Machine Learning (ML) Latest News and Trending articles from all top sources only on Techminis

A naukri.com initiative

New

Home

ML News

Arxiv

229

Image Credit: Arxiv

Impact of Data Duplication on Deep Neural Network-Based Image Classifiers: Robust vs. Standard Models

The impact of duplicated training data in image classification models has been studied.
Duplicated images in the training set negatively affect model training efficiency.
The presence of duplicated images may result in lower accuracy of the image classifier.
Even when duplicated samples are selected uniformly, accuracy does not significantly improve.

Read Full Article

13 Likes

Arxiv

173

Image Credit: Arxiv

Learning to Normalize on the SPD Manifold under Bures-Wasserstein Geometry

Covariance matrices within the Symmetric Positive Definite (SPD) manifold play a crucial role in various scientific fields.
Researchers have developed neural networks on the SPD manifold to accurately learn covariance embeddings.
The existing Riemannian batch normalization (RBN) approach may not effectively handle ill-conditioned SPD matrices.
A novel Riemannian batch normalization (RBN) algorithm based on the Generalized Bures-Wasserstein metric (GBWM) is proposed for improved performance.

Read Full Article

10 Likes

Arxiv

184

Image Credit: Arxiv

Sim-is-More: Randomizing HW-NAS with Synthetic Devices

Researchers propose a two-stage hardware-aware neural architecture search (HW-NAS) framework for optimizing the performance of target devices.
The first stage involves training an architecture controller on synthetic devices.
The second stage deploys the learned controller on the target device, without relying on pre-collected information.
The framework enables the controller to design architecture for the target device through high-fidelity latency measurements and in-context adaptation.

Read Full Article

11 Likes

Arxiv

199

Image Credit: Arxiv

GraphMaster: Automated Graph Synthesis via LLM Agents in Data-Limited Environments

GraphMaster is introduced as a multi-agent framework designed for graph data synthesis in data-limited environments.
GraphMaster utilizes four specialized LLM agents to optimize the synthesis process, ensuring semantic coherence and structural integrity.
New data-limited graph benchmarks are created to evaluate the performance of GraphMaster in realistic constraints.
Experimental results show that GraphMaster outperforms traditional synthesis methods, paving the way for advancements in Graph Foundation Models (GFMs) in data-scarce environments.

Read Full Article

12 Likes

Discover more

Arxiv

203

Image Credit: Arxiv

Spectral Normalization and Voigt-Reuss net: A universal approach to microstructure-property forecasting with physical guarantees

Heterogeneous materials play a crucial role in various applications.
Current simulation-based approaches for evaluating material properties can be computationally expensive and lack gradients with respect to microstructure and constitutive parameters.
To address this, a novel spectral normalization scheme is proposed, which enforces physical bounds on constitutive response.
The approach is agnostic to microstructural features and surrogate models, resulting in improved accuracy and robustness.

Read Full Article

12 Likes

Arxiv

316

Image Credit: Arxiv

Alleviating Performance Disparity in Adversarial Spatiotemporal Graph Learning Under Zero-Inflated Distribution

Spatiotemporal Graph Learning (SGL) under Zero-Inflated Distribution (ZID) is crucial for urban risk management tasks.
Traditional adversarial training (AT) exacerbates performance disparities between majority and minority classes under ZID.
The proposed MinGRE framework addresses the performance disparity by reweighting gradients and enhancing representations of the minority class.
MinGRE achieves enhanced robustness and reduces the performance disparity across classes compared to existing baselines.

Read Full Article

19 Likes

Arxiv

Image Credit: Arxiv

Detection of Disease on Nasal Breath Sound by New Lightweight Architecture: Using COVID-19 as An Example

This study aims to develop a novel, lightweight deep neural network for efficient, accurate, and cost-effective detection of COVID-19 using a nasal breathing audio data collected via smartphones.
The proposed model achieved 97% accuracy in detecting COVID-19 from nasal breathing sounds.
The Dense-ReLU-Dropout model, using RF and PCA for feature selection, achieves high accuracy with greater computational efficiency compared to existing methods.
The findings suggest that the proposed method holds significant potential for clinical implementation, advancing smartphone-based diagnostics in infectious diseases.

Read Full Article

3 Likes

Arxiv

312

Image Credit: Arxiv

Integrating Fourier Neural Operators with Diffusion Models to improve Spectral Representation of Synthetic Earthquake Ground Motion Response

Nuclear reactor buildings must be designed to withstand the dynamic load induced by strong ground motion earthquakes.
In this study, an AI physics-based approach is proposed to generate synthetic ground motion by integrating a neural operator and a denoising diffusion probabilistic model.
The neural operator approximates the elastodynamics Green's operator, while the diffusion model corrects the generated ground motion time series.
The approach enhances the realism of synthetic seismograms and improves the frequency biases and Goodness-Of-Fit (GOF) scores.

Read Full Article

18 Likes

Arxiv

Image Credit: Arxiv

Automated Explanation of Machine Learning Models of Footballing Actions in Words

Researchers introduce a novel approach called wordalizations to bridge the communication gap between machine learning and football coaching.
They build an expected goals model using logistic regression and utilize the coefficients of the regression model to describe factors influencing the model's prediction.
Large language models are employed to provide entertaining descriptions of the shots.
The approach is discussed in a model card and an interactive open-source application is provided for shots in recent tournaments.

Read Full Article

3 Likes

Arxiv

Image Credit: Arxiv

Mixture-of-Experts for Distributed Edge Computing with Channel-Aware Gating Function

Researchers have developed a novel channel-aware gating function for wireless distributed mixture-of-experts (MoE) system.
The system incorporates channel conditions into the MoE gating mechanism, enabling effective operation in wireless networks.
By considering signal-to-noise ratios (SNRs) and expert specializations, the proposed approach optimizes expert selection.
Experimental results indicate that the channel-aware gating scheme outperformed traditional MoE models.

Read Full Article

4 Likes

Arxiv

165

Image Credit: Arxiv

Deep Generative Models: Complexity, Dimensionality, and Approximation

Generative networks have shown success in learning complex data distributions, but their theoretical foundation is unclear.
Previous theory suggested that the latent dimension needs to be at least equal to the intrinsic dimension of the data manifold to approximate its distribution.
However, a new study challenges this requirement by demonstrating that generative networks can approximate distributions on lower-dimensional manifolds from inputs of any dimension.
This finding implies a trade-off between approximation error, dimensionality, and model complexity in generative networks.

Read Full Article

9 Likes

Arxiv

143

Image Credit: Arxiv

Generalized Tensor-based Parameter-Efficient Fine-Tuning via Lie Group Transformations

Adapting pre-trained foundation models for diverse downstream tasks is a core practice in artificial intelligence.
Parameter-efficient fine-tuning (PEFT) methods like LoRA have emerged and are becoming a growing research focus.
A generalization of matrix-based PEFT methods to higher-dimensional parameter spaces is proposed, preserving the structural properties.
Extensive experiments on computer vision and natural language processing validate the effectiveness and versatility of the approach.

Read Full Article

8 Likes

Arxiv

146

Image Credit: Arxiv

ReaLitE: Enrichment of Relation Embeddings in Knowledge Graphs using Numeric Literals

Most knowledge graph embedding (KGE) methods focus on entities and relations, neglecting literal values.
ReaLitE is a relation-centric KGE model that dynamically merges entities' numerical attributes with relation embeddings.
ReaLitE outperforms existing methods in link prediction and node classification tasks.
It supports various numerical aggregations, including a learnable method.

Read Full Article

8 Likes

Arxiv

Image Credit: Arxiv

P2NIA: Privacy-Preserving Non-Iterative Auditing

The emergence of AI legislation has increased the need to assess the ethical compliance of high-risk AI systems.
Traditional auditing methods rely on platforms' APIs, which can compromise privacy and result in data leaks.
P2NIA is a novel auditing scheme that proposes a mutually beneficial collaboration between auditors and platforms.
P2NIA enhances fairness assessments using synthetic or local data, avoiding the challenges of traditional API-based audits.

Read Full Article

Arxiv

116

Image Credit: Arxiv

Detection of Anomalous Vehicular Traffic and Sensor Failures Using Data Clustering Techniques

The study focuses on employing clustering techniques to analyze traffic flow data and detect anomalies, including sensor failures and irregular congestion events.
Multiple clustering approaches are explored, including partitioning and hierarchical methods, paired with various time-series representations and similarity measures.
Hierarchical clustering with symbolic representations is found to provide robust segmentation of traffic patterns, while partitioning methods like k-means and fuzzy c-means yield meaningful results when paired with Dynamic Time Warping.
The proposed anomaly detection strategy successfully identifies sensor malfunctions and abnormal traffic conditions with minimal false positives, making it useful for real-time monitoring.

Read Full Article

7 Likes

For uninterrupted reading, download the app