PyTorchKR Papers

최신 논문

총 9,217편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2412.07486

Real-time Sign Language Recognition Using MobileNetV2 and Transfer Learning

The hearing-impaired community in India deserves the access to tools that help them communicate, however, there is limited known technology solutions that make use of Indian Sign Language (ISL) at present. Even though there are many ISL users, ISL cannot access social and education arenas because there is not yet an efficient technology to convert the ISL signal into speech or text. We initiated this initiative owing to the rising demand for products and technologies that are inclusive and help ISL, filling the gap of communication for the ones with hearing disability. Our goal is to build an reliable sign language recognition system with the help of Convolutional Neural Networks (CNN) to . By expanding communication access, we aspire toward better educational opportunities and a more inclusive society for hearing impaired people in India.

arXiv:2412.07228공식 코드 · Papers with Code

T-TIME: Test-Time Information Maximization Ensemble for Plug-and-Play BCIs

Objective: An electroencephalogram (EEG)-based brain-computer interface (BCI) enables direct communication between the human brain and a computer. Due to individual differences and non-stationarity of EEG signals, such BCIs usually require a subject-specific calibration session before each use, which is time-consuming and user-unfriendly. Transfer learning (TL) has been proposed to shorten or eliminate this calibration, but existing TL approaches mainly consider offline settings, where all unlabeled EEG trials from the new user are available. Methods: This paper proposes Test-Time Information Maximization Ensemble (T-TIME) to accommodate the most challenging online TL scenario, where unlabeled EEG data from the new user arrive in a stream, and immediate classification is performed. T-TIME initializes multiple classifiers from the aligned source data. When an unlabeled test EEG trial arrives, T-TIME first predicts its labels using ensemble learning, and then updates each classifier by conditional entropy minimization and adaptive marginal distribution regularization. Our code is publicized. Results: Extensive experiments on three public motor imagery based BCI datasets demonstrated that T-TIME outperformed about 20 classical and state-of-the-art TL approaches. Significance: To our knowledge, this is the first work on test time adaptation for calibration-free EEG-based BCIs, making plug-and-play BCIs possible.

arXiv:2412.07155

Annotation Techniques for Judo Combat Phase Classification from Tournament Footage

This paper presents a semi-supervised approach to extracting and analyzing combat phases in judo tournaments using live-streamed footage. The objective is to automate the annotation and summarization of live streamed judo matches. We train models that extract relevant entities and classify combat phases from fixed-perspective judo recordings. We employ semi-supervised methods to address limited labeled data in the domain. We build a model of combat phases via transfer learning from a fine-tuned object detector to classify the presence, activity, and standing state of the match. We evaluate our approach on a dataset of 19 thirty-second judo clips, achieving an F1 score on a $20\%$ test hold-out of 0.66, 0.78, and 0.87 for the three classes, respectively. Our results show initial promise for automating more complex information retrieval tasks using rigorous methods with limited labeled data.

arXiv:2412.07771

PETALface: Parameter Efficient Transfer Learning for Low-resolution Face Recognition

Pre-training on large-scale datasets and utilizing margin-based loss functions have been highly successful in training models for high-resolution face recognition. However, these models struggle with low-resolution face datasets, in which the faces lack the facial attributes necessary for distinguishing different faces. Full fine-tuning on low-resolution datasets, a naive method for adapting the model, yields inferior performance due to catastrophic forgetting of pre-trained knowledge. Additionally the domain difference between high-resolution (HR) gallery images and low-resolution (LR) probe images in low resolution datasets leads to poor convergence for a single model to adapt to both gallery and probe after fine-tuning. To this end, we propose PETALface, a Parameter-Efficient Transfer Learning approach for low-resolution face recognition. Through PETALface, we attempt to solve both the aforementioned problems. (1) We solve catastrophic forgetting by leveraging the power of parameter efficient fine-tuning(PEFT). (2) We introduce two low-rank adaptation modules to the backbone, with weights adjusted based on the input image quality to account for the difference in quality for the gallery and probe images. To the best of our knowledge, PETALface is the first work leveraging the powers of PEFT for low resolution face recognition. Extensive experiments demonstrate that the proposed method outperforms full fine-tuning on low-resolution datasets while preserving performance on high-resolution and mixed-quality datasets, all while using only 0.48% of the parameters. Code: https://kartik-3004.github.io/PETALface/

arXiv:2412.07448

Dynamic Ensemble Reasoning for LLM Experts

Ensemble reasoning for the strengths of different LLM experts is critical to achieving consistent and satisfactory performance on diverse inputs across a wide range of tasks. However, existing LLM ensemble methods are either computationally intensive or incapable of leveraging complementary knowledge among LLM experts for various inputs. In this paper, we propose a Dynamic Ensemble Reasoning paradigm, called DER to integrate the strengths of multiple LLM experts conditioned on dynamic inputs. Specifically, we model the LLM ensemble reasoning problem as a Markov Decision Process (MDP), wherein an agent sequentially takes inputs to request knowledge from an LLM candidate and passes the output to a subsequent LLM candidate. Moreover, we devise a reward function to train a DER-Agent to dynamically select an optimal answering route given the input questions, aiming to achieve the highest performance with as few computational resources as possible. Last, to fully transfer the expert knowledge from the prior LLMs, we develop a Knowledge Transfer Prompt (KTP) that enables the subsequent LLM candidates to transfer complementary knowledge effectively. Experiments demonstrate that our method uses fewer computational resources to achieve better performance compared to state-of-the-art baselines.

arXiv:2412.07186공식 코드 · Papers with Code

Monte Carlo Tree Search based Space Transfer for Black-box Optimization

Bayesian optimization (BO) is a popular method for computationally expensive black-box optimization. However, traditional BO methods need to solve new problems from scratch, leading to slow convergence. Recent studies try to extend BO to a transfer learning setup to speed up the optimization, where search space transfer is one of the most promising approaches and has shown impressive performance on many tasks. However, existing search space transfer methods either lack an adaptive mechanism or are not flexible enough, making it difficult to efficiently identify promising search space during the optimization process. In this paper, we propose a search space transfer learning method based on Monte Carlo tree search (MCTS), called MCTS-transfer, to iteratively divide, select, and optimize in a learned subspace. MCTS-transfer can not only provide a well-performing search space for warm-start but also adaptively identify and leverage the information of similar source tasks to reconstruct the search space during the optimization process. Experiments on synthetic functions, real-world problems, Design-Bench and hyper-parameter optimization show that MCTS-transfer can demonstrate superior performance compared to other search space transfer methods under different settings. Our code is available at \url{https://github.com/lamda-bbo/mcts-transfer}.

arXiv:2412.07182

An Enhancement of CNN Algorithm for Rice Leaf Disease Image Classification in Mobile Applications

This study focuses on enhancing rice leaf disease image classification algorithms, which have traditionally relied on Convolutional Neural Network (CNN) models. We employed transfer learning with MobileViTV2_050 using ImageNet-1k weights, a lightweight model that integrates CNN's local feature extraction with Vision Transformers' global context learning through a separable self-attention mechanism. Our approach resulted in a significant 15.66% improvement in classification accuracy for MobileViTV2_050-A, our first enhanced model trained on the baseline dataset, achieving 93.14%. Furthermore, MobileViTV2_050-B, our second enhanced model trained on a broader rice leaf dataset, demonstrated a 22.12% improvement, reaching 99.6% test accuracy. Additionally, MobileViTV2-A attained an F1-score of 93% across four rice labels and a Receiver Operating Characteristic (ROC) curve ranging from 87% to 97%. In terms of resource consumption, our enhanced models reduced the total parameters of the baseline CNN model by up to 92.50%, from 14 million to 1.1 million. These results indicate that MobileViTV2_050 not only improves computational efficiency through its separable self-attention mechanism but also enhances global context learning. Consequently, it offers a lightweight and robust solution suitable for mobile deployment, advancing the interpretability and practicality of models in precision agriculture.

arXiv:2412.05719공식 코드 · Papers with Code

Finite Element Neural Network Interpolation. Part I: Interpretable and Adaptive Discretization for Solving PDEs

We present the Finite Element Neural Network Interpolation (FENNI) framework, a sparse neural network architecture extending previous work on Embedded Finite Element Neural Networks (EFENN) introduced with the Hierarchical Deep-learning Neural Networks (HiDeNN). Due to their mesh-based structure, EFENN requires significantly fewer trainable parameters than fully connected neural networks, with individual weights and biases having a clear interpretation. Our FENNI framework, within the EFENN framework, brings improvements to the HiDeNN approach. First, we propose a reference element-based architecture where shape functions are defined on a reference element, enabling variability in interpolation functions and straightforward use of Gaussian quadrature rules for evaluating the loss function. Second, we propose a pragmatic multigrid training strategy based on the framework's interpretability. Third, HiDeNN's combined rh-adaptivity is extended from 1D to 2D, with a new Jacobian-based criterion for adding nodes combining h- and r-adaptivity. From a deep learning perspective, adaptive mesh behavior through rh-adaptivity and the multigrid approach correspond to transfer learning, enabling FENNI to optimize the network's architecture dynamically during training. The framework's capabilities are demonstrated on 1D and 2D test cases, where its accuracy and computational cost are compared against an analytical solution and a classical FEM solver. On these cases, the multigrid training strategy drastically improves the training stage's efficiency and robustness. Finally, we introduce a variational loss within the EFENN framework, showing that it performs as well as energy-based losses and outperforms residual-based losses. This framework is extended to surrogate modeling over the parametric space in Part II.

arXiv:2412.06233

Representational Transfer Learning for Matrix Completion

We propose to transfer representational knowledge from multiple sources to a target noisy matrix completion task by aggregating singular subspaces information. Under our representational similarity framework, we first integrate linear representation information by solving a two-way principal component analysis problem based on a properly debiased matrix-valued dataset. After acquiring better column and row representation estimators from the sources, the original high-dimensional target matrix completion problem is then transformed into a low-dimensional linear regression, of which the statistical efficiency is guaranteed. A variety of extensional arguments, including post-transfer statistical inference and robustness against negative transfer, are also discussed alongside. Finally, extensive simulation results and a number of real data cases are reported to support our claims.

arXiv:2412.05825공식 코드 · Papers with Code

Self-Supervised Learning with Probabilistic Density Labeling for Rainfall Probability Estimation

Numerical weather prediction (NWP) models are fundamental in meteorology for simulating and forecasting the behavior of various atmospheric variables. The accuracy of precipitation forecasts and the acquisition of sufficient lead time are crucial for preventing hazardous weather events. However, the performance of NWP models is limited by the nonlinear and unpredictable patterns of extreme weather phenomena driven by temporal dynamics. In this regard, we propose a \textbf{S}elf-\textbf{S}upervised \textbf{L}earning with \textbf{P}robabilistic \textbf{D}ensity \textbf{L}abeling (SSLPDL) for estimating rainfall probability by post-processing NWP forecasts. Our post-processing method uses self-supervised learning (SSL) with masked modeling for reconstructing atmospheric physics variables, enabling the model to learn the dependency between variables. The pre-trained encoder is then utilized in transfer learning to a precipitation segmentation task. Furthermore, we introduce a straightforward labeling approach based on probability density to address the class imbalance in extreme weather phenomena like heavy rain events. Experimental results show that SSLPDL surpasses other precipitation forecasting models in regional precipitation post-processing and demonstrates competitive performance in extending forecast lead times. Our code is available at https://github.com/joonha425/SSLPDL

arXiv:2412.06138

SGIA: Enhancing Fine-Grained Visual Classification with Sequence Generative Image Augmentation

In Fine-Grained Visual Classification (FGVC), distinguishing highly similar subcategories remains a formidable challenge, often necessitating datasets with extensive variability. The acquisition and annotation of such FGVC datasets are notably difficult and costly, demanding specialized knowledge to identify subtle distinctions among closely related categories. Our study introduces a novel approach employing the Sequence Latent Diffusion Model (SLDM) for augmenting FGVC datasets, called Sequence Generative Image Augmentation (SGIA). Our method features a unique Bridging Transfer Learning (BTL) process, designed to minimize the domain gap between real and synthetically augmented data. This approach notably surpasses existing methods in generating more realistic image samples, providing a diverse range of pose transformations that extend beyond the traditional rigid transformations and style changes in generative augmentation. We demonstrate the effectiveness of our augmented dataset with substantial improvements in FGVC tasks on various datasets, models, and training strategies, especially in few-shot learning scenarios. Our method outperforms conventional image augmentation techniques in benchmark tests on three FGVC datasets, showcasing superior realism, variability, and representational quality. Our work sets a new benchmark and outperforms the previous state-of-the-art models in classification accuracy by 0.5% for the CUB-200-2011 dataset and advances the application of generative models in FGVC data augmentation.

arXiv:2412.06664

Knowledge Transfer and Domain Adaptation for Fine-Grained Remote Sensing Image Segmentation

Fine-grained remote sensing image segmentation is essential for accurately identifying detailed objects in remote sensing images. Recently, vision transformer models (VTMs) pre-trained on large-scale datasets have demonstrated strong zero-shot generalization. However, directly applying them to specific tasks may lead to domain shift. We introduce a novel end-to-end learning paradigm combining knowledge guidance with domain refinement to enhance performance. We present two key components: the Feature Alignment Module (FAM) and the Feature Modulation Module (FMM). FAM aligns features from a CNN-based backbone with those from the pretrained VTM's encoder using channel transformation and spatial interpolation, and transfers knowledge via KL divergence and L2 normalization constraint. FMM further adapts the knowledge to the specific domain to address domain shift. We also introduce a fine-grained grass segmentation dataset and demonstrate, through experiments on two datasets, that our method achieves a significant improvement of 2.57 mIoU on the grass dataset and 3.73 mIoU on the cloud dataset. The results highlight the potential of combining knowledge transfer and domain adaptation to overcome domain-related challenges and data limitations. The project page is available at https://xavierjiezou.github.io/KTDA/.

arXiv:2412.05021

Improving data sharing and knowledge transfer via the Neuroelectrophysiology Analysis Ontology (NEAO)

Describing the processes involved in analyzing data from electrophysiology experiments to investigate the function of neural systems is inherently challenging. On the one hand, data can be analyzed by distinct methods that serve a similar purpose, such as different algorithms to estimate the spectral power content of a measured time series. On the other hand, different software codes can implement the same algorithm for the analysis while adopting different names to identify functions and parameters. Having reproducibility in mind, with these ambiguities the outcomes of the analysis are difficult to report, e.g., in the methods section of a manuscript or on a platform for scientific findings. Here, we illustrate how using an ontology to describe the analysis process can assist in improving clarity, rigour and comprehensibility by complementing, simplifying and classifying the details of the implementation. We implemented the Neuroelectrophysiology Analysis Ontology (NEAO) to define a unified vocabulary and to standardize the descriptions of the processes involved in analyzing data from neuroelectrophysiology experiments. Real-world examples demonstrate how the NEAO can be employed to annotate provenance information describing an analysis process. Based on such provenance, we detail how it can be used to query various types of information (e.g., using knowledge graphs) that enable researchers to find, understand and reuse prior analysis results.

논문 ID: 2022.91043

Completing Single-Cell DNA Methylome Profiles via Transfer Learning Together With KL-Divergence

The high level of sparsity in methylome profiles obtained using whole-genome bisulfite sequencing in the case of low biological material amount limits its value in the study of systems in which large samples are difficult to assemble, such as mammalian preimplantation embryonic development. The recently developed computational methods for addressing the sparsity by imputing missing have their limits when the required minimum data coverage or profiles of the same tissue in other modalities are not available. In this study, we explored the use of transfer learning together with Kullback-Leibler (KL) divergence to train predictive models for completing methylome profiles with very low coverage (below 2%). Transfer learning was used to leverage less sparse profiles that are typically available for different tissues for the same species, while KL divergence was employed to maximize the usage of information carried in the input data. A deep neural network was adopted to extract both DNA sequence and local methylation patterns for imputation. Our study of training models for completing methylome profiles of bovine oocytes and early embryos demonstrates the effectiveness of transfer learning and KL divergence, with individual increase of 29.98 and 29.43%, respectively, in prediction performance and 38.70% increase when the two were used together. The drastically increased data coverage (43.80–73.6%) after imputation powers downstream analyses involving methylomes that cannot be effectively done using the very low coverage profiles (0.06–1.47%) before imputation.

arXiv:2412.04616

Assessing and Learning Alignment of Unimodal Vision and Language Models

How well are unimodal vision and language models aligned? Although prior work have approached answering this question, their assessment methods do not directly translate to how these models are used in practical vision-language tasks. In this paper, we propose a direct assessment method, inspired by linear probing, to assess vision-language alignment. We identify that the degree of alignment of the SSL vision models depends on their SSL training objective, and we find that the clustering quality of SSL representations has a stronger impact on alignment performance than their linear separability. Next, we introduce Swift Alignment of Image and Language (SAIL), a efficient transfer learning framework that aligns pretrained unimodal vision and language models for downstream vision-language tasks. Since SAIL leverages the strengths of pretrained unimodal models, it requires significantly fewer (6%) paired image-text data for the multimodal alignment compared to models like CLIP which are trained from scratch. SAIL training only requires a single A100 GPU, 5 hours of training and can accommodate a batch size up to 32,768. SAIL achieves 73.4% zero-shot accuracy on ImageNet (vs. CLIP's 72.7%) and excels in zero-shot retrieval, complex reasoning, and semantic segmentation. Additionally, SAIL improves the language-compatibility of vision encoders that in turn enhance the performance of multimodal large language models. The entire codebase and model weights are open-source: https://lezhang7.github.io/sail.github.io/

arXiv:2412.04060

Expanding Deep Learning-based Sensing Systems with Multi-Source Knowledge Transfer

Expanding the existing sensing systems to provide high-quality deep learning models for more domains, such as new users or environments, is challenged by the limited labeled data and the data and device heterogeneities. While knowledge distillation methods could overcome label scarcity and device heterogeneity, they assume the teachers are fully reliable and overlook the data heterogeneity, which prevents the direct adoption of existing models. To address this problem, this paper proposes an efficient knowledge transfer framework, HaKT, to expand sensing systems. It first selects multiple high-quality models from the system at a low cost and then fuses their knowledge by assigning sample-wise weights to their predictions. Later, the fused knowledge is selectively injected into the customized models for new domains based on the knowledge quality. Extensive experiments on different tasks, modalities, and settings show that HaKT outperforms stat-of-the-art baselines by at most 16.5% accuracy and saves up to 39% communication traffic.

arXiv:2412.03851

FedMetaMed: Federated Meta-Learning for Personalized Medication in Distributed Healthcare Systems

Personalized medication aims to tailor healthcare to individual patient characteristics. However, the heterogeneity of patient data across healthcare systems presents significant challenges to achieving accurate and effective personalized treatments. Ethical concerns further complicate the aggregation of large volumes of data from diverse institutions. Federated Learning (FL) offers a promising decentralized solution by enabling collaborative model training through the exchange of client models rather than raw data, thus preserving privacy. However, existing FL methods often suffer from retrogression during server aggregation, leading to a decline in model performance in real-world medical FL settings. To address data variability in distributed healthcare systems, we introduce Federated Meta-Learning for Personalized Medication (FedMetaMed), which combines federated learning and meta-learning to create models that adapt to diverse patient data across healthcare systems. The FedMetaMed framework aims to produce superior personalized models for individual clients by addressing these limitations. Specifically, we introduce Cumulative Fourier Aggregation (CFA) at the server to improve stability and effectiveness in global knowledge aggregation. CFA achieves this by gradually integrating client models from low to high frequencies. At the client level, we implement a Collaborative Transfer Optimization (CTO) strategy with a three-step process - Retrieve, Reciprocate, and Refine - to enhance the personalized local model through seamless global knowledge transfer. Experiments on real-world medical imaging datasets demonstrate that FedMetaMed outperforms state-of-the-art FL methods, showing superior generalization even on out-of-distribution cohorts.

arXiv:2412.04276공식 코드 · Papers with Code

Graph-Sequential Alignment and Uniformity: Toward Enhanced Recommendation Systems

Graph-based and sequential methods are two popular recommendation paradigms, each excelling in its domain but lacking the ability to leverage signals from the other. To address this, we propose a novel method that integrates both approaches for enhanced performance. Our framework uses Graph Neural Network (GNN)-based and sequential recommenders as separate submodules while sharing a unified embedding space optimized jointly. To enable positive knowledge transfer, we design a loss function that enforces alignment and uniformity both within and across submodules. Experiments on three real-world datasets demonstrate that the proposed method significantly outperforms using either approach alone and achieves state-of-the-art results. Our implementations are publicly available at https://github.com/YuweiCao-UIC/GSAU.git.

arXiv:2412.04111

Adult Glioma Segmentation in Sub-Saharan Africa using Transfer Learning on Stratified Finetuning Data

Gliomas, a kind of brain tumor characterized by high mortality, present substantial diagnostic challenges in low- and middle-income countries, particularly in Sub-Saharan Africa. This paper introduces a novel approach to glioma segmentation using transfer learning to address challenges in resource-limited regions with minimal and low-quality MRI data. We leverage pre-trained deep learning models, nnU-Net and MedNeXt, and apply a stratified fine-tuning strategy using the BraTS2023-Adult-Glioma and BraTS-Africa datasets. Our method exploits radiomic analysis to create stratified training folds, model training on a large brain tumor dataset, and transfer learning to the Sub-Saharan context. A weighted model ensembling strategy and adaptive post-processing are employed to enhance segmentation accuracy. The evaluation of our proposed method on unseen validation cases on the BraTS-Africa 2024 task resulted in lesion-wise mean Dice scores of 0.870, 0.865, and 0.926, for enhancing tumor, tumor core, and whole tumor regions and was ranked first for the challenge. Our approach highlights the ability of integrated machine-learning techniques to bridge the gap between the medical imaging capabilities of resource-limited countries and established developed regions. By tailoring our methods to a target population's specific needs and constraints, we aim to enhance diagnostic capabilities in isolated environments. Our findings underscore the importance of approaches like local data integration and stratification refinement to address healthcare disparities, ensure practical applicability, and enhance impact. A dockerized version of the BraTS-Africa 2024 winning algorithm is available at https://hub.docker.com/r/aparida12/brats-ssa-2024 .

arXiv:2412.04266

Representation Purification for End-to-End Speech Translation

Speech-to-text translation (ST) is a cross-modal task that involves converting spoken language into text in a different language. Previous research primarily focused on enhancing speech translation by facilitating knowledge transfer from machine translation, exploring various methods to bridge the gap between speech and text modalities. Despite substantial progress made, factors in speech that are not relevant to translation content, such as timbre and rhythm, often limit the efficiency of knowledge transfer. In this paper, we conceptualize speech representation as a combination of content-agnostic and content-relevant factors. We examine the impact of content-agnostic factors on translation performance through preliminary experiments and observe a significant performance deterioration when content-agnostic perturbations are introduced to speech signals. To address this issue, we propose a \textbf{S}peech \textbf{R}epresentation \textbf{P}urification with \textbf{S}upervision \textbf{E}nhancement (SRPSE) framework, which excludes the content-agnostic components within speech representations to mitigate their negative impact on ST. Experiments on MuST-C and CoVoST-2 datasets demonstrate that SRPSE significantly improves translation performance across all translation directions in three settings and achieves preeminent performance under a \textit{transcript-free} setting.