Overlapping sound events are ubiquitous in real-world environments, but existing end-to-end sound event detection (SED) methods still struggle to detect them effectively. A critical reason is that these methods represent overlapping events using shared and entangled frame-wise features, which degrades the feature discrimination. To solve the problem, we propose a disentangled feature learning framework to learn a category-specific representation. Specifically, we employ different projectors to learn the frame-wise features for each category. To ensure that these feature does not contain information of other categories, we maximize the common information between frame-wise features within the same category and propose a frame-wise contrastive loss. In addition, considering that the labeled data used by the proposed method is limited, we propose a semi-supervised frame-wise contrastive loss that can leverage large amounts of unlabeled data to achieve feature disentanglement. The experimental results demonstrate the effectiveness of our method.
相關內容
Millimeter-wave (mmWave) and terahertz (THz) communication systems require large antenna arrays and use narrow directive beams to ensure sufficient receive signal power. However, selecting the optimal beams for these large antenna arrays incurs a significant beam training overhead, making it challenging to support applications involving high mobility. In recent years, machine learning (ML) solutions have shown promising results in reducing the beam training overhead by utilizing various sensing modalities such as GPS position and RGB images. However, the existing approaches are mainly limited to scenarios with only a single object of interest present in the wireless environment and focus only on co-located sensing, where all the sensors are installed at the communication terminal. This brings key challenges such as the limited sensing coverage compared to the coverage of the communication system and the difficulty in handling non-line-of-sight scenarios. To overcome these limitations, our paper proposes the deployment of multiple distributed sensing nodes, each equipped with an RGB camera. These nodes focus on extracting environmental semantics from the captured RGB images. The semantic data, rather than the raw images, are then transmitted to the basestation. This strategy significantly alleviates the overhead associated with the data storage and transmission of the raw images. Furthermore, semantic communication enhances the system's adaptability and responsiveness to dynamic environments, allowing for prioritization and transmission of contextually relevant information. Experimental results on the DeepSense 6G dataset demonstrate the effectiveness of the proposed solution in reducing the sensing data transmission overhead while accurately predicting the optimal beams in realistic communication environments.
This paper presents a neural vocoder based on a denoising diffusion probabilistic model (DDPM) incorporating explicit periodic signals as auxiliary conditioning signals. Recently, DDPM-based neural vocoders have gained prominence as non-autoregressive models that can generate high-quality waveforms. The neural vocoders based on DDPM have the advantage of training with a simple time-domain loss. In practical applications, such as singing voice synthesis, there is a demand for neural vocoders to generate high-fidelity speech waveforms with flexible pitch control. However, conventional DDPM-based neural vocoders struggle to generate speech waveforms under such conditions. Our proposed model aims to accurately capture the periodic structure of speech waveforms by incorporating explicit periodic signals. Experimental results show that our model improves sound quality and provides better pitch control than conventional DDPM-based neural vocoders.
Visual hallucination (VH) means that a multi-modal LLM (MLLM) imagines incorrect details about an image in visual question answering. Existing studies find VH instances only in existing image datasets, which results in biased understanding of MLLMs' performance under VH due to limited diversity of such VH instances. In this work, we propose a tool called VHTest to generate a diverse set of VH instances. Specifically, VHTest finds some initial VH instances in existing image datasets (e.g., COCO), generates a text description for each VH mode, and uses a text-to-image generative model (e.g., DALL-E-3) to generate VH images based on the text descriptions. We collect a benchmark dataset with 1,200 VH instances in 8 VH modes using VHTest. We find that existing MLLMs such as GPT-4V, LLaVA-1.5, and MiniGPT-v2 hallucinate for a large fraction of the instances in our benchmark. Moreover, we find that fine-tuning an MLLM using our benchmark dataset reduces its likelihood to hallucinate without sacrificing its performance on other benchmarks. Our benchmarks are publicly available: //github.com/wenhuang2000/VHTest.
We present for the first time a novel method that utilizes the chest movement-modulated radio signals for non-contact acquisition of the photoplethysmography (PPG) signal. Under the proposed method, a software-defined radio (SDR) exposes the chest of a subject sitting nearby to an orthogonal frequency division multiplexing signal with 64 sub-carriers at a center frequency 5.24 GHz, while another SDR in the close vicinity collects the modulated radio signal reflected off the chest. This way, we construct a custom dataset by collecting 160 minutes of labeled data (both raw radio data as well as the reference PPG signal) from 16 healthy young subjects. With this, we first utilize principal component analysis for dimensionality reduction of the radio data. Next, we denoise the radio signal and reference PPG signal using wavelet technique, followed by segmentation and Z-score normalization. We then synchronize the radio and PPG segments using cross-correlation method. Finally, we proceed to the waveform translation (regression) task, whereby we first convert the radio and PPG segments into frequency domain using discrete cosine transform (DCT), and then learn the non-linear regression between them. Eventually, we reconstruct the synthetic PPG signal by taking inverse DCT of the output of regression block, with a mean absolute error of 8.1294. The synthetic PPG waveform has a great clinical significance as it could be used for non-contact performance assessment of cardiovascular and respiratory systems of patients suffering from infectious diseases, e.g., covid19.
Recommender systems have been widely used for various scenarios, such as e-commerce, news, and music, providing online contents to help and enrich users' daily life. Different scenarios hold distinct and unique characteristics, calling for domain-specific investigations and corresponding designed recommender systems. Therefore, in this paper, we focus on food delivery recommendations to unveil unique features in this domain, where users order food online and enjoy their meals shortly after delivery. We first conduct an in-depth analysis on food delivery datasets. The analysis shows that repeat orders are prevalent for both users and stores, and situations' differently influence repeat and exploration consumption in the food delivery recommender systems. Moreover, we revisit the ability of existing situation-aware methods for repeat and exploration recommendations respectively, and find them unable to effectively solve both tasks simultaneously. Based on the analysis and experiments, we have designed two separate recommendation models -- ReRec for repeat orders and ExpRec for exploration orders; both are simple in their design and computation. We conduct experiments on three real-world food delivery datasets, and our proposed models outperform various types of baselines on repeat, exploration, and combined recommendation tasks. This paper emphasizes the importance of dedicated analyses and methods for domain-specific characteristics for the recommender system studies.
We propose two novel extensions of the Wyner common information optimization problem. Each relaxes one fundamental constraints in Wyner's formulation. The \textit{Variational Wyner Common Information} relaxes the matching constraint to the known distribution while imposing conditional independence to the feasible solution set. We derive a tight surrogate upper bound of the obtained unconstrained Lagrangian via the theory of variational inference, which can be minimized efficiently. Our solver caters to problems where conditional independence holds with significantly reduced computation complexity; On the other hand, the \textit{Bipartite Wyner Common Information} relaxes the conditional independence constraint whereas the matching condition is enforced on the feasible set. By leveraging the difference-of-convex structure of the formulated optimization problem, we show that our solver is resilient to conditional dependent sources. Both solvers are provably convergent (local stationary points), and empirically, they obtain more accurate solutions to Wyner's formulation with substantially less runtime. Moreover, them can be extended to unknown distribution settings by parameterizing the common randomness as a member of the exponential family of distributions. Our approaches apply to multi-modal clustering problems, where multiple modalities of observations come from the same cluster. Empirically, our solvers outperform the state-of-the-art multi-modal clustering algorithms with significantly improved performance.
Can uncorrelated surrounding sound sources be used to generate extended diffuse sound fields? By definition, targets are a constant sound pressure level, a vanishing average sound intensity, uncorrelated sound waves arriving isotropically from all directions. Does this require specific sources and geometries for surrounding 2D and 3D source layouts? As methods, we employ numeric simulations and undertake a series of calculations with uncorrelated circular/spherical source layouts, or such with infinite excess dimensions, and we point out relations to potential theory. Using a radial decay 1/r^b modified by the exponent b, the representation of the resulting fields with hypergeometric functions, Gegenbauer polynomials, and circular as well as spherical harmonics yields fruitful insights. In circular layouts, waves decaying by the exponent b=1/2 synthesize ideally extended, diffuse sound fields; spherical layouts do so with b=1. None of the layouts synthesizes a perfectly constant expected sound pressure level but its flatness is acceptable. Spherical t-designs describe optimal source layouts with well-described area of high diffuseness, and non-spherical, convex layouts can be improved by restoring isotropy or by mode matching for a maximally diffuse synthesis. Theory and simulation offer a basis for loudspeaker-based synthesis of diffuse sound fields and contribute physical reasons to recent psychoacoustic findings in spatial audio.
Existing knowledge graph (KG) embedding models have primarily focused on static KGs. However, real-world KGs do not remain static, but rather evolve and grow in tandem with the development of KG applications. Consequently, new facts and previously unseen entities and relations continually emerge, necessitating an embedding model that can quickly learn and transfer new knowledge through growth. Motivated by this, we delve into an expanding field of KG embedding in this paper, i.e., lifelong KG embedding. We consider knowledge transfer and retention of the learning on growing snapshots of a KG without having to learn embeddings from scratch. The proposed model includes a masked KG autoencoder for embedding learning and update, with an embedding transfer strategy to inject the learned knowledge into the new entity and relation embeddings, and an embedding regularization method to avoid catastrophic forgetting. To investigate the impacts of different aspects of KG growth, we construct four datasets to evaluate the performance of lifelong KG embedding. Experimental results show that the proposed model outperforms the state-of-the-art inductive and lifelong embedding baselines.
We propose a novel single shot object detection network named Detection with Enriched Semantics (DES). Our motivation is to enrich the semantics of object detection features within a typical deep detector, by a semantic segmentation branch and a global activation module. The segmentation branch is supervised by weak segmentation ground-truth, i.e., no extra annotation is required. In conjunction with that, we employ a global activation module which learns relationship between channels and object classes in a self-supervised manner. Comprehensive experimental results on both PASCAL VOC and MS COCO detection datasets demonstrate the effectiveness of the proposed method. In particular, with a VGG16 based DES, we achieve an mAP of 81.7 on VOC2007 test and an mAP of 32.8 on COCO test-dev with an inference speed of 31.5 milliseconds per image on a Titan Xp GPU. With a lower resolution version, we achieve an mAP of 79.7 on VOC2007 with an inference speed of 13.0 milliseconds per image.
High spectral dimensionality and the shortage of annotations make hyperspectral image (HSI) classification a challenging problem. Recent studies suggest that convolutional neural networks can learn discriminative spatial features, which play a paramount role in HSI interpretation. However, most of these methods ignore the distinctive spectral-spatial characteristic of hyperspectral data. In addition, a large amount of unlabeled data remains an unexploited gold mine for efficient data use. Therefore, we proposed an integration of generative adversarial networks (GANs) and probabilistic graphical models for HSI classification. Specifically, we used a spectral-spatial generator and a discriminator to identify land cover categories of hyperspectral cubes. Moreover, to take advantage of a large amount of unlabeled data, we adopted a conditional random field to refine the preliminary classification results generated by GANs. Experimental results obtained using two commonly studied datasets demonstrate that the proposed framework achieved encouraging classification accuracy using a small number of data for training.
小貼士
登錄享
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191