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

A naukri.com initiative

New

Home

ML News

Arxiv

Image Credit: Arxiv

Learning Sign Language Representation using CNN LSTM, 3DCNN, CNN RNN LSTM and CCN TD

Existing Sign Language Learning applications focus on the demonstration of the sign in the hope that the student will copy a sign correctly.
This paper explores algorithms for real-time, video sign translation, and grading of sign language accuracy for new users.
The study compares popular algorithms including CNN and 3DCNN on Trinidad and Tobago Sign Language and American Sign Language datasets.
The 3DCNN algorithm achieved 91% accuracy in the TTSL dataset and 83% accuracy in the ASL dataset, making it the best performing neural network algorithm.

Read Full Article

Arxiv

306

Image Credit: Arxiv

Developing Cryptocurrency Trading Strategy Based on Autoencoder-CNN-GANs Algorithms

This paper utilizes machine learning algorithms to forecast and analyze financial time series.
It employs a denoising autoencoder to filter out random noise fluctuations from the main contract price data.
The filtered data is then processed using one-dimensional convolution to extract key information.
A GANs network is utilized to further analyze the filtered and dimensionality-reduced price data for predicting significant price changes in real-time.

Read Full Article

18 Likes

Arxiv

289

Image Credit: Arxiv

Sharper Error Bounds in Late Fusion Multi-view Clustering Using Eigenvalue Proportion

Late Fusion Multi-View Clustering (LFMVC) aims to integrate complementary information from multiple views to enhance clustering performance.
Current LFMVC methods struggle with noisy and redundant partitions and often fail to capture high-order correlations across views.
A novel theoretical framework is presented for analyzing the generalization error bounds of multiple kernel k-means, leveraging local Rademacher complexity and principal eigenvalue proportions.
Experimental results on benchmark datasets confirm that the proposed approach outperforms state-of-the-art methods in clustering performance and robustness.

Read Full Article

17 Likes

Arxiv

204

Image Credit: Arxiv

Accelerating AIGC Services with Latent Action Diffusion Scheduling in Edge Networks

AIGC services in edge networks face challenges due to high service delay and resource demands.
A novel method called LAD-TS is proposed for task scheduling in edge networks.
LAD-TS minimizes service delays by leveraging diffusion models and reinforcement learning.
DEdgeAI, a prototype edge system, demonstrates shorter service delays compared to existing AIGC platforms.

Read Full Article

12 Likes

Discover more

Arxiv

184

Image Credit: Arxiv

On the Effectiveness of Adversarial Training on Malware Classifiers

Adversarial Training (AT) has been widely applied to harden malware classifiers against adversarial evasive attacks.
The effectiveness of AT in identifying and strengthening vulnerable areas of the model's decision space while maintaining high performance on clean data remains underexplored.
Robustness achieved by AT has often been assessed against unrealistic or weak adversarial attacks, negatively affecting performance on clean data.
Factors such as data, feature representations, classifiers, and robust optimization settings influence the effectiveness of AT.

Read Full Article

11 Likes

Arxiv

Image Credit: Arxiv

Towards Macro-AUC oriented Imbalanced Multi-Label Continual Learning

In Continual Learning (CL), there has been limited research on Multi-Label Learning (MLL).
MLL datasets are often class-imbalanced, making it challenging in CL.
To optimize Macro-AUC in MLCL, a new memory replay-based method called RLDAM is proposed.
Experimental results demonstrate the effectiveness of the proposed method over baselines.

Read Full Article

Arxiv

111

Image Credit: Arxiv

Robust Semi-Supervised Learning in Open Environments

Semi-supervised learning (SSL) aims to improve performance by exploiting unlabeled data when labels are scarce.
Conventional SSL studies assume consistent important factors between labeled and unlabeled data in close environments.
This paper discusses robust SSL in open environments with inconsistent important factors.
Advances in techniques addressing label, feature, and data distribution inconsistency in SSL are introduced, along with evaluation benchmarks.

Read Full Article

6 Likes

Arxiv

Image Credit: Arxiv

Free the Design Space of Equivariant Graph Neural Networks: High-Rank Irreducible Cartesian Tensor Decomposition and Bases of Equivariant Spaces

Irreducible Cartesian tensors (ICTs) are important in the design of equivariant graph neural networks and theoretical chemistry.
The ICT decomposition and basis for equivariant spaces are challenging to obtain for high-order tensors.
Researchers have achieved an explicit ICT decomposition for $n=5$ with factorial complexity and obtained decomposition matrices for ICTs up to rank $n=9$ with reduced complexity.
They used path matrices obtained through chain-like contraction with Clebsch-Gordan matrices to establish an orthonormal change-of-basis matrix and a complete orthogonal basis for the equivariant space.

Read Full Article

1 Like

Arxiv

Image Credit: Arxiv

NoiseHGNN: Synthesized Similarity Graph-Based Neural Network For Noised Heterogeneous Graph Representation Learning

Real-world graph data environments often contain noise that affects the effectiveness of graph representation and downstream learning tasks.
Existing methods for homogeneous graphs synthesize a similarity graph based on original node features to correct the structure of the noisy graph.
However, similar nodes in heterogeneous graphs do not have direct links, posing a challenge for noise correction in heterogeneous graph learning.
This paper proposes a novel synthesized similarity-based graph neural network for learning from noisy heterogeneous graphs, achieving state-of-the-art results in various real-world datasets.

Read Full Article

Arxiv

144

Image Credit: Arxiv

On the Local Complexity of Linear Regions in Deep ReLU Networks

Researchers have defined the local complexity of a neural network with continuous piecewise linear activations as a measure of the density of linear regions over an input data distribution.
They have theoretically shown that ReLU networks learning low-dimensional feature representations have a lower local complexity.
This connects recent empirical observations on feature learning with concrete properties of the learned functions.
The local complexity also serves as an upper bound on the total variation of the function over the input data distribution, linking feature learning to adversarial robustness.
The researchers also consider how optimization drives ReLU networks towards solutions with lower local complexity, contributing a theoretical framework for understanding geometric properties of ReLU networks in relation to learning.

Read Full Article

8 Likes

Arxiv

131

Image Credit: Arxiv

Semi-supervised Credit Card Fraud Detection via Attribute-Driven Graph Representation

Researchers propose a semi-supervised graph neural network for credit card fraud detection.
The method utilizes transaction records to construct a temporal transaction graph.
A Gated Temporal Attention Network (GTAN) is used to learn transaction representations.
The proposed method outperforms other baselines on fraud detection datasets.

Read Full Article

7 Likes

Arxiv

Image Credit: Arxiv

Navigating Data Corruption in Machine Learning: Balancing Quality, Quantity, and Imputation Strategies

Data corruption, including missing and noisy data, poses challenges in machine learning.
This study investigates the effects of data corruption on model performance and explores strategies to mitigate them.
Results show that noisy data has a more severe impact than missing data, especially in sequential decision-making tasks.
Increasing dataset size helps mitigate but cannot fully overcome the effects of data corruption.

Read Full Article

1 Like

Arxiv

319

Image Credit: Arxiv

Data-Driven Self-Supervised Graph Representation Learning

Self-supervised graph representation learning (SSGRL) is a representation learning paradigm used to reduce or avoid manual labeling.
Existing methods of graph data augmentation rely on heuristics and are effective only within specific application domains.
This study proposes a data-driven SSGRL approach that automatically learns graph augmentation from the graph's signal.
The proposed method outperforms baselines and performs similarly to semi-supervised methods in various experiments.

Read Full Article

19 Likes

Arxiv

151

Image Credit: Arxiv

Point-DeepONet: A Deep Operator Network Integrating PointNet for Nonlinear Analysis of Non-Parametric 3D Geometries and Load Conditions

Nonlinear structural analyses in engineering often require extensive finite element simulations.
Point-DeepONet is an operator-learning-based surrogate that integrates PointNet into the DeepONet framework.
It accurately predicts three-dimensional displacement and von Mises stress fields without mesh parameterization or retraining.
Point-DeepONet provides predictions in mere seconds, approximately 400 times faster than nonlinear finite element analyses.

Read Full Article

9 Likes

Arxiv

141

Image Credit: Arxiv

Hypergraph Attacks via Injecting Homogeneous Nodes into Elite Hyperedges

Recent studies have shown that Hypergraph Neural Networks (HGNNs) are vulnerable to adversarial attacks.
A novel framework called Hypergraph Attacks via Injecting Homogeneous Nodes into Elite Hyperedges (IE-Attack) is proposed to tackle these challenges.
IE-Attack utilizes node spanning in the hypergraph to identify hyperedges to be injected and generates a homogeneous node with the group identity of hyperedges using Kernel Density Estimation (KDE).
By injecting the homogeneous node into elite hyperedges, IE-Attack improves attack performance and enhances the imperceptibility of attacks.

Read Full Article

8 Likes

For uninterrupted reading, download the app