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

Enhancing Time Series Forecasting via Multi-Level Text Alignment with LLMs

The adaptation of large language models (LLMs) to time series forecasting poses unique challenges.
A multi-level text alignment framework for time series forecasting using LLMs is proposed.
The method decomposes time series into trend, seasonal, and residual components.
Experiments show that the proposed method outperforms state-of-the-art models in accuracy and interpretability.

Read Full Article

Arxiv

294

Image Credit: Arxiv

Min-Max Optimisation for Nonconvex-Nonconcave Functions Using a Random Zeroth-Order Extragradient Algorithm

This study explores the performance of the ZO-EG scheme for min-max optimization problems with NonConvex-NonConcave (NC-NC) objective functions.
The study considers both unconstrained and constrained, differentiable and non-differentiable settings.
For the unconstrained problem, the ZO-EG algorithm is proven to converge to the neighborhood of an ε-stationary point of the NC-NC objective function.
For the constrained problem, the study introduces the concept of proximal variational inequalities and provides analogous results to the unconstrained case.

Read Full Article

17 Likes

Arxiv

Image Credit: Arxiv

Leveraging LLMs for Multimodal Retrieval-Augmented Radiology Report Generation via Key Phrase Extraction

Automated radiology report generation (RRG) using large language models (LLMs) has the potential to reduce radiologists' workload.
A retrieval-augmented generation approach is proposed, leveraging multimodal retrieval and LLMs for radiology report generation.
The method extracts key phrases from radiology reports using LLMs, focusing on essential diagnostic information.
The approach achieves state-of-the-art results on evaluation metrics and demonstrates robust generalization for multi-view report generation.

Read Full Article

2 Likes

Arxiv

Image Credit: Arxiv

RadZero: Similarity-Based Cross-Attention for Explainable Vision-Language Alignment in Radiology with Zero-Shot Multi-Task Capability

RadZero is a similarity-based cross-attention framework designed for vision-language alignment in radiology with zero-shot multi-task capability.
It addresses the challenges of effectively utilizing complex radiology reports, relying on low-resolution images, and limited interpretability in attention mechanisms.
RadZero leverages large language models to extract semantic sentences from radiology reports and employs a multi-positive contrastive learning strategy to capture relationships between images and textual descriptions.
Experimental results show that RadZero outperforms state-of-the-art methods in zero-shot classification, grounding, segmentation, and improves explainability in vision-language alignment.

Read Full Article

3 Likes

Discover more

Arxiv

176

Image Credit: Arxiv

Enhancing Player Enjoyment with a Two-Tier DRL and LLM-Based Agent System for Fighting Games

A Two-Tier DRL and LLM-Based Agent System for Fighting Games has been proposed to enhance player enjoyment.
The system consists of a task-oriented network architecture, hybrid training, and modularized reward functions for producing diverse and skilled DRL agents.
A Large Language Model Hyper-Agent is used to dynamically select suitable DRL opponents based on players' data and feedback.
Experiments show significant improvements in executing advanced skills and overall enjoyment, validating the effectiveness of the system.

Read Full Article

10 Likes

Arxiv

180

Image Credit: Arxiv

Conditional Data Synthesis Augmentation

Conditional Data Synthesis Augmentation (CoDSA) is a framework that uses generative models to synthesize diverse and well-distributed data for improving machine learning and statistical analysis.
CoDSA focuses on addressing data limitations and biases by generating synthetic samples that capture the conditional distributions of the original data, particularly in under-sampled or high-interest regions.
CoDSA leverages transfer learning to enhance the realism of synthetic data and increase sample density in sparse areas, preserving inter-modal relationships and improving domain adaptation and generalization of models.
Experiments indicate that CoDSA consistently outperforms non-adaptive augmentation strategies and state-of-the-art baselines in both supervised and unsupervised settings.

Read Full Article

10 Likes

Arxiv

102

Image Credit: Arxiv

A Mechanism-Learning Deeply Coupled Model for Remote Sensing Retrieval of Global Land Surface Temperature

Land surface temperature (LST) retrieval from remote sensing data is pivotal for analyzing climate processes and surface energy budgets.
A deeply coupled framework integrating mechanistic modeling and machine learning is proposed to enhance the accuracy and generalizability of single-channel LST retrieval.
Global validation demonstrated a 30% reduction in root-mean-square error versus standalone methods, and a 53% improvement in mean absolute error under extreme humidity.
Continental-scale tests across five continents confirmed the superior generalizability of this model.

Read Full Article

6 Likes

Arxiv

130

Image Credit: Arxiv

AI-Slop to AI-Polish? Aligning Language Models through Edit-Based Writing Rewards and Test-time Computation

AI-generated text is proliferating across domains, from creative writing and journalism to marketing content and scientific articles.
Writing quality assessment has received less attention from the community, in part because it is fundamentally subjective and requires expertise.
Researchers have introduced the Writing Quality Benchmark (WQ) by consolidating five writing-preference datasets into 4,729 writing quality judgments.
Specialized Writing Quality Reward Models (WQRM) demonstrate strong generalization and practical benefits during inference, producing preferred writing samples.

Read Full Article

7 Likes

Arxiv

Image Credit: Arxiv

PhaseGen: A Diffusion-Based Approach for Complex-Valued MRI Data Generation

A novel complex-valued diffusion model, PhaseGen, has been introduced for generating synthetic MRI raw data conditioned on magnitude images.
PhaseGen allows pretraining for models that require k-Space information, enabling downstream tasks such as tumor segmentation and classification.
The evaluation of PhaseGen on skull-stripping and MRI reconstruction tasks showed significant improvements in segmentation accuracy and reconstruction when combined with limited real-world data.
This approach bridges the gap between magnitude-based datasets and the complex-valued nature of MRI raw data, enabling more accurate and efficient diagnostic tasks.

Read Full Article

5 Likes

Arxiv

256

Image Credit: Arxiv

Privacy-Preserving Vertical K-Means Clustering

Clustering is a fundamental data processing task used for grouping records based on one or more features.
In the vertically partitioned setting, computing distances between records requires access to all distributed features, which may be privacy-sensitive and cannot be directly shared.
A novel solution based on homomorphic encryption and differential privacy (DP) is proposed, reducing communication complexity and ensuring privacy with minimal impact on utility.
The proposed solution allows for practical deployments even in WAN settings, maintaining accuracy comparable to plaintext k-means algorithms.

Read Full Article

15 Likes

Arxiv

365

Image Credit: Arxiv

Do LLMs Understand Your Translations? Evaluating Paragraph-level MT with Question Answering

A new framework, TREQA, is introduced for evaluating translation quality at the paragraph-level.
TREQA assesses how accurately candidate translations answer reading comprehension questions that target key information in the original source or reference texts.
In challenging domains, TREQA is shown to be competitive with, and sometimes outperforms, state-of-the-art neural and LLM-based metrics in ranking alternative paragraph-level translations.
The generated questions and answers provide interpretability by effectively targeting translation errors identified in evaluated datasets.

Read Full Article

21 Likes

Arxiv

Image Credit: Arxiv

Stochastic Smoothed Primal-Dual Algorithms for Nonconvex Optimization with Linear Inequality Constraints

Researchers propose smoothed primal-dual algorithms for solving nonconvex optimization problems with linear inequality constraints.
The algorithms are single-loop and utilize one stochastic gradient based on one sample per iteration.
Estimation of the gradient of the Moreau envelope is performed using a stochastic primal-dual augmented Lagrangian method.
The algorithms provide optimal sample complexity guarantees for obtaining ɛ-stationary points and offer an improved complexity by using variance reduction and expected smoothness assumption.

Read Full Article

1 Like

Arxiv

155

Image Credit: Arxiv

Beating Transformers using Synthetic Cognition

A novel approach called Synthetic Cognition has been proposed to develop cognitive architectures for Artificial General Intelligence.
Researchers explore the use of Synthetic Cognition to develop episodic reactive behaviors and propose a mechanism to deal with sequences.
In experiments, the proposed method outperforms DNA foundation models in DNA sequence classification tasks.
The study achieves the goals of expanding Synthetic Cognition to sequences and beating the Transformer architecture for sequence classification.

Read Full Article

9 Likes

Arxiv

365

Image Credit: Arxiv

Data Requirement Goal Modeling for Machine Learning Systems

Machine Learning (ML) has been integrated into various software and systems.
Data requirement goal modeling (DRGM) is proposed to guide non-experts in identifying data requirements for ML systems.
DRGM is built by surveying white literature and allows customization based on different project needs.
The proposed approach aligns with real-world projects, demonstrating its practicality and effectiveness.

Read Full Article

21 Likes

Arxiv

344

Image Credit: Arxiv

Conformalized Generative Bayesian Imaging: An Uncertainty Quantification Framework for Computational Imaging

A new framework for uncertainty quantification in computational imaging has been introduced.
The proposed framework combines generative model-based methods and Bayesian neural networks.
The framework can jointly quantify aleatoric and epistemic uncertainties in image reconstruction.
Experiments on different imaging problems demonstrate the effectiveness and calibration of the proposed framework.

Read Full Article

20 Likes

For uninterrupted reading, download the app