PyTorchKR Papers

최신 논문

총 61편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2308.11918공식 코드 · Papers with Code

AMSP-UOD: When Vortex Convolution and Stochastic Perturbation Meet Underwater Object Detection

In this paper, we present a novel Amplitude-Modulated Stochastic Perturbation and Vortex Convolutional Network, AMSP-UOD, designed for underwater object detection. AMSP-UOD specifically addresses the impact of non-ideal imaging factors on detection accuracy in complex underwater environments. To mitigate the influence of noise on object detection performance, we propose AMSP Vortex Convolution (AMSP-VConv) to disrupt the noise distribution, enhance feature extraction capabilities, effectively reduce parameters, and improve network robustness. We design the Feature Association Decoupling Cross Stage Partial (FAD-CSP) module, which strengthens the association of long and short range features, improving the network performance in complex underwater environments. Additionally, our sophisticated post-processing method, based on Non-Maximum Suppression (NMS) with aspect-ratio similarity thresholds, optimizes detection in dense scenes, such as waterweed and schools of fish, improving object detection accuracy. Extensive experiments on the URPC and RUOD datasets demonstrate that our method outperforms existing state-of-the-art methods in terms of accuracy and noise immunity. AMSP-UOD proposes an innovative solution with the potential for real-world applications. Our code is available at https://github.com/zhoujingchun03/AMSP-UOD.

arXiv:2307.11108

Flatness-Aware Minimization for Domain Generalization

Domain generalization (DG) seeks to learn robust models that generalize well under unknown distribution shifts. As a critical aspect of DG, optimizer selection has not been explored in depth. Currently, most DG methods follow the widely used benchmark, DomainBed, and utilize Adam as the default optimizer for all datasets. However, we reveal that Adam is not necessarily the optimal choice for the majority of current DG methods and datasets. Based on the perspective of loss landscape flatness, we propose a novel approach, Flatness-Aware Minimization for Domain Generalization (FAD), which can efficiently optimize both zeroth-order and first-order flatness simultaneously for DG. We provide theoretical analyses of the FAD's out-of-distribution (OOD) generalization error and convergence. Our experimental results demonstrate the superiority of FAD on various DG datasets. Additionally, we confirm that FAD is capable of discovering flatter optima in comparison to other zeroth-order and first-order flatness-aware optimization methods.

arXiv:2306.14221

Feature Adversarial Distillation for Point Cloud Classification

Due to the point cloud's irregular and unordered geometry structure, conventional knowledge distillation technology lost a lot of information when directly used on point cloud tasks. In this paper, we propose Feature Adversarial Distillation (FAD) method, a generic adversarial loss function in point cloud distillation, to reduce loss during knowledge transfer. In the feature extraction stage, the features extracted by the teacher are used as the discriminator, and the students continuously generate new features in the training stage. The feature of the student is obtained by attacking the feedback from the teacher and getting a score to judge whether the student has learned the knowledge well or not. In experiments on standard point cloud classification on ModelNet40 and ScanObjectNN datasets, our method reduced the information loss of knowledge transfer in distillation in 40x model compression while maintaining competitive performance.

arXiv:2303.07914공식 코드 · Papers with Code

Adapting Offline Speech Translation Models for Streaming with Future-Aware Distillation and Inference

A popular approach to streaming speech translation is to employ a single offline model with a wait-k policy to support different latency requirements, which is simpler than training multiple online models with different latency constraints. However, there is a mismatch problem in using a model trained with complete utterances for streaming inference with partial input. We demonstrate that speech representations extracted at the end of a streaming input are significantly different from those extracted from a complete utterance. To address this issue, we propose a new approach called Future-Aware Streaming Translation (FAST) that adapts an offline ST model for streaming input. FAST includes a Future-Aware Inference (FAI) strategy that incorporates future context through a trainable masked embedding, and a Future-Aware Distillation (FAD) framework that transfers future context from an approximation of full speech to streaming input. Our experiments on the MuST-C EnDe, EnEs, and EnFr benchmarks show that FAST achieves better trade-offs between translation quality and latency than strong baselines. Extensive analyses suggest that our methods effectively alleviate the aforementioned mismatch problem between offline training and online inference.

arXiv:2303.02396

A General Framework for Learning Procedural Audio Models of Environmental Sounds

This paper introduces the Procedural (audio) Variational autoEncoder (ProVE) framework as a general approach to learning Procedural Audio PA models of environmental sounds with an improvement to the realism of the synthesis while maintaining provision of control over the generated sound through adjustable parameters. The framework comprises two stages: (i) Audio Class Representation, in which a latent representation space is defined by training an audio autoencoder, and (ii) Control Mapping, in which a joint function of static/temporal control variables derived from the audio and a random sample of uniform noise is learned to replace the audio encoder. We demonstrate the use of ProVE through the example of footstep sound effects on various surfaces. Our results show that ProVE models outperform both classical PA models and an adversarial-based approach in terms of sound fidelity, as measured by Fr\'echet Audio Distance (FAD), Maximum Mean Discrepancy (MMD), and subjective evaluations, making them feasible tools for sound design workflows.

arXiv:2301.07806

Federated Automatic Differentiation

Federated learning (FL) is a general framework for learning across an axis of group partitioned data (heterogeneous clients) while preserving data privacy, under the orchestration of a central server. FL methods often compute gradients of loss functions purely locally (ie. entirely at each client, or entirely at the server), typically using automatic differentiation (AD) techniques. We propose a federated automatic differentiation (FAD) framework that 1) enables computing derivatives of functions involving client and server computation as well as communication between them and 2) operates in a manner compatible with existing federated technology. In other words, FAD computes derivatives across communication boundaries. We show, in analogy with traditional AD, that FAD may be implemented using various accumulation modes, which introduce distinct computation-communication trade-offs and systems requirements. Further, we show that a broad class of federated computations is closed under these various modes of FAD, implying in particular that if the original computation can be implemented using privacy-preserving primitives, its derivative may be computed using only these same primitives. We then show how FAD can be used to create algorithms that dynamically learn components of the algorithm itself. In particular, we show that FedAvg-style algorithms can exhibit significantly improved performance by using FAD to adjust the server optimization step automatically, or by using FAD to learn weighting schemes for computing weighted averages across clients.

arXiv:2212.09478공식 코드 · Papers with Code

MM-Diffusion: Learning Multi-Modal Diffusion Models for Joint Audio and Video Generation

We propose the first joint audio-video generation framework that brings engaging watching and listening experiences simultaneously, towards high-quality realistic videos. To generate joint audio-video pairs, we propose a novel Multi-Modal Diffusion model (i.e., MM-Diffusion), with two-coupled denoising autoencoders. In contrast to existing single-modal diffusion models, MM-Diffusion consists of a sequential multi-modal U-Net for a joint denoising process by design. Two subnets for audio and video learn to gradually generate aligned audio-video pairs from Gaussian noises. To ensure semantic consistency across modalities, we propose a novel random-shift based attention block bridging over the two subnets, which enables efficient cross-modal alignment, and thus reinforces the audio-video fidelity for each other. Extensive experiments show superior results in unconditional audio-video generation, and zero-shot conditional tasks (e.g., video-to-audio). In particular, we achieve the best FVD and FAD on Landscape and AIST++ dancing datasets. Turing tests of 10k votes further demonstrate dominant preferences for our model. The code and pre-trained models can be downloaded at https://github.com/researchmm/MM-Diffusion.

arXiv:2211.15076공식 코드 · Papers with Code

Refined Semantic Enhancement towards Frequency Diffusion for Video Captioning

Video captioning aims to generate natural language sentences that describe the given video accurately. Existing methods obtain favorable generation by exploring richer visual representations in encode phase or improving the decoding ability. However, the long-tailed problem hinders these attempts at low-frequency tokens, which rarely occur but carry critical semantics, playing a vital role in the detailed generation. In this paper, we introduce a novel Refined Semantic enhancement method towards Frequency Diffusion (RSFD), a captioning model that constantly perceives the linguistic representation of the infrequent tokens. Concretely, a Frequency-Aware Diffusion (FAD) module is proposed to comprehend the semantics of low-frequency tokens to break through generation limitations. In this way, the caption is refined by promoting the absorption of tokens with insufficient occurrence. Based on FAD, we design a Divergent Semantic Supervisor (DSS) module to compensate for the information loss of high-frequency tokens brought by the diffusion process, where the semantics of low-frequency tokens is further emphasized to alleviate the long-tailed problem. Extensive experiments indicate that RSFD outperforms the state-of-the-art methods on two benchmark datasets, i.e., MSR-VTT and MSVD, demonstrate that the enhancement of low-frequency tokens semantics can obtain a competitive generation effect. Code is available at https://github.com/lzp870/RSFD.

arXiv:2210.17496

A Fast Automatic Method for Deconvoluting Macro X-ray Fluorescence Data Collected from Easel Paintings

Macro X-ray Fluorescence (MA-XRF) scanning is increasingly widely used by researchers in heritage science to analyse easel paintings as one of a suite of non-invasive imaging techniques. The task of processing the resulting MA-XRF datacube generated in order to produce individual chemical element maps is called MA-XRF deconvolution. While there are several existing methods that have been proposed for MA-XRF deconvolution, they require a degree of manual intervention from the user that can affect the final results. The state-of-the-art AFRID approach can automatically deconvolute the datacube without user input, but it has a long processing time and does not exploit spatial dependency. In this paper, we propose two versions of a fast automatic deconvolution (FAD) method for MA-XRF datacubes collected from easel paintings with ADMM (alternating direction method of multipliers) and FISTA (fast iterative shrinkage-thresholding algorithm). The proposed FAD method not only automatically analyses the datacube and produces element distribution maps of high-quality with spatial dependency considered, but also significantly reduces the running time. The results generated on the MA-XRF datacubes collected from two easel paintings from the National Gallery, London, verify the performance of the proposed FAD method.

arXiv:2210.15028

FaD-VLP: Fashion Vision-and-Language Pre-training towards Unified Retrieval and Captioning

Multimodal tasks in the fashion domain have significant potential for e-commerce, but involve challenging vision-and-language learning problems - e.g., retrieving a fashion item given a reference image plus text feedback from a user. Prior works on multimodal fashion tasks have either been limited by the data in individual benchmarks, or have leveraged generic vision-and-language pre-training but have not taken advantage of the characteristics of fashion data. Additionally, these works have mainly been restricted to multimodal understanding tasks. To address these gaps, we make two key contributions. First, we propose a novel fashion-specific pre-training framework based on weakly-supervised triplets constructed from fashion image-text pairs. We show the triplet-based tasks are an effective addition to standard multimodal pre-training tasks. Second, we propose a flexible decoder-based model architecture capable of both fashion retrieval and captioning tasks. Together, our model design and pre-training approach are competitive on a diverse set of fashion tasks, including cross-modal retrieval, image retrieval with text feedback, image captioning, relative image captioning, and multimodal categorization.

arXiv:2206.12563공식 코드 · Papers with Code

Generating Diverse Vocal Bursts with StyleGAN2 and MEL-Spectrograms

We describe our approach for the generative emotional vocal burst task (ExVo Generate) of the ICML Expressive Vocalizations Competition. We train a conditional StyleGAN2 architecture on mel-spectrograms of preprocessed versions of the audio samples. The mel-spectrograms generated by the model are then inverted back to the audio domain. As a result, our generated samples substantially improve upon the baseline provided by the competition from a qualitative and quantitative perspective for all emotions. More precisely, even for our worst-performing emotion (awe), we obtain an FAD of 1.76 compared to the baseline of 4.81 (as a reference, the FAD between the train/validation sets for awe is 0.776).

arXiv:2205.11811

Sensing Performance of Multi-Channel RFID-based Finger Augmentation Devices for Tactile Internet

Radiofrequency finger augmentation devices (R-FADs) are a recently introduced class of epidermal radiofrequency identification (RFID) sensor-tags attached to the fingers, communicating with a body-worn reader. These devices are promising candidates to enable Tactile Internet (TI) applications in the short term. R-FAD based on auto-tuning RFID microchips can be used as dielectric probes for the material of touched objects. However, due to the nearly unpredictable intrinsic variability of finger-object interaction, a single sensorized finger (single-channel device) is not enough to guarantee reliable data sampling. These limitations can be overcome by exploiting a multi-channel R-FAD sensorizing multiple fingers of the hand. In this paper, the dielectric-sensing performance of a multi-channel R-FAD, composed of sensors encapsulated into soft elastomers, is numerically and experimentally characterized, involving a set of volunteers. The inter-sensor coupling is negligible, thus enabling simultaneous independent dielectric measurements. The multi-sensor configuration allows for 100% reliability of the on-hand communication link for touched objects in a wide range of permittivity. Experiments moreover demonstrate that multi-channel measurements can halve the measurement uncertainty of the single-channel case. The achievable precision is suitable to discriminate among low-, medium-, and high-permittivity materials.

arXiv:2203.10425

A Study on Robustness to Perturbations for Representations of Environmental Sound

Audio applications involving environmental sound analysis increasingly use general-purpose audio representations, also known as embeddings, for transfer learning. Recently, Holistic Evaluation of Audio Representations (HEAR) evaluated twenty-nine embedding models on nineteen diverse tasks. However, the evaluation's effectiveness depends on the variation already captured within a given dataset. Therefore, for a given data domain, it is unclear how the representations would be affected by the variations caused by myriad microphones' range and acoustic conditions -- commonly known as channel effects. We aim to extend HEAR to evaluate invariance to channel effects in this work. To accomplish this, we imitate channel effects by injecting perturbations to the audio signal and measure the shift in the new (perturbed) embeddings with three distance measures, making the evaluation domain-dependent but not task-dependent. Combined with the downstream performance, it helps us make a more informed prediction of how robust the embeddings are to the channel effects. We evaluate two embeddings -- YAMNet, and OpenL3 on monophonic (UrbanSound8K) and polyphonic (SONYC-UST) urban datasets. We show that one distance measure does not suffice in such task-independent evaluation. Although Fr\'echet Audio Distance (FAD) correlates with the trend of the performance drop in the downstream task most accurately, we show that we need to study FAD in conjunction with the other distances to get a clear understanding of the overall effect of the perturbation. In terms of the embedding performance, we find OpenL3 to be more robust than YAMNet, which aligns with the HEAR evaluation.

arXiv:2109.06732

Tuna-AI: tuna biomass estimation with Machine Learning models trained on oceanography and echosounder FAD data

Echo-sounder data registered by buoys attached to drifting FADs provide a very valuable source of information on populations of tuna and their behaviour. This value increases whenthese data are supplemented with oceanographic data coming from CMEMS. We use these sources to develop Tuna-AI, a Machine Learning model aimed at predicting tuna biomass under a given buoy, which uses a 3-day window of echo-sounder data to capture the daily spatio-temporal patterns characteristic of tuna schools. As the supervised signal for training, we employ more than 5000 set events with their corresponding tuna catch reported by the AGAC tuna purse seine fleet.

arXiv:2109.02096공식 코드 · Papers with Code

Timbre Transfer with Variational Auto Encoding and Cycle-Consistent Adversarial Networks

This research project investigates the application of deep learning to timbre transfer, where the timbre of a source audio can be converted to the timbre of a target audio with minimal loss in quality. The adopted approach combines Variational Autoencoders with Generative Adversarial Networks to construct meaningful representations of the source audio and produce realistic generations of the target audio and is applied to the Flickr 8k Audio dataset for transferring the vocal timbre between speakers and the URMP dataset for transferring the musical timbre between instruments. Furthermore, variations of the adopted approach are trained, and generalised performance is compared using the metrics SSIM (Structural Similarity Index) and FAD (Frech\'et Audio Distance). It was found that a many-to-many approach supersedes a one-to-one approach in terms of reconstructive capabilities, and that the adoption of a basic over a bottleneck residual block design is more suitable for enriching content information about a latent space. It was also found that the decision on whether cyclic loss takes on a variational autoencoder or vanilla autoencoder approach does not have a significant impact on reconstructive and adversarial translation aspects of the model.

arXiv:2108.03169

Responding to Illegal Activities Along the Canadian Coastlines Using Reinforcement Learning

This article elaborates on how machine learning (ML) can leverage the solution of a contemporary problem related to the security of maritime domains. The worldwide ``Illegal, Unreported, and Unregulated'' (IUU) fishing incidents have led to serious environmental and economic consequences which involve drastic changes in our ecosystems in addition to financial losses caused by the depletion of natural resources. The Fisheries and Aquatic Department (FAD) of the United Nation's Food and Agriculture Organization (FAO) issued a report which indicated that the annual losses due to IUU fishing reached $25 Billion. This imposes negative impacts on the future-biodiversity of the marine ecosystem and domestic Gross National Product (GNP). Hence, robust interception mechanisms are increasingly needed for detecting and pursuing the unrelenting illegal fishing incidents in maritime territories. This article addresses the problem of coordinating the motion of a fleet of marine vessels (pursuers) to catch an IUU vessel while still in local waters. The problem is formulated as a pursuer-evader problem that is tackled within an ML framework. One or more pursuers, such as law enforcement vessels, intercept an evader (i.e., the illegal fishing ship) using an online reinforcement learning mechanism that is based on a value iteration process. It employs real-time navigation measurements of the evader ship as well as those of the pursuing vessels and returns back model-free interception strategies.

arXiv:2106.10714

Quantum Machine Learning: Fad or Future?

For the last few decades, classical machine learning has allowed us to improve the lives of many through automation, natural language processing, predictive analytics and much more. However, a major concern is the fact that we're fast approach the threshold of the maximum possible computational capacity available to us by the means of classical computing devices including CPUs, GPUs and Application Specific Integrated Circuits (ASICs). This is due to the exponential increase in model sizes which now have parameters in the magnitude of billions and trillions, requiring a significant amount of computing resources across a significant amount of time, just to converge one single model. To observe the efficacy of using quantum computing for certain machine learning tasks and explore the improved potential of convergence, error reduction and robustness to noisy data, this paper will look forth to test and verify the aspects in which quantum machine learning can help improve over classical machine learning approaches while also shedding light on the likely limitations that have prevented quantum approaches to become the mainstream. A major focus will be to recreate the work by Farhi et al and conduct experiments using their theory of performing machine learning in a quantum context, with assistance from the Tensorflow Quantum documentation.

arXiv:1810.12538

Phase asymmetry guided adaptive fractional-order total variation and diffusion for feature-preserving ultrasound despeckling

It is essential for ultrasound despeckling to remove speckle noise while simultaneously preserving edge features for accurate diagnosis and analysis in many applications. To preserve real edges such as ramp edges and low contrast edges, we first detect edges using a phase-based measure called phase asymmetry (PAS), which can distinguish small differences in transition border regions and varies from $0$ to $1$, taking $0$ in ideal smooth regions and taking $1$ at ideal step edges. We further propose three strategies to properly preserve edges. First, in observing that fractional-order anisotropic diffusion (FAD) filter has good performance in smooth regions while the fractional-order TV (FTV) filter performs better at edges, we leverage the PAS metric to keep a balance between FAD filter and FTV filter for achieving the best performance of preserving ramp edges. Second, considering that the FAD filter fails to protect low contrast edges by solely integrating gradient information into the diffusion coefficient, we integrate the PAS metric into the diffusion coefficient to properly preserve low contrast edges. Finally, different from fixed fractional order diffusion filters neglecting the differences between smooth regions and transition border regions, an adaptive fractional order is implemented based on the PAS metric to enhance edges. The experimental results show that our method outperforms other state-of-the-art ultrasound despeckling filters in both speckle reduction and feature preservation.

arXiv:2007.12075공식 코드 · Papers with Code

Representation Sharing for Fast Object Detector Search and Beyond

Region Proposal Network (RPN) provides strong support for handling the scale variation of objects in two-stage object detection. For one-stage detectors which do not have RPN, it is more demanding to have powerful sub-networks capable of directly capturing objects of unknown sizes. To enhance such capability, we propose an extremely efficient neural architecture search method, named Fast And Diverse (FAD), to better explore the optimal configuration of receptive fields and convolution types in the sub-networks for one-stage detectors. FAD consists of a designed search space and an efficient architecture search algorithm. The search space contains a rich set of diverse transformations designed specifically for object detection. To cope with the designed search space, a novel search algorithm termed Representation Sharing (RepShare) is proposed to effectively identify the best combinations of the defined transformations. In our experiments, FAD obtains prominent improvements on two types of one-stage detectors with various backbones. In particular, our FAD detector achieves 46.4 AP on MS-COCO (under single-scale testing), outperforming the state-of-the-art detectors, including the most recent NAS-based detectors, Auto-FPN (searched for 16 GPU-days) and NAS-FCOS (28 GPU-days), while significantly reduces the search cost to 0.6 GPU-days. Beyond object detection, we further demonstrate the generality of FAD on the more challenging instance segmentation, and expect it to benefit more tasks.

arXiv:1902.06208공식 코드 · Papers with Code

Twitch Plays Pokemon, Machine Learns Twitch: Unsupervised Context-Aware Anomaly Detection for Identifying Trolls in Streaming Data

With the increasing importance of online communities, discussion forums, and customer reviews, Internet "trolls" have proliferated thereby making it difficult for information seekers to find relevant and correct information. In this paper, we consider the problem of detecting and identifying Internet trolls, almost all of which are human agents. Identifying a human agent among a human population presents significant challenges compared to detecting automated spam or computerized robots. To learn a troll's behavior, we use contextual anomaly detection to profile each chat user. Using clustering and distance-based methods, we use contextual data such as the group's current goal, the current time, and the username to classify each point as an anomaly. A user whose features significantly differ from the norm will be classified as a troll. We collected 38 million data points from the viral Internet fad, Twitch Plays Pokemon. Using clustering and distance-based methods, we develop heuristics for identifying trolls. Using MapReduce techniques for preprocessing and user profiling, we are able to classify trolls based on 10 features extracted from a user's lifetime history.