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

A naukri.com initiative

New

Home

ML News

Arxiv

228

Image Credit: Arxiv

Symbolic Representation for Any-to-Any Generative Tasks

Researchers have proposed a symbolic generative task description language and inference engine for any-to-any multimodal tasks.
The framework utilizes a symbolic representation comprising functions, parameters, and topological logic.
A pre-trained language model is used to map natural language instructions to symbolic workflows without the need for training.
Experiments demonstrate strong performance, efficiency, editability, and interruptibility of the proposed method for multimodal generative tasks.

Read Full Article

13 Likes

Arxiv

356

Image Credit: Arxiv

Signal Recovery from Random Dot-Product Graphs Under Local Differential Privacy

The paper proposes a method for recovering latent information from graphs under local differential privacy.
The authors show that a standard local differential privacy mechanism induces a specific geometric distortion in the latent positions of generalized random dot-product graphs.
They demonstrate that consistent recovery of the latent positions can be achieved by adjusting the statistical inference procedure for the privatized graph.
The proposed procedure is nearly minimax-optimal under local edge differential privacy constraints and allows for consistent recovery of geometric and topological information.

Read Full Article

21 Likes

Arxiv

363

Image Credit: Arxiv

HeRB: Heterophily-Resolved Structure Balancer for Graph Neural Networks

Recent research has witnessed the progress of Graph Neural Networks (GNNs) in graph data representation.
GNNs face the challenge of structural imbalance, and existing solutions do not account for graph heterophily.
The HeRB (Heterophily-Resolved Structure Balancer) method is proposed to address this problem.
Experimental results show that HeRB outperforms other methods on benchmark datasets.

Read Full Article

21 Likes

Arxiv

367

Image Credit: Arxiv

ExOSITO: Explainable Off-Policy Learning with Side Information for Intensive Care Unit Blood Test Orders

Ordering a minimal subset of lab tests for patients in the intensive care unit (ICU) can be challenging.
A novel method called ExOSITO has been developed, which combines off-policy learning with privileged information to identify the optimal set of ICU lab tests to order.
ExOSITO creates an interpretable assistive tool for clinicians to order lab tests by considering both the observed and predicted future status of each patient.
The learned policy function provides interpretable clinical information and reduces costs without omitting any vital lab orders, outperforming both a physician's policy and prior approaches to this practical problem.

Read Full Article

22 Likes

Discover more

Arxiv

131

Image Credit: Arxiv

The Ultimate Cookbook for Invisible Poison: Crafting Subtle Clean-Label Text Backdoors with Style Attributes

Backdoor attacks on text classifiers can be made more effective and subtle by crafting trigger attributes that are indistinguishable from normal texts.
Previous attacks often rely on triggers that are ungrammatical or unusual, making them easily detectable by human annotators during manual inspection.
The study proposes 'AttrBkd', a method for crafting subtle trigger attributes by extracting fine-grained attributes from existing backdoor attacks.
Human evaluations show that AttrBkd with baseline-derived attributes is more effective and subtle compared to original baseline backdoor attacks.

Read Full Article

7 Likes

Arxiv

149

Image Credit: Arxiv

Machine learning-based condition monitoring of powertrains in modern electric drives

Advances in digitalization have transformed the industrial sector.
Machine learning models can be integrated into modern electric drives to enhance performance.
A data-driven thermal model of a power module was developed using data from electric drives.
Different approaches, including linear models and deep neural networks, were evaluated for estimating case temperature.

Read Full Article

9 Likes

Arxiv

262

Image Credit: Arxiv

Class-Conditional Distribution Balancing for Group Robust Classification

Spurious correlations pose a challenge for robust real-world generalization in machine learning.
Existing methods address this issue by maximizing group-balanced or worst-group accuracy, but they heavily rely on expensive bias annotations.
A new method is proposed to tackle spurious correlations by reframing them as imbalances or mismatches in class-conditional distributions, eliminating the need for bias annotations or predictions.
The proposed method achieves class-conditional distribution balancing and produces a debiased data distribution for classification, delivering state-of-the-art performance.

Read Full Article

15 Likes

Arxiv

358

Image Credit: Arxiv

Collaborative Multi-Agent Reinforcement Learning for Automated Feature Transformation with Graph-Driven Path Optimization

Feature transformation methods aim to find an optimal mathematical feature-feature crossing process that generates high-value features and improves the performance of downstream machine learning tasks.
TCTO is a collaborative multi-agent reinforcement learning framework that automates feature engineering through graph-driven path optimization.
The framework utilizes an evolving interaction graph to model features as nodes and transformations as edges.
Comprehensive experiments and case studies demonstrate the efficacy and adaptability of TCTO, showing superior performance across a range of datasets.

Read Full Article

21 Likes

Arxiv

194

Image Credit: Arxiv

Doubly Adaptive Social Learning

In social learning, a network of agents assigns probability scores (beliefs) to hypotheses of interest for generating streaming data.
The traditional approach may fail to adapt to dynamic drifts in the data, leading to incorrect decision making.
The Doubly Adaptive Social Learning (A2SL) strategy is proposed to overcome this limitation by incorporating two adaptation stages.
The A2SL strategy ensures consistent learning by tracking and adapting to changes in the decision model and the true hypothesis over time.

Read Full Article

11 Likes

Arxiv

104

Image Credit: Arxiv

CHASe: Client Heterogeneity-Aware Data Selection for Effective Federated Active Learning

Active learning reduces human annotation costs for machine learning systems by strategically selecting the most informative unlabeled data for annotation, but performing it individually may still be insufficient due to restricted data diversity and annotation budget.
Federated Active Learning (FAL) facilitates collaborative data selection and model training while preserving the confidentiality of raw data samples.
Existing FAL methods fail to account for the heterogeneity of data distribution across clients and the associated fluctuations in global and local model parameters, adversely affecting model accuracy.
To address these challenges, CHASe (Client Heterogeneity-Aware Data Selection) is proposed, which focuses on identifying unlabeled samples with high epistemic variations and incorporates techniques for tracking variations, calibrating decision boundaries, and enhancing data selection efficiency.

Read Full Article

6 Likes

Arxiv

108

Image Credit: Arxiv

HMI: Hierarchical Knowledge Management for Efficient Multi-Tenant Inference in Pretrained Language Models

The significant computational demands of pretrained language models (PLMs) pose a challenge in efficient inference, especially in multi-tenant environments.
HMI (Hierarchical knowledge management-based Multi-tenant Inference) is introduced as a system to manage tenants with distinct PLMs resource-efficiently.
HMI utilizes hierarchical PLMs (hPLMs) by categorizing PLM knowledge into general, domain-specific, and task-specific, reducing GPU memory usage per tenant.
System optimizations like hierarchical knowledge prefetching and parallel implementations improve resource utilization and inference throughput in HMI.

Read Full Article

6 Likes

Arxiv

311

Image Credit: Arxiv

Evaluating Time Series Models for Urban Wastewater Management: Predictive Performance, Model Complexity and Resilience

Climate change increases the frequency of extreme rainfall, straining urban infrastructures, especially Combined Sewer Systems (CSS).
Machine Learning (ML) offers cost-efficient alternatives to traditional physics-based models for urban wastewater management.
Neural Network architectures are evaluated for CSS time series forecasting, considering predictive performance, model complexity, and resilience to perturbations.
Local models provide sufficient resilience in decentralized scenarios, ensuring robust modeling of urban infrastructure for sustainable wastewater management.

Read Full Article

18 Likes

Arxiv

Image Credit: Arxiv

GRANITE : a Byzantine-Resilient Dynamic Gossip Learning Framework

Gossip Learning (GL) is a decentralized learning paradigm where users iteratively exchange and aggregate models with a small set of neighboring peers.
GRANITE is a framework for robust learning over sparse, dynamic graphs in the presence of Byzantine nodes.
GRANITE relies on a History-aware Byzantine-resilient Peer Sampling protocol (HaPS) to reduce adversarial influence over time.
Empirical results show that GRANITE maintains convergence with up to 30% Byzantine nodes and improves learning speed in sparser graphs.

Read Full Article

3 Likes

Arxiv

292

Image Credit: Arxiv

Goal-Oriented Time-Series Forecasting: Foundation Framework Design

Traditional time-series forecasting often focuses only on minimizing prediction errors, ignoring the specific requirements of real-world applications that employ them.
A new training methodology is presented in this paper which allows a forecasting model to dynamically adjust its focus based on the importance of forecast ranges specified by the end application.
Unlike previous methods, this approach breaks down predictions over the entire signal range into smaller segments, which are then dynamically weighted and combined to produce accurate forecasts.
Testing on standard datasets, including a new dataset from wireless communication, showed that this method not only improves prediction accuracy but also enhances the performance of end applications using the forecasting model.

Read Full Article

17 Likes

Arxiv

307

Image Credit: Arxiv

Combining GCN Structural Learning with LLM Chemical Knowledge for or Enhanced Virtual Screening

Virtual screening plays a critical role in modern drug discovery by enabling the identification of promising candidate molecules for experimental validation.
Traditional machine learning methods such as support vector machines (SVM) and XGBoost rely on predefined molecular representations, often leading to information loss and potential bias.
In contrast, utilizing Graph Convolutional Networks (GCNs) and Large Language Models (LLMs) can provide a more expressive and unbiased alternative by operating directly on molecular graphs and capturing complex chemical patterns.
A hybrid architecture that integrates GCNs with LLM-derived embeddings has been proposed, achieving superior results and outperforming standalone GCN, XGBoost, and SVM baselines.

Read Full Article

18 Likes

For uninterrupted reading, download the app