without losing parallel teacher forcing, we use a scheduled multi-pass objective that introduces latent feedback late in pretraining and mixes a small fraction of deeper feedback passes for stability. We train 1B-parameter full-bandwidth transformers on up to 400B tokens and find that latent feedback improves validation loss, 5-shot language-model evaluation, math and coding generation, and instruction-tuned performance. With negligible per-token decoding overhead, full-bandwidth transformers match or approach standard transformers trained with roughly 1.5x more tokens, and manage to produce shorter reasoning when no off-policy templates are provided.

cs.AI↗

arXiv2026

Generative midtended cognition and Artificial Intelligence. Thinging with thinging things

This paper introduces the concept of ``generative midtended cognition'', exploring the integration of generative AI with human cognition. The term "generative" reflects AI's ability to iteratively produce structured outputs, while "midtended" captures the potential hybrid (human-AI) nature of the process. It stands between traditional conceptions of intended creation, understood directed from within, and extended processes that bring exo-biological processes into the creative process. We examine current generative technologies (based on multimodal transformer architectures typical of large language models like ChatGPT), to explain how they can transform human cognitive agency beyond what standard theories of extended cognition can capture. We suggest that the type of cognitive activity typical of the coupling between a human and generative technologies is closer (but not equivalent) to social cognition than to classical extended cognitive paradigms. Yet, it deserves a specific treatment. We provide an explicit definition of generative midtended cognition in which we treat interventions by AI systems as constitutive of the agent's intentional creative processes. Furthermore, we distinguish two dimensions of generative hybrid creativity: 1. Width: captures the sensitivity of the context of the generative process (from the single letter to the whole historical and surrounding data), 2. Depth: captures the granularity of iteration loops involved in the process. Generative midtended cognition stands in the middle depth between conversational forms of cognition in which complete utterances or creative units are exchanged, and micro-cognitive (e.g. neural) subpersonal processes. Finally, the paper discusses the potential risks and benefits of widespread generative AI adoption, including the challenges of authenticity, generative power asymmetry, and creative boost or atrophy.

cs.AI↗

arXiv2022

Stabilizing Voltage in Power Distribution Networks via Multi-Agent Reinforcement Learning with Transformer

The increased integration of renewable energy poses a slew of technical challenges for the operation of power distribution networks. Among them, voltage fluctuations caused by the instability of renewable energy are receiving increasing attention. Utilizing MARL algorithms to coordinate multiple control units in the grid, which is able to handle rapid changes of power systems, has been widely studied in active voltage control task recently. However, existing approaches based on MARL ignore the unique nature of the grid and achieve limited performance. In this paper, we introduce the transformer architecture to extract representations adapting to power network problems and propose a Transformer-based Multi-Agent Actor-Critic framework (T-MAAC) to stabilize voltage in power distribution networks. In addition, we adopt a novel auxiliary-task training process tailored to the voltage control task, which improves the sample efficiency and facilitating the representation learning of the transformer-based model. We couple T-MAAC with different multi-agent actor-critic algorithms, and the consistent improvements on the active voltage control task demonstrate the effectiveness of the proposed method.

cs.MA↗

arXiv2025

Reactive Transformer (RxT) -- Stateful Real-Time Processing for Event-Driven Reactive Language Models

The Transformer architecture has become the de facto standard for Large Language Models (LLMs), demonstrating remarkable capabilities in language understanding and generation. However, its application in conversational AI is fundamentally constrained by its stateless nature and the quadratic computational complexity ($O(L^2)$) with respect to sequence length $L$. Current models emulate memory by reprocessing an ever-expanding conversation history with each turn, leading to prohibitive costs and latency in long dialogues. This paper introduces the Reactive Transformer (RxT), a novel architecture designed to overcome these limitations by shifting from a data-driven to an event-driven paradigm. RxT processes each conversational turn as a discrete event in real-time, maintaining context in an integrated, fixed-size Short-Term Memory (STM) system. The architecture features a distinct operational cycle where a generator-decoder produces a response based on the current query and the previous memory state, after which a memory-encoder and a dedicated Memory Attention network asynchronously update the STM with a representation of the complete interaction. This design fundamentally alters the scaling dynamics, reducing the total user-facing cost of a conversation from quadratic ($O(N^2 \cdot T)$) to linear ($O(N \cdot T)$) with respect to the number of interactions $N$. By decoupling response generation from memory updates, RxT achieves low latency, enabling truly real-time, stateful, and economically viable long-form conversations. We validated our architecture with a series of proof-of-concept experiments on synthetic data, demonstrating superior performance and constant-time inference latency compared to a baseline stateless model of comparable size.

cs.CL↗

arXiv2026

PolarSym: Polar Geometry-aware Attention for CAD Floorplan Parsing

CAD plan parsing is a fundamental task in Building Information Modeling (BIM), aiming to automatically extract architectural elements including walls, doors, windows, and furniture from 2D engineering drawings. Existing Transformer-based methods capture global semantic dependencies via self-attention, yet they infer spatial relationships merely from semantic features without explicitly characterizing the intrinsic geometric symmetry of building layouts. Such methods tend to produce mismatched correspondences in long-range matching and complex symmetric spatial layouts. To tackle this limitation, we propose PolarSym, a polar-coordinate geometry-aware attention framework for CAD plan parsing. The framework decouples geometric relationships of buildings into two complementary components, direction and distance, which are modeled independently. Structural consistency is strengthened by directional constraints, while long-range symmetric correspondences are built with distance constraints. A dynamic gating mechanism is adopted to synergistically fuse the two geometric information branches while maintaining the vanilla Transformer architecture. This design boosts geometric modeling capacity with negligible extra computation. Experiments on a public CAD plan parsing dataset show that PolarSym surpasses the reproduced SymPoint V2 baseline by 1.73% PQ, 1.54% RQ and 4.31% mIoU under identical training settings. PolarSym also converges faster and yields more stable optimization. Ablation experiments verify the complementary effects of direction and distance modeling. Our results reveal that PolarSym improves the geometric awareness of Transformers at low computational cost, offering an effective geometric modeling paradigm for CAD plan parsing.

cs.CV↗

arXiv2026

Sharper Generalization Bounds for Transformer

This paper studies generalization error bounds for Transformer models. Based on the offset Rademacher complexity, we derive sharper generalization bounds for different Transformer architectures, including single-layer single-head, single-layer multi-head, and multi-layer Transformers. We first express the excess risk of Transformers in terms of the offset Rademacher complexity. By exploiting its connection with the empirical covering numbers of the corresponding hypothesis spaces, we obtain excess risk bounds that achieve optimal convergence rates up to constant factors. We then derive refined excess risk bounds by upper bounding the covering numbers of Transformer hypothesis spaces using matrix ranks and matrix norms, leading to precise, architecture-dependent generalization bounds. Finally, we relax the boundedness assumption on feature mappings and extend our theoretical results to settings with unbounded (sub-Gaussian) features and heavy-tailed distributions.

cs.LG↗

arXiv2026

Fusing Memory and Attention: A study on LSTM, Transformer and Hybrid Architectures for Symbolic Music Generation

Machine learning techniques, such as Transformers and Long Short-Term Memory (LSTM) networks, play a crucial role in Symbolic Music Generation (SMG). Existing literature indicates a difference between LSTMs and Transformers regarding their ability to model local melodic continuity versus maintaining global structural coherence. However, their specific properties within the context of SMG have not been systematically studied. This paper addresses this gap by providing a fine-grained comparative analysis of LSTMs versus Transformers for SMG, examining local and global properties in detail using 17 musical quality metrics on the Deutschl dataset. We find that LSTM networks excel at capturing local patterns but fail to preserve long-range dependencies, while Transformers model global structure effectively but tend to produce irregular phrasing. Based on this analysis and leveraging their respective strengths, we propose a Hybrid architecture combining a Transformer Encoder with an LSTM Decoder and evaluate it against both baselines. We evaluated 1,000 generated melodies from each of the three architectures on the Deutschl dataset. The results show that the hybrid method achieves better local and global continuity and coherence compared to the baselines. Our work highlights the key characteristics of these models and demonstrates how their properties can be leveraged to design superior models. We also supported the experiments with ablation studies and human perceptual evaluations, which statistically support the findings and provide robust validation for this work.

cs.LG↗

arXiv2024

Optimizing Robotic Manipulation with Decision-RWKV: A Recurrent Sequence Modeling Approach for Lifelong Learning

Models based on the Transformer architecture have seen widespread application across fields such as natural language processing, computer vision, and robotics, with large language models like ChatGPT revolutionizing machine understanding of human language and demonstrating impressive memory and reproduction capabilities. Traditional machine learning algorithms struggle with catastrophic forgetting, which is detrimental to the diverse and generalized abilities required for robotic deployment. This paper investigates the Receptance Weighted Key Value (RWKV) framework, known for its advanced capabilities in efficient and effective sequence modeling, and its integration with the decision transformer and experience replay architectures. It focuses on potential performance enhancements in sequence decision-making and lifelong robotic learning tasks. We introduce the Decision-RWKV (DRWKV) model and conduct extensive experiments using the D4RL database within the OpenAI Gym environment and on the D'Claw platform to assess the DRWKV model's performance in single-task tests and lifelong learning scenarios, showcasing its ability to handle multiple subtasks efficiently. The code for all algorithms, training, and image rendering in this study is open-sourced at https://github.com/ancorasir/DecisionRWKV.

cs.RO↗

arXiv2025

Scaling White-Box Transformers for Vision

CRATE, a white-box transformer architecture designed to learn compressed and sparse representations, offers an intriguing alternative to standard vision transformers (ViTs) due to its inherent mathematical interpretability. Despite extensive investigations into the scaling behaviors of language and vision transformers, the scalability of CRATE remains an open question which this paper aims to address. Specifically, we propose CRATE-$α$, featuring strategic yet minimal modifications to the sparse coding block in the CRATE architecture design, and a light training recipe designed to improve the scalability of CRATE. Through extensive experiments, we demonstrate that CRATE-$α$ can effectively scale with larger model sizes and datasets. For example, our CRATE-$α$-B substantially outperforms the prior best CRATE-B model accuracy on ImageNet classification by 3.7%, achieving an accuracy of 83.2%. Meanwhile, when scaling further, our CRATE-$α$-L obtains an ImageNet classification accuracy of 85.1%. More notably, these model performance improvements are achieved while preserving, and potentially even enhancing the interpretability of learned CRATE models, as we demonstrate through showing that the learned token representations of increasingly larger trained CRATE-$α$ models yield increasingly higher-quality unsupervised object segmentation of images. The project page is https://rayjryang.github.io/CRATE-alpha/.

cs.CV↗

arXiv2024

Evolution Transformer: In-Context Evolutionary Optimization

Evolutionary optimization algorithms are often derived from loose biological analogies and struggle to leverage information obtained during the sequential course of optimization. An alternative promising approach is to leverage data and directly discover powerful optimization principles via meta-optimization. In this work, we follow such a paradigm and introduce Evolution Transformer, a causal Transformer architecture, which can flexibly characterize a family of Evolution Strategies. Given a trajectory of evaluations and search distribution statistics, Evolution Transformer outputs a performance-improving update to the search distribution. The architecture imposes a set of suitable inductive biases, i.e. the invariance of the distribution update to the order of population members within a generation and equivariance to the order of the search dimensions. We train the model weights using Evolutionary Algorithm Distillation, a technique for supervised optimization of sequence models using teacher algorithm trajectories. The resulting model exhibits strong in-context optimization performance and shows strong generalization capabilities to otherwise challenging neuroevolution tasks. We analyze the resulting properties of the Evolution Transformer and propose a technique to fully self-referentially train the Evolution Transformer, starting from a random initialization and bootstrapping its own learning progress. We provide an open source implementation under https://github.com/RobertTLange/evosax.

cs.AI↗

arXiv2024

WaveMamba: Spatial-Spectral Wavelet Mamba for Hyperspectral Image Classification

Hyperspectral Imaging (HSI) has proven to be a powerful tool for capturing detailed spectral and spatial information across diverse applications. Despite the advancements in Deep Learning (DL) and Transformer architectures for HSI classification, challenges such as computational efficiency and the need for extensive labeled data persist. This paper introduces WaveMamba, a novel approach that integrates wavelet transformation with the spatial-spectral Mamba architecture to enhance HSI classification. WaveMamba captures both local texture patterns and global contextual relationships in an end-to-end trainable model. The Wavelet-based enhanced features are then processed through the state-space architecture to model spatial-spectral relationships and temporal dependencies. The experimental results indicate that WaveMamba surpasses existing models, achieving an accuracy improvement of 4.5\% on the University of Houston dataset and a 2.0\% increase on the Pavia University dataset.

cs.CV↗

arXiv2024

Benchmarking Image Transformers for Prostate Cancer Detection from Ultrasound Data

PURPOSE: Deep learning methods for classifying prostate cancer (PCa) in ultrasound images typically employ convolutional networks (CNNs) to detect cancer in small regions of interest (ROI) along a needle trace region. However, this approach suffers from weak labelling, since the ground-truth histopathology labels do not describe the properties of individual ROIs. Recently, multi-scale approaches have sought to mitigate this issue by combining the context awareness of transformers with a CNN feature extractor to detect cancer from multiple ROIs using multiple-instance learning (MIL). In this work, we present a detailed study of several image transformer architectures for both ROI-scale and multi-scale classification, and a comparison of the performance of CNNs and transformers for ultrasound-based prostate cancer classification. We also design a novel multi-objective learning strategy that combines both ROI and core predictions to further mitigate label noise. METHODS: We evaluate 3 image transformers on ROI-scale cancer classification, then use the strongest model to tune a multi-scale classifier with MIL. We train our MIL models using our novel multi-objective learning strategy and compare our results to existing baselines. RESULTS: We find that for both ROI-scale and multi-scale PCa detection, image transformer backbones lag behind their CNN counterparts. This deficit in performance is even more noticeable for larger models. When using multi-objective learning, we can improve performance of MIL, with a 77.9% AUROC, a sensitivity of 75.9%, and a specificity of 66.3%. CONCLUSION: Convolutional networks are better suited for modelling sparse datasets of prostate ultrasounds, producing more robust features than transformers in PCa detection. Multi-scale methods remain the best architecture for this task, with multi-objective learning presenting an effective way to improve performance.

eess.IV↗

arXiv2025

Sonnet: Spectral Operator Neural Network for Multivariable Time Series Forecasting

Multivariable time series forecasting methods can integrate information from exogenous variables, leading to significant prediction accuracy gains. The transformer architecture has been widely applied in various time series forecasting models due to its ability to capture long-range sequential dependencies. However, a naïve application of transformers often struggles to effectively model complex relationships among variables over time. To mitigate against this, we propose a novel architecture, termed Spectral Operator Neural Network (Sonnet). Sonnet applies learnable wavelet transformations to the input and incorporates spectral analysis using the Koopman operator. Its predictive skill relies on the Multivariable Coherence Attention (MVCA), an operation that leverages spectral coherence to model variable dependencies. Our empirical analysis shows that Sonnet yields the best performance on $34$ out of $47$ forecasting tasks with an average mean absolute error (MAE) reduction of $2.2\%$ against the most competitive baseline. We further show that MVCA can remedy the deficiencies of naïve attention in various deep learning models, reducing MAE by $10.7\%$ on average in the most challenging forecasting tasks.

cs.LG↗

arXiv2025

Bridging the Dimensional Chasm: Uncover Layer-wise Dimensional Reduction in Transformers through Token Correlation

The geometric evolution of token representations in large language models (LLMs) presents a fundamental paradox: while human language inherently organizes semantic information in low-dimensional spaces ($\sim 10^1$ dimensions), modern LLMs employ high-dimensional embeddings ($\sim 10^3$ dimensions) processed through Transformer architectures. To resolve this paradox, this work bridges this conceptual gap by developing a geometric framework that tracks token dynamics across Transformers layers. Through layer-wise analysis of intrinsic dimensions across multiple architectures, we reveal an expansion-contraction pattern where tokens diffuse to a "working space" and then progressively project onto lower-dimensional submanifolds. Our finding implies a negative correlation between the working space dimension and parameter-sensitive performance of the LLMs, and indicates that effective models tend to compress tokens into approximately 10-dimensional submanifolds, closely resembling human semantic spaces. This work not only advances LLM interpretability by reframing Transformers layers as projectors that mediate between high-dimensional computation and low-dimensional semantics, but also provides practical tools for model diagnostics that do not rely on task-specific evaluations.

cs.CL↗

arXiv2026

Towards Understanding What State Space Models Learn About Code

State Space Models (SSMs) have emerged as an efficient alternative to the Transformer architecture. Prior work shows that, when trained under comparable conditions, SSMs can match or surpass Transformers on code understanding tasks. However, their internal mechanisms remain a black box. We present the first systematic analysis of what SSM-based code models learn along with the direct comparison between SSM and Transformer models in this domain. Our analysis shows that SSMs capture syntactic and semantic structure more effectively than Transformers during pretraining but forgets certain relations during fine-tuning on some tasks. To investigate this behavior, we introduce SSM-Interpret, a frequency-domain framework that exposes a spectral shift toward short-range dependencies during fine-tuning. Guided by these findings, we propose architectural modifications that significantly improve the performance of SSM-based code model by upto +6 MRR on NLCodeSearch. This demonstrates that our analysis not only explains model behavior but also leads directly to better designs.

cs.AI↗

arXiv2019

Character-based NMT with Transformer

Character-based translation has several appealing advantages, but its performance is in general worse than a carefully tuned BPE baseline. In this paper we study the impact of character-based input and output with the Transformer architecture. In particular, our experiments on EN-DE show that character-based Transformer models are more robust than their BPE counterpart, both when translating noisy text, and when translating text from a different domain. To obtain comparable BLEU scores in clean, in-domain data and close the gap with BPE-based models we use known techniques to train deeper Transformer models.

cs.CL↗

arXiv2020

The Chess Transformer: Mastering Play using Generative Language Models

This work demonstrates that natural language transformers can support more generic strategic modeling, particularly for text-archived games. In addition to learning natural language skills, the abstract transformer architecture can generate meaningful moves on a chessboard. With further fine-tuning, the transformer learns complex gameplay by training on 2.8 million chess games in Portable Game Notation. After 30,000 training steps, OpenAI's Generative Pre-trained Transformer (GPT-2) optimizes weights for 774 million parameters. This fine-tuned Chess Transformer generates plausible strategies and displays game formations identifiable as classic openings, such as English or the Slav Exchange. Finally, in live play, the novel model demonstrates a human-to-transformer interface that correctly filters illegal moves and provides a novel method to challenge the transformer's chess strategies. We anticipate future work will build on this transformer's promise, particularly in other strategy games where features can capture the underlying complex rule syntax from simple but expressive player annotations.

cs.AI↗

arXiv2022

SwinTrack: A Simple and Strong Baseline for Transformer Tracking

Recently Transformer has been largely explored in tracking and shown state-of-the-art (SOTA) performance. However, existing efforts mainly focus on fusing and enhancing features generated by convolutional neural networks (CNNs). The potential of Transformer in representation learning remains under-explored. In this paper, we aim to further unleash the power of Transformer by proposing a simple yet efficient fully-attentional tracker, dubbed SwinTrack, within classic Siamese framework. In particular, both representation learning and feature fusion in SwinTrack leverage the Transformer architecture, enabling better feature interactions for tracking than pure CNN or hybrid CNN-Transformer frameworks. Besides, to further enhance robustness, we present a novel motion token that embeds historical target trajectory to improve tracking by providing temporal context. Our motion token is lightweight with negligible computation but brings clear gains. In our thorough experiments, SwinTrack exceeds existing approaches on multiple benchmarks. Particularly, on the challenging LaSOT, SwinTrack sets a new record with 0.713 SUC score. It also achieves SOTA results on other benchmarks. We expect SwinTrack to serve as a solid baseline for Transformer tracking and facilitate future research. Our codes and results are released at https://github.com/LitingLin/SwinTrack.

cs.CV↗