This paper investigates over-the-air computation (AirComp) over multiple-access time-varying channels, where devices with high mobility transmit their sensing data to a fusion center (FC) for averaging. To combat the Doppler shift induced by time-varying channels, each device adopts orthogonal time frequency space (OTFS) modulation. Our objective is minimizing the mean squared error (MSE) for the target function estimation. Due to the multipath time-varying channels, the OTFS-based AirComp not only suffers from noise but also interference. Specifically, we propose three schemes, namely S1, S2, and S3, for the target function estimation. S1 directly estimates the target function under the impacts of noise and interference. S2 mitigates the interference by introducing a zero padding-assisted OTFS. In S3, we propose an iterative algorithm to estimate the function in a matrix form. In the numerical results, we evaluate the performance of S1, S2, and S3 from the perspectives of MSE and computational complexity, and compare them with benchmarks. Specifically, compared to benchmarks, S3 outperforms them with a significantly lower MSE but incurs a higher computational complexity. In contrast, S2 demonstrates a reduction in both MSE and computational complexity. Lastly, S1 shows superior error performance at small SNR and reduced computational complexity.
相關內容
In this paper, we focus on addressing the constraints faced when applying LLMs to ASR. Recent works utilize prefixLM-type models, which directly apply speech as a prefix to LLMs for ASR. We have found that optimizing speech prefixes leads to better ASR performance and propose applying RNNT loss to perform speech prefix-tuning. This is a simple approach and does not increase the model complexity or alter the inference pipeline. We also propose language-based soft prompting to further improve with frozen LLMs. Empirical analysis on realtime testset from 10 Indic languages demonstrate that our proposed speech prefix-tuning yields improvements with both frozen and fine-tuned LLMs. Our recognition results on an average of 10 Indics show that the proposed prefix-tuning with RNNT loss results in a 12\% relative improvement in WER over the baseline with a fine-tuned LLM. Our proposed approches with the frozen LLM leads to a 31\% relative improvement over basic soft-prompting prefixLM.
This paper examines gradient flow dynamics of two-homogeneous neural networks for small initializations, where all weights are initialized near the origin. For both square and logistic losses, it is shown that for sufficiently small initializations, the gradient flow dynamics spend sufficient time in the neighborhood of the origin to allow the weights of the neural network to approximately converge in direction to the Karush-Kuhn-Tucker (KKT) points of a neural correlation function that quantifies the correlation between the output of the neural network and corresponding labels in the training data set. For square loss, it has been observed that neural networks undergo saddle-to-saddle dynamics when initialized close to the origin. Motivated by this, this paper also shows a similar directional convergence among weights of small magnitude in the neighborhood of certain saddle points.
This paper describes a novel semi-autonomous mobile robot system designed to assist search and rescue (SAR) first responders in disaster scenarios. While robots offer significant potential in SAR missions, current solutions are limited in their ability to handle a diverse range of tasks. This gap is addressed by presenting a system capable of (1) autonomous navigation and mapping, allowing the robot to autonomously explore and map areas affected by catastrophic events, (2) radiation mapping, enabling the system to triangulate a radiation map from discrete radiation measurements to aid in identifying hazardous areas, (3) semi-autonomous substance sampling, allowing the robot to collect samples of suspicious substances and analyze them onboard with immediate classification, and (4) valve manipulation, enabling teleoperated closing of valves that control hazardous material flow. This semi-autonomous approach balances human control over critical tasks like substance sampling with efficient robot navigation in low-risk areas. The system is evaluated during three trials that simulate possible disaster scenarios, two of which have been recorded during the European Robotics Hackathon (EnRicH). Furthermore, we provide recorded sensor data as well as the implemented software system as supplemental material through a GitHub repository: //github.com/TW-Robotics/search-and-rescue-robot-IROS2024.
This paper presents novel techniques for improving the error correction performance and reducing the complexity of coarsely quantized 5G-LDPC decoders. The proposed decoder design supports arbitrary message-passing schedules on a base-matrix level by modeling exchanged messages with entry-specific discrete random variables. Variable nodes (VNs) and check nodes (CNs) involve compression operations designed using the information bottleneck method to maximize preserved mutual information between code bits and quantized messages. We introduce alignment regions that assign the messages to groups with aligned reliability levels to decrease the number of individual design parameters. Group compositions with degree-specific separation of messages improve performance by up to 0.4 dB. Further, we generalize our recently proposed CN-aware quantizer design to irregular LDPC codes and layered schedules. The method optimizes the VN quantizer to maximize preserved mutual information at the output of the subsequent CN update, enhancing performance by up to 0.2 dB. A schedule optimization modifies the order of layer updates, reducing the average iteration count by up to 35 %. We integrate all new techniques in a rate-compatible decoder design by extending the alignment regions along a rate-dimension. Our complexity analysis for 2-bit decoding estimates up to 64 % higher throughput versus 4-bit decoding at similar performance.
This paper addresses the challenge of developing a robust audio-visual deepfake detection model. In practical use cases, new generation algorithms are continually emerging, and these algorithms are not encountered during the development of detection methods. This calls for the generalization ability of the method. Additionally, to ensure the credibility of detection methods, it is beneficial for the model to interpret which cues from the video indicate it is fake. Motivated by these considerations, we then propose a multi-stream fusion approach with one-class learning as a representation-level regularization technique. We study the generalization problem of audio-visual deepfake detection by creating a new benchmark by extending and re-splitting the existing FakeAVCeleb dataset. The benchmark contains four categories of fake video(Real Audio-Fake Visual, Fake Audio-Fake Visual, Fake Audio-Real Visual, and unsynchronized video). The experimental results show that our approach improves the model's detection of unseen attacks by an average of 7.31% across four test sets, compared to the baseline model. Additionally, our proposed framework offers interpretability, indicating which modality the model identifies as fake.
This paper introduces a novel method for the stability analysis of positive feedback systems with a class of fully connected feedforward neural networks (FFNN) controllers. By establishing sector bounds for fully connected FFNNs without biases, we present a stability theorem that demonstrates the global exponential stability of linear systems under fully connected FFNN control. Utilizing principles from positive Lur'e systems and the positive Aizerman conjecture, our approach effectively addresses the challenge of ensuring stability in highly nonlinear systems. The crux of our method lies in maintaining sector bounds that preserve the positivity and Hurwitz property of the overall Lur'e system. We showcase the practical applicability of our methodology through its implementation in a linear system managed by a FFNN trained on output feedback controller data, highlighting its potential for enhancing stability in dynamic systems.
This paper investigates the efficacy of jointly optimizing content-specific post-processing filters to adapt a human oriented video/image codec into a codec suitable for machine vision tasks. By observing that artifacts produced by video/image codecs are content-dependent, we propose a novel training strategy based on competitive learning principles. This strategy assigns training samples to filters dynamically, in a fuzzy manner, which further optimizes the winning filter on the given sample. Inspired by simulated annealing optimization techniques, we employ a softmax function with a temperature variable as the weight allocation function to mitigate the effects of random initialization. Our evaluation, conducted on a system utilizing multiple post-processing filters within a Versatile Video Coding (VVC) codec framework, demonstrates the superiority of content-specific filters trained with our proposed strategies, specifically, when images are processed in blocks. Using VVC reference software VTM 12.0 as the anchor, experiments on the OpenImages dataset show an improvement in the BD-rate reduction from -41.3% and -44.6% to -42.3% and -44.7% for object detection and instance segmentation tasks, respectively, compared to independently trained filters. The statistics of the filter usage align with our hypothesis and underscore the importance of jointly optimizing filters for both content and reconstruction quality. Our findings pave the way for further improving the performance of video/image codecs.
This paper develops rollover prevention guarantees for mobile robots using control barrier function (CBF) theory, and demonstrates the method experimentally. We consider a safety measure based on a zero moment point condition through the lens of CBFs. However, these conditions depend on time-varying and noisy parameters. To address this issue, we present a differentiator-based safety-critical controller that estimates these parameters and pairs Input-to-State Stable (ISS) differentiator dynamics with CBFs to achieve rigorous safety guarantees. Additionally, to ensure safety in the presence of disturbances, we utilize a time-varying extension of Projection-to-State Safety (PSSf). The effectiveness of the proposed method is demonstrated via experiments on a tracked robot with a rollover potential on steep slopes.
Bayesian Image-on-Scalar Regression (ISR) offers significant advantages for neuroimaging data analysis, including flexibility and the ability to quantify uncertainty. However, its application to large-scale imaging datasets, such as found in the UK Biobank, is hindered by the computational demands of traditional posterior computation methods, as well as the challenge of individual-specific brain masks that deviate from the common mask typically used in standard ISR approaches. To address these challenges, we introduce a novel Bayesian ISR model that is scalable and accommodates inconsistent brain masks across subjects in large-scale imaging studies. Our model leverages Gaussian process priors and integrates salience area indicators to facilitate ISR. We develop a cutting-edge scalable posterior computation algorithm that employs stochastic gradient Langevin dynamics coupled with memory mapping techniques, ensuring that computation time scales linearly with subsample size and memory usage is constrained only by the batch size. Our approach uniquely enables direct spatial posterior inferences on brain activation regions. The efficacy of our method is demonstrated through simulations and analysis of the UK Biobank task fMRI data, encompassing 38,639 subjects and over 120,000 voxels per image, showing that it can achieve a speed increase of 4 to 11 times and enhance statistical power by 8% to 18% compared to traditional Gibbs sampling with zero-imputation in various simulation scenarios.
Vast amount of data generated from networks of sensors, wearables, and the Internet of Things (IoT) devices underscores the need for advanced modeling techniques that leverage the spatio-temporal structure of decentralized data due to the need for edge computation and licensing (data access) issues. While federated learning (FL) has emerged as a framework for model training without requiring direct data sharing and exchange, effectively modeling the complex spatio-temporal dependencies to improve forecasting capabilities still remains an open problem. On the other hand, state-of-the-art spatio-temporal forecasting models assume unfettered access to the data, neglecting constraints on data sharing. To bridge this gap, we propose a federated spatio-temporal model -- Cross-Node Federated Graph Neural Network (CNFGNN) -- which explicitly encodes the underlying graph structure using graph neural network (GNN)-based architecture under the constraint of cross-node federated learning, which requires that data in a network of nodes is generated locally on each node and remains decentralized. CNFGNN operates by disentangling the temporal dynamics modeling on devices and spatial dynamics on the server, utilizing alternating optimization to reduce the communication cost, facilitating computations on the edge devices. Experiments on the traffic flow forecasting task show that CNFGNN achieves the best forecasting performance in both transductive and inductive learning settings with no extra computation cost on edge devices, while incurring modest communication cost.
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191