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

Unveiling the Threat of Fraud Gangs to Graph Neural Networks: Multi-Target Graph Injection Attacks against GNN-Based Fraud Detectors

Graph neural networks (GNNs) are used for fraud detection, but attacks against GNN-based fraud detectors are understudied.
This study focuses on multi-target graph injection attacks by fraud gangs aiming to evade detection.
They propose MonTi, a transformer-based attack model that generates attributes and edges of attack nodes simultaneously.
Experimental results show that MonTi outperforms existing methods on real-world graphs.

Read Full Article

3 Likes

Arxiv

289

Image Credit: Arxiv

GeFL: Model-Agnostic Federated Learning with Generative Models

Federated learning (FL) is a promising paradigm in distributed learning while preserving user privacy.
The increasing size of models makes it difficult for users with limited resources to participate in FL.
The authors propose GeFL, a model-agnostic federated learning approach that incorporates a generative model to aggregate global knowledge across users with heterogeneous models.
Experimental results show improved performance of GeFL compared to baselines, along with a novel framework GeFL-F that addresses privacy and scalability concerns.

Read Full Article

17 Likes

Arxiv

Image Credit: Arxiv

An Empirical Analysis of Federated Learning Models Subject to Label-Flipping Adversarial Attack

This paper presents an empirical analysis of federated learning models subjected to label-flipping adversarial attacks.
Various models such as MLR, SVC, MLP, CNN, RNN, Random Forest, XGBoost, and LSTM are considered.
Experiments are conducted with different percentages of adversarial clients and flipped labels.
The study reveals variations in the robustness of models to these attack vectors.

Read Full Article

1 Like

Arxiv

Image Credit: Arxiv

Characterizations of Language Generation With Breadth

Researchers study language generation in the limit, building on classical works of Gold and Angluin.
Different notions of consistent generation with breadth are proposed, including exact breadth, approximate breadth, and unambiguous generation.
Language generation with exact breadth is characterized by Angluin's condition for identification.
Unambiguous generation is also characterized by Angluin's condition, and there is a separation between stable and unstable generation with approximate breadth.

Read Full Article

3 Likes

Discover more

Arxiv

Image Credit: Arxiv

Graph Structure Learning for Spatial-Temporal Imputation: Adapting to Node and Feature Scales

Spatial-temporal data often have missing values, making data analysis challenging.
Existing methods assume the same spatial relationship for all features across different locations.
The multi-scale Graph Structure Learning framework for spatial-temporal Imputation (GSLI) dynamically adapts to diverse spatial correlations.
GSLI incorporates node-scale and feature-scale graph structure learning, prominence modeling, and cross-feature and cross-temporal representation learning.

Read Full Article

2 Likes

Arxiv

279

Image Credit: Arxiv

Consistency Checks for Language Model Forecasters

Forecasting is a difficult task to evaluate as the ground truth is only known in the future.
A new consistency check framework is proposed to benchmark and evaluate forecasters instantaneously.
A general consistency metric based on arbitrage is introduced to measure the consistency of predictions.
An automated evaluation system is built to measure consistency and correlate it with ground truth Brier scores.

Read Full Article

16 Likes

Arxiv

166

Image Credit: Arxiv

FedVCK: Non-IID Robust and Communication-Efficient Federated Learning via Valuable Condensed Knowledge for Medical Image Analysis

Federated learning has become a promising solution for collaboration among medical institutions. However, data owned by each institution would be highly heterogeneous and the distribution is always non-independent and identical distribution (non-IID), resulting in client drift and unsatisfactory performance.
In this paper, a novel federated learning method called FedVCK (Federated learning via Valuable Condensed Knowledge) is proposed. It aims to tackle the non-IID problem within limited communication budgets effectively.
FedVCK condenses the knowledge of each client into a small dataset and enhances the condensation procedure with latent distribution constraints. It prevents unnecessary repetition of homogeneous knowledge and minimizes the frequency of communications required.
Experiments showed that FedVCK outperforms state-of-the-art methods in various medical tasks, demonstrating its non-IID robustness and communication efficiency.

Read Full Article

10 Likes

Arxiv

Image Credit: Arxiv

Structure Learning in Gaussian Graphical Models from Glauber Dynamics

Gaussian graphical model selection is addressed under observations from the Glauber dynamics.
The Gibbs sampler, a Markov chain, is used to update variables in the model.
The proposed algorithm for structure learning in Gaussian graphical models under Glauber dynamics is introduced.
The algorithm shows good computational and statistical performance and is nearly minimax optimal for various problems.

Read Full Article

5 Likes

Arxiv

213

Image Credit: Arxiv

GOPT: Generalizable Online 3D Bin Packing via Transformer-based Deep Reinforcement Learning

Researchers propose GOPT, a generalizable online 3D Bin Packing approach via Transformer-based deep reinforcement learning (DRL), for robotic object packing.
GOPT consists of a Placement Generator module and a Packing Transformer, enabling generalization across multiple environments with different bin dimensions.
Extensive experiments demonstrate that GOPT outperforms baselines and exhibits excellent generalization capabilities.
The practical applicability of GOPT is showcased through deployment with a robot.

Read Full Article

12 Likes

Arxiv

Image Credit: Arxiv

KG4Diagnosis: A Hierarchical Multi-Agent LLM Framework with Knowledge Graph Enhancement for Medical Diagnosis

KG4Diagnosis is a hierarchical multi-agent framework for medical diagnosis.
It combines Large Language Models (LLMs) with automated knowledge graph construction.
The framework includes a general practitioner (GP) agent for initial assessment and specialized agents for in-depth diagnosis.
KG4Diagnosis facilitates the generation of knowledge graphs and can incorporate new diseases and medical knowledge.

Read Full Article

3 Likes

Arxiv

269

Image Credit: Arxiv

EEG-Based Mental Imagery Task Adaptation via Ensemble of Weight-Decomposed Low-Rank Adapters

Researchers have proposed a novel technique called EDoRA for adapting EEG-based mental imagery tasks.
The technique, based on parameter-efficient fine-tuning (PEFT), aims to improve neural decoding in Brain Computer Interfaces (BCIs).
The study focuses on two mental imagery tasks - speech imagery and motor imagery, which are important for post-stroke neuro-rehabilitation.
In experiments using publicly available datasets, EDoRA outperformed full fine-tuning and other PEFT methods for mental imagery EEG classification.

Read Full Article

16 Likes

Arxiv

183

Image Credit: Arxiv

RUL forecasting for wind turbine predictive maintenance based on deep learning

Predictive maintenance (PdM) is pursued to reduce wind farm operation and maintenance costs by accurately predicting the remaining useful life (RUL) and strategically scheduling maintenance.
A novel deep learning (DL) methodology, ForeNet-2d and ForeNet-3d, is proposed for future RUL forecasting.
The models successfully forecast RUL for seven wind turbine failures with a 2-week forecast window.
The methodology offers a substantial time frame for remote wind turbines maintenance, enabling the practical implementation of PdM.

Read Full Article

11 Likes

Arxiv

169

Image Credit: Arxiv

CPFI-EIT: A CNN-PINN Framework for Full-Inverse Electrical Impedance Tomography on Non-Smooth Conductivity Distributions

This paper presents a hybrid learning framework, CPFI-EIT, that combines convolutional neural networks (CNNs) and physics-informed neural networks (PINNs) for full-inverse electrical impedance tomography (EIT).
EIT is a noninvasive imaging technique used for reconstructing the spatial distribution of internal conductivity based on boundary voltage measurements.
The proposed framework employs a forward CNN-based supervised network for mapping differential boundary voltage measurements to a discrete potential distribution, and an inverse PINN-based unsupervised network for enforcing PDE constraints for conductivity reconstruction.
The framework utilizes discrete numerical differentiation to bridge the forward and inverse networks, improving modularity and reducing computational demands.

Read Full Article

10 Likes

Arxiv

163

Image Credit: Arxiv

Decoding individual words from non-invasive brain recordings across 723 participants

A novel deep learning pipeline has been introduced to decode individual words from non-invasive brain recordings.
The approach has been trained and evaluated on 723 participants exposed to five million words in different languages.
The model outperforms existing methods consistently across participants, devices, languages, and tasks.
The study highlights the importance of the recording device, experimental protocol, and amount of data used for training and testing.

Read Full Article

9 Likes

Arxiv

Image Credit: Arxiv

A Beginner's Guide to Power and Energy Measurement and Estimation for Computing and Machine Learning

Concerns about the environmental footprint of machine learning are increasing.
Most ML researchers and developers do not incorporate energy measurement in their work practices.
This paper introduces considerations for using energy measurement tools and interpreting energy estimates.
It calls for improving measurement methods and standards for robust comparisons between hardware and software environments.

Read Full Article

5 Likes

For uninterrupted reading, download the app