We consider a problem of statistical mean estimation in which the samples are not observed directly, but are instead observed by a relay (``teacher'') that transmits information through a memoryless channel to the decoder (``student''), who then produces the final estimate. We consider the minimax estimation error in the large deviations regime, and establish achievable error exponents that are tight in broad regimes of the estimation accuracy and channel quality. In contrast, two natural baseline methods are shown to yield strictly suboptimal error exponents. We initially focus on Bernoulli sources and binary symmetric channels, and then generalize to sub-Gaussian and heavy-tailed settings along with arbitrary discrete memoryless channels.
State-of-the-art spatio-temporal action detection (STAD) methods show promising results for extracting soccer events from broadcast videos. However, when operated in the high-recall, low-precision regime required for exhaustive event coverage in soccer analytics, their lack of contextual understanding becomes apparent: many false positives could be resolved by considering a broader sequence of actions and game-state information. In this work, we address this limitation by reasoning at the game level and improving STAD through the addition of a denoising sequence transduction task. Sequences of noisy, context-free player-centric predictions are processed alongside clean game state information using a Transformer-based encoder-decoder model. By modeling extended temporal context and reasoning jointly over team-level dynamics, our method leverages the "language of soccer" - its tactical regularities and inter-player dependencies - to generate "denoised" sequences of actions. This approach improves both precision and recall in low-confidence regimes, enabling more reliable event extraction from broadcast video and complementing existing pixel-based methods.
In recent years, diffusion model has shown its potential across diverse domains from vision generation to language modeling. Transferring its capabilities to modern autonomous driving systems has also emerged as a promising direction.In this work, we propose TransDiffuser, an encoder-decoder based generative trajectory planning model for end-to-end autonomous driving. The encoded scene information serves as the multi-modal conditional input of the denoising decoder. To tackle the mode collapse dilemma in generating high-quality diverse trajectories, we introduce a simple yet effective multi-modal representation decorrelation optimization mechanism during the training process.TransDiffuser achieves PDMS of 94.85 on the NAVSIM benchmark, surpassing previous state-of-the-art methods without any anchor-based prior trajectories.
Unmanned Aerial Vehicles (UAVs) are increasingly used for reforestation and forest monitoring, including seed dispersal in hard-to-reach terrains. However, a detailed understanding of the forest floor remains a challenge due to high natural variability, quickly changing environmental parameters, and ambiguous annotations due to unclear definitions. To address this issue, we adapt the Segment Anything Model (SAM), a vision foundation model with strong generalization capabilities, to segment forest floor objects such as tree stumps, vegetation, and woody debris. To this end, we employ parameter-efficient fine-tuning (PEFT) to fine-tune a small subset of additional model parameters while keeping the original weights fixed. We adjust SAM's mask decoder to generate masks corresponding to our dataset categories, allowing for automatic segmentation without manual prompting. Our results show that the adapter-based PEFT method achieves the highest mean intersection over union (mIoU), while Low-rank Adaptation (LoRA), with fewer parameters, offers a lightweight alternative for resource-constrained UAV platforms.
Cloud-based multilingual translation services like Google Translate and Microsoft Translator achieve state-of-the-art translation capabilities. These services inherently use large multilingual language models such as GRU, LSTM, BERT, GPT, T5, or similar encoder-decoder architectures with attention mechanisms as the backbone. Also, new age natural language systems, for instance ChatGPT and DeepSeek, have established huge potential in multiple tasks in natural language processing. At the same time, they also possess outstanding multilingual translation capabilities. However, these models use the classical computing realm as a backend. QEDACVC (Quantum Encoder Decoder Attention-based Convolutional Variational Circuits) is an alternate solution that explores the quantum computing realm instead of the classical computing realm to study and demonstrate multilingual machine translation. QEDACVC introduces the quantum encoder-decoder architecture that simulates and runs on quantum computing hardware via quantum convolution, quantum pooling, quantum variational circuit, and quantum attention as software alterations. QEDACVC achieves an Accuracy of 82% when trained on the OPUS dataset for English, French, German, and Hindi corpora for multilingual translations.
Current applications of self-supervised learning to wireless channel representation often borrow paradigms developed for text and image processing, without fully addressing the unique characteristics and constraints of wireless communications. Aiming to fill this gap, we first propose WiMAE (Wireless Masked Autoencoder), a transformer-based encoder-decoder foundation model pretrained on a realistic open-source multi-antenna wireless channel dataset. Building upon this foundation, we develop ContraWiMAE, which enhances WiMAE by incorporating a contrastive learning objective alongside the reconstruction task in a unified multi-task framework. By warm-starting from pretrained WiMAE weights and generating positive pairs via noise injection, the contrastive component enables the model to capture both structural and discriminative features, enhancing representation quality beyond what reconstruction alone can achieve. Through extensive evaluation on unseen scenarios, we demonstrate the effectiveness of both approaches across multiple downstream tasks, with ContraWiMAE showing further improvements in linear separability and adaptability in diverse wireless environments. Comparative evaluations against a state-of-the-art wireless channel foundation model confirm the superior performance and data efficiency of our models, highlighting their potential as powerful baselines for future research in self-supervised wireless channel representation learning.
A transformer-based deep learning approach is presented that enables the diagnosis of fault cases in optical fiber amplifiers using condition-based monitoring time series data. The model, Inverse Triple-Aspect Self-Attention Transformer (ITST), uses an encoder-decoder architecture, utilizing three feature extraction paths in the encoder, feature-engineered data for the decoder and a self-attention mechanism. The results show that ITST outperforms state-of-the-art models in terms of classification accuracy, which enables predictive maintenance for optical fiber amplifiers, reducing network downtimes and maintenance costs.
Speech compression is commonly used to send voice over radio channels in applications such as mobile telephony and two-way push-to-talk (PTT) radio. In classical systems, the speech codec is combined with forward error correction, modulation and radio hardware. In this paper we describe an autoencoder that replaces many of the traditional signal processing elements with a neural network. The encoder takes a vocoder feature set (short term spectrum, pitch, voicing), and produces discrete time, but continuously valued quadrature amplitude modulation (QAM) symbols. We use orthogonal frequency domain multiplexing (OFDM) to send and receive these symbols over high frequency (HF) radio channels. The decoder converts received QAM symbols to vocoder features suitable for synthesis. The autoencoder has been trained to be robust to additive Gaussian noise and multipath channel impairments while simultaneously maintaining a Peak To Average Power Ratio (PAPR) of less than 1~dB. Over simulated and real world HF radio channels we have achieved output speech intelligibility that clearly surpasses existing analog and digital radio systems over a range of SNRs.
Agricultural monitoring is critical for ensuring food security, maintaining sustainable farming practices, informing policies on mitigating food shortage, and managing greenhouse gas emissions. Traditional process-based physical models are often designed and implemented for specific situations, and their parameters could also be highly uncertain. In contrast, data-driven models often use black-box structures and does not explicitly model the inter-dependence between different ecological variables. As a result, they require extensive training data and lack generalizability to different tasks with data distribution shifts and inconsistent observed variables. To address the need for more universal models, we propose a knowledge-guided encoder-decoder model, which can predict key crop variables by leveraging knowledge of underlying processes from multiple physical models. The proposed method also integrates a language model to process complex and inconsistent inputs and also utilizes it to implement a model selection mechanism for selectively combining the knowledge from different physical models. Our evaluations on predicting carbon and nitrogen fluxes for multiple sites demonstrate the effectiveness and robustness of the proposed model under various scenarios.
Singing melody extraction (SME) is a key task in the field of music information retrieval. However, existing methods are facing several limitations: firstly, prior models use transformers to capture the contextual dependencies, which requires quadratic computation resulting in low efficiency in the inference stage. Secondly, prior works typically rely on frequencysupervised methods to estimate the fundamental frequency (f0), which ignores that the musical performance is actually based on notes. Thirdly, transformers typically require large amounts of labeled data to achieve optimal performances, but the SME task lacks of sufficient annotated data. To address these issues, in this paper, we propose a mamba-based network, called SpectMamba, for semi-supervised singing melody extraction using confidence binary regularization. In particular, we begin by introducing vision mamba to achieve computational linear complexity. Then, we propose a novel note-f0 decoder that allows the model to better mimic the musical performance. Further, to alleviate the scarcity of the labeled data, we introduce a confidence binary regularization (CBR) module to leverage the unlabeled data by maximizing the probability of the correct classes. The proposed method is evaluated on several public datasets and the conducted experiments demonstrate the effectiveness of our proposed method.
Fine-tuning pre-trained language models (PLMs) has become a dominant paradigm in applying PLMs to downstream tasks. However, with limited fine-tuning, PLMs still struggle with the discrepancies between the representation obtained from the PLMs' encoder and the optimal input to the PLMs' decoder. This paper tackles this challenge by learning to calibrate the representation of PLMs in the latent space. In the proposed representation calibration method (RepCali), we integrate a specific calibration block to the latent space after the encoder and use the calibrated output as the decoder input. The merits of the proposed RepCali include its universality to all PLMs with encoder-decoder architectures, its plug-and-play nature, and ease of implementation. Extensive experiments on 25 PLM-based models across 8 tasks (including both English and Chinese datasets) demonstrate that the proposed RepCali offers desirable enhancements to PLMs (including LLMs) and significantly improves the performance of downstream tasks. Comparison experiments across 4 benchmark tasks indicate that RepCali is superior to the representative fine-tuning baselines.
Dense retrievers utilize pre-trained backbone language models (e.g., BERT, LLaMA) that are fine-tuned via contrastive learning to perform the task of encoding text into sense representations that can be then compared via a shallow similarity operation, e.g. inner product. Recent research has questioned the role of fine-tuning vs. that of pre-training within dense retrievers, specifically arguing that retrieval knowledge is primarily gained during pre-training, meaning knowledge not acquired during pre-training cannot be sub-sequentially acquired via fine-tuning. We revisit this idea here as the claim was only studied in the context of a BERT-based encoder using DPR as representative dense retriever. We extend the previous analysis by testing other representation approaches (comparing the use of CLS tokens with that of mean pooling), backbone architectures (encoder-only BERT vs. decoder-only LLaMA), and additional datasets (MSMARCO in addition to Natural Questions). Our study confirms that in DPR tuning, pre-trained knowledge underpins retrieval performance, with fine-tuning primarily adjusting neuron activation rather than reorganizing knowledge. However, this pattern does not hold universally, such as in mean-pooled (Contriever) and decoder-based (LLaMA) models. We ensure full reproducibility and make our implementation publicly available at https://github.com/ielab/DenseRetriever-Knowledge-Acquisition.
Precise forecasting of significant wave height (Hs) is essential for the development and utilization of wave energy. The challenges in predicting Hs arise from its non-linear and non-stationary characteristics. The combination of decomposition preprocessing and machine learning models have demonstrated significant effectiveness in Hs prediction by extracting data features. However, decomposing the unknown data in the test set can lead to data leakage issues. To simultaneously achieve data feature extraction and prevent data leakage, a novel Adaptive Feature Extraction Time-Frequency Network (AFE-TFNet) is proposed to improve prediction accuracy and stability. It is encoder-decoder rolling framework. The encoder consists of two stages: feature extraction and feature fusion. In the feature extraction stage, global and local frequency domain features are extracted by combining Wavelet Transform (WT) and Fourier Transform (FT), and multi-scale frequency analysis is performed using Inception blocks. In the feature fusion stage, time-domain and frequency-domain features are integrated through dominant harmonic sequence energy weighting (DHSEW). The decoder employed an advanced long short-term memory (LSTM) model. Hourly measured wind speed (Ws), dominant wave period (DPD), average wave period (APD) and Hs from three stations are used as the dataset, and the four metrics are employed to evaluate the forecasting performance. Results show that AFE-TFNet significantly outperforms benchmark methods in terms of prediction accuracy. Feature extraction can significantly improve the prediction accuracy. DHSEW has substantially increased the accuracy of medium-term to long-term forecasting. The prediction accuracy of AFE-TFNet does not demonstrate significant variability with changes of rolling time window size. Overall, AFE-TFNet shows strong potential for handling complex signal forecasting.
Decoder-only transformer networks have become incredibly popular for language modeling tasks. State-of-the-art models can have over a hundred transformer blocks, containing billions of trainable parameters, and are trained on trillions of tokens of text. Each transformer block typically consists of a multi-head attention (MHA) mechanism and a two-layer fully connected feedforward network (FFN). In this paper, we examine the importance of the FFN during the model pre-training process through a series of experiments, confirming that the FFN is important to model performance. Furthermore, we show that models using a transformer block configuration with three-layer FFNs with fewer such blocks outperform the standard two-layer configuration delivering lower training loss with fewer total parameters in less time.
Visible and infrared image fusion (VIF) has attracted significant attention in recent years. Traditional VIF methods primarily focus on generating fused images with high visual quality, while recent advancements increasingly emphasize incorporating semantic information into the fusion model during training. However, most existing segmentation-oriented VIF methods adopt a cascade structure comprising separate fusion and segmentation models, leading to increased network complexity and redundancy. This raises a critical question: can we design a more concise and efficient structure to integrate semantic information directly into the fusion model during training-Inspired by multi-task learning, we propose a concise and universal training framework, MultiTaskVIF, for segmentation-oriented VIF models. In this framework, we introduce a multi-task head decoder (MTH) to simultaneously output both the fused image and the segmentation result during training. Unlike previous cascade training frameworks that necessitate joint training with a complete segmentation model, MultiTaskVIF enables the fusion model to learn semantic features by simply replacing its decoder with MTH. Extensive experimental evaluations validate the effectiveness of the proposed method. Our code will be released upon acceptance.
Automated Radiology report generation (RRG) aims at producing detailed descriptions of medical images, reducing radiologists' workload and improving access to high-quality diagnostic services. Existing encoder-decoder models only rely on visual features extracted from raw input images, which can limit the understanding of spatial structures and semantic relationships, often resulting in suboptimal text generation. To address this, we propose Anatomical Attention Alignment Network (A3Net), a framework that enhance visual-textual understanding by constructing hyper-visual representations. Our approach integrates a knowledge dictionary of anatomical structures with patch-level visual features, enabling the model to effectively associate image regions with their corresponding anatomical entities. This structured representation improves semantic reasoning, interpretability, and cross-modal alignment, ultimately enhancing the accuracy and clinical relevance of generated reports. Experimental results on IU X-Ray and MIMIC-CXR datasets demonstrate that A3Net significantly improves both visual perception and text generation quality. Our code is available at \href{https://github.com/Vinh-AI/A3Net}{GitHub}.
We introduce xGen-small, a family of 4B and 9B Transformer decoder models optimized for long-context applications. Our vertically integrated pipeline unites domain-balanced, frequency-aware data curation; multi-stage pre-training with quality annealing and length extension to 128k tokens; and targeted post-training via supervised fine-tuning, preference learning, and online reinforcement learning. xGen-small delivers strong performance across various tasks, especially in math and coding domains, while excelling at long context benchmarks.
The automated summarisation of long legal documents can be a great aid for legal experts in their daily work. We automatically create summaries (guiding principles) of German judgments by fine-tuning a decoder-based large language model. We enrich the judgments with information about legal entities before the training. For the evaluation of the created summaries, we define a set of evaluation classes which allows us to measure their language, pertinence, completeness and correctness. Our results show that employing legal entities helps the generative model to find the relevant content, but the quality of the created summaries is not yet sufficient for a use in practice.
This work presents an efficient decoding pipeline for neuromorphic implantable brain-machine interfaces (Neu-iBMI), leveraging sparse neural event data from an event-based neural sensing scheme. We introduce a tunable event filter (EvFilter), which also functions as a spike detector (EvFilter-SPD), significantly reducing the number of events processed for decoding by 192X and 554X, respectively. The proposed pipeline achieves high decoding performance, up to R^2=0.73, with ANN- and SNN-based decoders, eliminating the need for signal recovery, spike detection, or sorting, commonly performed in conventional iBMI systems. The SNN-Decoder reduces computations and memory required by 5-23X compared to NN-, and LSTM-Decoders, while the ST-NN-Decoder delivers similar performance to an LSTM-Decoder requiring 2.5X fewer resources. This streamlined approach significantly reduces computational and memory demands, making it ideal for low-power, on-implant, or wearable iBMIs.
In this paper, we examine a key limitation in query-based detectors for temporal action detection (TAD), which arises from their direct adaptation of originally designed architectures for object detection. Despite the effectiveness of the existing models, they struggle to fully address the unique challenges of TAD, such as the redundancy in multi-scale features and the limited ability to capture sufficient temporal context. To address these issues, we propose a multi-dilated gated encoder and central-adjacent region integrated decoder for temporal action detection transformer (DiGIT). Our approach replaces the existing encoder that consists of multi-scale deformable attention and feedforward network with our multi-dilated gated encoder. Our proposed encoder reduces the redundant information caused by multi-level features while maintaining the ability to capture fine-grained and long-range temporal information. Furthermore, we introduce a central-adjacent region integrated decoder that leverages a more comprehensive sampling strategy for deformable cross-attention to capture the essential information. Extensive experiments demonstrate that DiGIT achieves state-of-the-art performance on THUMOS14, ActivityNet v1.3, and HACS-Segment. Code is available at: https://github.com/Dotori-HJ/DiGIT