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 1,747 records · Page 97Linked to original sources

arXiv2025

Adaptive parameters identification for nonlinear dynamics using deep permutation invariant networks

The promising outcomes of dynamical system identification techniques, such as SINDy [Brunton et al. 2016], highlight their advantages in providing qualitative interpretability and extrapolation compared to non-interpretable deep neural networks [Rudin 2019]. These techniques suffer from parameter updating in real-time use cases, especially when the system parameters are likely to change during or between processes. Recently, the OASIS [Bhadriraju et al. 2020] framework introduced a data-driven technique to address the limitations of real-time dynamical system parameters updating, yielding interesting results. Nevertheless, we show in this work that superior performance can be achieved using more advanced model architectures. We present an innovative encoding approach, based mainly on the use of Set Encoding methods of sequence data, which give accurate adaptive model identification for complex dynamic systems, with variable input time series length. Two Set Encoding methods are used, the first is Deep Set [Zaheer et al. 2017], and the second is Set Transformer [Lee et al. 2019]. Comparing Set Transformer to OASIS framework on Lotka Volterra for real-time local dynamical system identification and time series forecasting, we find that the Set Transformer architecture is well adapted to learning relationships within data sets. We then compare the two Set Encoding methods based on the Lorenz system for online global dynamical system identification. Finally, we trained a Deep Set model to perform identification and characterization of abnormalities for 1D heat-transfer problem.

cs.LG↗

arXiv2024

Efficient transformer with reinforced position embedding for language models

In this paper, we propose an efficient transformer architecture that uses reinforced positional embedding to obtain superior performance with half the number of encoder decoder layers. We demonstrate that concatenating positional encoding with trainable token embeddings, normalizing columns in the token embedding matrix, and using the normalized token embedding matrix as the value of the attention layer improve the training and validation loss and the training time in an encoder-decoder Transformer model for a Portuguese-English translation task with 10 epochs or 12 hours of training across 10 trials. Our method, with roughly a threefold parameter reduction compared to the baseline model, yields a mean training loss of 1.21, a mean validation loss of 1.51, and an average training time of 1352.27 seconds per epoch, surpassing the baseline model with the same embedding dimension that employs addition of positional encoding and token embeddings, which achieves a mean training loss of 1.96, a validation loss of 2.18, and an average training time of 4297.79 seconds per epoch. Additionally, we evaluated our proposed architecture and the baseline across 14 diverse translation datasets from TensorFlow. The results indicate that our method consistently achieves lower or comparable training and validation losses, suggesting enhanced learning efficiency.

cs.CL↗

arXiv2026

MambaVoice: Lightweight Audiovisual Singing Voice Separation Via A Hybrid Mamba-Transformer Model

Isolating a target singing voice from a music video remains challenging, particularly in the presence of multiple vocalists and dense instrumental accompaniment. We propose MambaVoice, a lightweight audiovisual framework that leverages a hybrid Mamba--Transformer architecture for targeted singing voice separation. The model jointly encodes audio and visual streams using an attention-based band-split audio encoder and a spatio-temporal graph convolutional network (ST-GCN) for facial motion features. These modalities are fused through a multiplicative gating mechanism, enabling visual cues to selectively modulate audio representations. The fused features are processed by a hybrid backbone that combines Transformer self-attention with Selective State Space Models (SSMs), achieving efficient long-range temporal modeling with linear complexity. We evaluated MambaVoice on the Acappella and URSing datasets under challenging conditions, including mixtures with interfering singers. At 16.2 million parameters, the model demonstrates comparable performance, achieving 14.18 dB SDR on Acappella and strong cross-dataset performance on URSing, comparable to larger models at a fraction of the parameter count. These findings highlight the effectiveness of hybrid SSM--attention architectures for scalable, efficient audiovisual source separation, suggesting they are well-suited as lightweight components within larger pipelines. We conduct a perceptual study that further supports our improvements in objective metrics. We provide our implementation online.

cs.SD↗

arXiv2026

Dual-Locking Learned AI Models: A PIN-Based Sparse QIM Watermarking and Adaptive Index Permutation Approach

We present a dual-locking method for securing trained neural networks that combines key-driven index permutation with PIN-based watermarking based on Sparse Quantization Index Modulation (QIM). Cryptographic randomness is introduced by independently applying a uniform random permutation to each row of adaptively selected index vectors. A robust blind binary watermark is then embedded into the bias coefficients by modulating their quantized values, binding the network to a user-defined Personal Identification Number (PIN). Without the correct key, the network retains its architecture but becomes functionally impaired due to disrupted internal representations. Inverse permutation fully restores the original model accuracy, while the embedded watermark remains imperceptible and enables blind verification of key association and model authorship. To improve both locking effectiveness and recoverability, an adaptive key selection strategy redistributes high-magnitude weights to low-sensitivity positions and vice versa, increasing degradation in the locked state while preserving full recovery. Experiments on MNIST, CIFAR-10/100, and ImageNet-1K using fully connected networks, ResNet CNNs, and transformer architectures show that locking reduces accuracy below 10\%, and even below 0.5\% for CNNs, while the correct key fully restores performance. The watermark introduces no measurable accuracy degradation and reliably authenticates ownership. Analysis of embedding distributions across CNNs and transformers further indicates potential diagnostic value for identifying undertrained or suboptimally designed models. The proposed approach therefore provides simultaneous model protection, recovery, and ownership verification.

cs.CR↗

arXiv2019

Do Transformer Attention Heads Provide Transparency in Abstractive Summarization?

Learning algorithms become more powerful, often at the cost of increased complexity. In response, the demand for algorithms to be transparent is growing. In NLP tasks, attention distributions learned by attention-based deep learning models are used to gain insights in the models' behavior. To which extent is this perspective valid for all NLP tasks? We investigate whether distributions calculated by different attention heads in a transformer architecture can be used to improve transparency in the task of abstractive summarization. To this end, we present both a qualitative and quantitative analysis to investigate the behavior of the attention heads. We show that some attention heads indeed specialize towards syntactically and semantically distinct input. We propose an approach to evaluate to which extent the Transformer model relies on specifically learned attention distributions. We also discuss what this implies for using attention distributions as a means of transparency.

cs.CL↗

arXiv2023

Character-level NMT and language similarity

We explore the effectiveness of character-level neural machine translation using Transformer architecture for various levels of language similarity and size of the training dataset on translation between Czech and Croatian, German, Hungarian, Slovak, and Spanish. We evaluate the models using automatic MT metrics and show that translation between similar languages benefits from character-level input segmentation, while for less related languages, character-level vanilla Transformer-base often lags behind subword-level segmentation. We confirm previous findings that it is possible to close the gap by finetuning the already trained subword-level models to character-level.

cs.CL↗

arXiv2023

Converting Transformers to Polynomial Form for Secure Inference Over Homomorphic Encryption

Designing privacy-preserving deep learning models is a major challenge within the deep learning community. Homomorphic Encryption (HE) has emerged as one of the most promising approaches in this realm, enabling the decoupling of knowledge between the model owner and the data owner. Despite extensive research and application of this technology, primarily in convolutional neural networks, incorporating HE into transformer models has been challenging because of the difficulties in converting these models into a polynomial form. We break new ground by introducing the first polynomial transformer, providing the first demonstration of secure inference over HE with transformers. This includes a transformer architecture tailored for HE, alongside a novel method for converting operators to their polynomial equivalent. This innovation enables us to perform secure inference on LMs with WikiText-103. It also allows us to perform image classification with CIFAR-100 and Tiny-ImageNet. Our models yield results comparable to traditional methods, bridging the performance gap with transformers of similar scale and underscoring the viability of HE for state-of-the-art applications. Finally, we assess the stability of our models and conduct a series of ablations to quantify the contribution of each model component.

cs.LG↗

arXiv2024

Transformer-based Drum-level Prediction in a Boiler Plant with Delayed Relations among Multivariates

The steam drum water level is a critical parameter that directly impacts the safety and efficiency of power plant operations. However, predicting the drum water level in boilers is challenging due to complex non-linear process dynamics originating from long-time delays and interrelations, as well as measurement noise. This paper investigates the application of Transformer-based models for predicting drum water levels in a steam boiler plant. Leveraging the capabilities of Transformer architectures, this study aims to develop an accurate and robust predictive framework to anticipate water level fluctuations and facilitate proactive control strategies. To this end, a prudent pipeline is proposed, including 1) data preprocess, 2) causal relation analysis, 3) delay inference, 4) variable augmentation, and 5) prediction. Through extensive experimentation and analysis, the effectiveness of Transformer-based approaches in steam drum water level prediction is evaluated, highlighting their potential to enhance operational stability and optimize plant performance.

cs.LG↗

arXiv2024

Low Latency Transformer Inference on FPGAs for Physics Applications with hls4ml

This study presents an efficient implementation of transformer architectures in Field-Programmable Gate Arrays(FPGAs) using hls4ml. We demonstrate the strategy for implementing the multi-head attention, softmax, and normalization layer and evaluate three distinct models. Their deployment on VU13P FPGA chip achieved latency less than 2us, demonstrating the potential for real-time applications. HLS4ML compatibility with any TensorFlow-built transformer model further enhances the scalability and applicability of this work. Index Terms: FPGAs, machine learning, transformers, high energy physics, LIGO

cs.LG↗

arXiv2024

Expanding Expressivity in Transformer Models with MöbiusAttention

Attention mechanisms and Transformer architectures have revolutionized Natural Language Processing (NLP) by enabling exceptional modeling of long-range dependencies and capturing intricate linguistic patterns. However, their inherent reliance on linear operations in the form of matrix multiplications limits their ability to fully capture inter-token relationships on their own. We propose MöbiusAttention, a novel approach that integrates Möbius transformations within the attention mechanism of Transformer-based models. Möbius transformations are non-linear operations in spaces over complex numbers with the ability to map between various geometries. By incorporating these properties, MöbiusAttention empowers models to learn more intricate geometric relationships between tokens and capture a wider range of information through complex-valued weight vectors. We build and pre-train a BERT and a RoFormer version enhanced with MöbiusAttention, which we then finetune on the GLUE benchmark. We evaluate empirically our approach against the baseline BERT and RoFormer models on a range of downstream tasks. Our approach compares favorably against the baseline models, even with smaller number of parameters suggesting the enhanced expressivity of MöbiusAttention. This research paves the way for exploring the potential of Möbius transformations in the complex projective space to enhance the expressivity and performance of foundation models.

cs.LG↗

arXiv2025

Revisiting Funnel Transformers for Modern LLM Architectures with Comprehensive Ablations in Training and Inference Configurations

Transformer-based Large Language Models, which suffer from high computational costs, advance so quickly that techniques proposed to streamline earlier iterations are not guaranteed to benefit more modern models. Building upon the Funnel Transformer proposed by Dai and Le (2020), which progressively compresses intermediate representations, we investigate the impact of funneling in contemporary Gemma2 Transformer architectures. We systematically evaluate various funnel configurations and recovery methods, comparing: (1) standard pretraining to funnel-aware pretraining strategies, (2) the impact of funnel-aware fine-tuning, and (3) the type of sequence recovery operation. Our results demonstrate that funneling creates information bottlenecks that propagate through deeper network layers, particularly in larger models (e.g., Gemma 7B), leading to at times unmanageable performance lost. However, carefully selecting the funneling layer and employing effective recovery strategies, can substantially mitigate performance losses, achieving up to a 44\% reduction in latency. Our findings highlight key trade-offs between computational efficiency and model accuracy, providing practical guidance for deploying funnel-based approaches in large-scale natural language applications.

cs.CL↗

arXiv2026

Orthogonal Self-Attention

Softmax Self-Attention (SSA) is a key component of Transformer architectures. However, when utilised within skipless architectures, which aim to improve representation learning, recent work has highlighted the inherent instability of SSA due to inducing rank collapse and poorly-conditioned Jacobians. In this work, we design a novel attention mechanism: Orthogonal Self-Attention (OSA), which aims to bypass these issues with SSA, in order to allow for (non-causal) Transformers without skip connections and normalisation layers to be more easily trained. In particular, OSA parametrises the attention matrix to be orthogonal via mapping a skew-symmetric matrix, formed from query-key values, through the matrix exponential. We show that this can be practically implemented, by exploiting the low-rank structure of our query-key values, resulting in the computational complexity and memory cost of OSA scaling linearly with sequence length. Furthermore, we derive an initialisation scheme for which we prove ensures that the Jacobian of OSA is well-conditioned.

cs.LG↗

arXiv2024

Stochastic Spiking Attention: Accelerating Attention with Stochastic Computing in Spiking Networks

Spiking Neural Networks (SNNs) have been recently integrated into Transformer architectures due to their potential to reduce computational demands and to improve power efficiency. Yet, the implementation of the attention mechanism using spiking signals on general-purpose computing platforms remains inefficient. In this paper, we propose a novel framework leveraging stochastic computing (SC) to effectively execute the dot-product attention for SNN-based Transformers. We demonstrate that our approach can achieve high classification accuracy ($83.53\%$) on CIFAR-10 within 10 time steps, which is comparable to the performance of a baseline artificial neural network implementation ($83.66\%$). We estimate that the proposed SC approach can lead to over $6.3\times$ reduction in computing energy and $1.7\times$ reduction in memory access costs for a digital CMOS-based ASIC design. We experimentally validate our stochastic attention block design through an FPGA implementation, which is shown to achieve $48\times$ lower latency as compared to a GPU implementation, while consuming $15\times$ less power.

cs.AR↗

arXiv2026

Post-LayerNorm Is Back: Stable, ExpressivE, and Deep

Large language model (LLM) scaling is hitting a wall. Widening models yields diminishing returns, and extending context length does not improve fundamental expressivity. In contrast, depth scaling offers theoretically superior expressivity, yet current Transformer architectures struggle to train reliably at extreme depths. We revisit the Post-LayerNorm (Post-LN) formulation, whose instability at scale caused its replacement by Pre-LN in modern LLMs. We show that the central failure mode of Post-LN arises from the ResNet-style residual pathway, which introduces gradient vanishing in deep networks. We present Keel, a Post-LN Transformer that replaces this residual path with a Highway-style connection. This modification preserves the gradient flow through the residual branch, preventing signal vanishing from the top layers to the bottom. Unlike prior methods, Keel enables stable training at extreme depths without requiring specialized initialization or complex optimization tricks. Keel trains robustly at depths exceeding 1000 layers and consistently improves perplexity and depth-scaling characteristics over Pre-LN. These findings indicate that Post-LN, when paired with a Highway-style connection, provides a simple and effective foundation for building deeply scalable LLMs, opening the possibility for future infinite-depth architectures.

cs.LG↗

arXiv2024

Just a Simple Transformation is Enough for Data Protection in Vertical Federated Learning

Vertical Federated Learning (VFL) aims to enable collaborative training of deep learning models while maintaining privacy protection. However, the VFL procedure still has components that are vulnerable to attacks by malicious parties. In our work, we consider feature reconstruction attacks, a common risk targeting input data compromise. We theoretically claim that feature reconstruction attacks cannot succeed without knowledge of the prior distribution on data. Consequently, we demonstrate that even simple model architecture transformations can significantly impact the protection of input data during VFL. Confirming these findings with experimental results, we show that MLP-based models are resistant to state-of-the-art feature reconstruction attacks.

cs.LG↗

arXiv2026

Same Architecture, Different Capacity: Optimizer-Induced Spectral Scaling Laws

Scaling laws have made language-model performance predictable from model size, data, and compute, but they typically treat the optimizer as a fixed training detail. We show that this assumption misses a fundamental axis of representation scaling: how effectively the optimizer converts added FFN width into utilized spectral capacity. Using eigenspectra of feed-forward network representations, measured through soft and hard spectral-ranks, we find that \emph{the same Transformer architecture realizes markedly different spectral scaling laws when trained with different optimizers}. Holding architecture and width schedule fixed, AdamW exhibits weak hard-rank scaling ($β$=0.44) on rare-token (TAIL) representations where learning is known to be hardest, whereas Muon achieves linear scaling ($β$=1.02) in the same regimes, a $2.3\times$ increase in the scaling exponent. This difference is not reducible to validation loss: AdamW configurations can match low-rank Dion variants in perplexity, under extended training, while exhibiting sharply different spectral geometry, demonstrating that matched loss does not imply matched representation structure. Hard--soft rank asymmetry further reveals that optimizers differ not only in how much capacity is realized, but also in how that capacity is structured across eigenmodes. To disentangle optimizer effects from architectural ones, we compare against architectural interventions (e.g., attention rank and positional encoding), and find that optimizer-induced spectral shifts often exceed the architectural effects. These results suggest optimization as a first-class axis of representation scaling, motivating optimizer--architecture co-design.

cs.LG↗

arXiv2025

Long-context Protein Language Modeling Using Bidirectional Mamba with Shared Projection Layers

Self-supervised training of language models (LMs) has seen great success for protein sequences in learning meaningful representations and for generative drug design. Most protein LMs are based on the Transformer architecture trained on individual proteins with short context lengths. Such protein LMs cannot extrapolate to longer proteins and protein complexes well. They also fail to account for the underlying biological mechanisms carried out by biomolecular interactions and dynamics i.e., proteins often interact with other proteins, molecules, and pathways in complex biological systems. In this work, we propose LC-PLM based on an alternative protein LM architecture, BiMamba-S, built upon selective structured state-space models, to learn high-quality universal protein representations at the amino acid token level using masked language modeling. We also introduce its graph-contextual variant, LC-PLM, which contextualizes protein-protein interaction (PPI) graphs for a second stage of training. LC-PLM demonstrates favorable neural scaling laws, better length extrapolation capability, and up to 30% and 16% improvements on protein downstream tasks compared to Transformer-based ESM-2 when trained with 100B and 1T tokens, respectively. LC-PLM-G further trained within the context of PPI graphs shows promising results on protein structure and function prediction tasks. Our study demonstrates the benefit of increasing the context size with computationally efficient LM architecture (e.g., structured state space models) in learning universal protein representations and incorporating molecular interaction contexts contained in biological graphs.

q-bio.BM↗

arXiv2021

DBIA: Data-free Backdoor Injection Attack against Transformer Networks

Recently, transformer architecture has demonstrated its significance in both Natural Language Processing (NLP) and Computer Vision (CV) tasks. Though other network models are known to be vulnerable to the backdoor attack, which embeds triggers in the model and controls the model behavior when the triggers are presented, little is known whether such an attack is still valid on the transformer models and if so, whether it can be done in a more cost-efficient manner. In this paper, we propose DBIA, a novel data-free backdoor attack against the CV-oriented transformer networks, leveraging the inherent attention mechanism of transformers to generate triggers and injecting the backdoor using the poisoned surrogate dataset. We conducted extensive experiments based on three benchmark transformers, i.e., ViT, DeiT and Swin Transformer, on two mainstream image classification tasks, i.e., CIFAR10 and ImageNet. The evaluation results demonstrate that, consuming fewer resources, our approach can embed backdoors with a high success rate and a low impact on the performance of the victim transformers. Our code is available at https://anonymous.4open.science/r/DBIA-825D.

cs.CV↗