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

A naukri.com initiative

New

Home

ML News

Arxiv

109

Image Credit: Arxiv

Quantum Machine Learning: Unveiling Trends, Impacts through Bibliometric Analysis

Quantum Machine Learning (QML) is the intersection of quantum computing and machine learning, unlocking unprecedented capabilities.
A comprehensive bibliometric analysis of QML research from 2000 to 2023 reveals notable trends and impact factors.
Prominent contributors in QML research include the United States and China, with substantial publication and citation metrics.
QML is currently in a formative stage, characterized by robust scholarly activity and ongoing development.

Read Full Article

6 Likes

Arxiv

348

Image Credit: Arxiv

Harnessing Equivariance: Modeling Turbulence with Graph Neural Networks

This work proposes a novel methodology for turbulence modeling in Large Eddy Simulation (LES) based on Graph Neural Networks (GNNs).
The proposed approach embeds the symmetries of the Navier-Stokes equations into the model architecture, resulting in a symmetry-preserving simulation setup.
The GNN models are trained successfully in actual simulations using Reinforcement Learning (RL), ensuring consistency with the underlying LES formulation and discretization.
The GNN model demonstrates the potential for turbulence modeling, particularly in the context of LES and RL.

Read Full Article

20 Likes

Arxiv

352

Image Credit: Arxiv

Gradient-based Sample Selection for Faster Bayesian Optimization

Bayesian optimization is an effective technique for black-box optimization, but its applicability is limited for large-budget problems due to computational complexity.
Researchers propose Gradient-based Sample Selection Bayesian Optimization (GSSBO) to improve the computational efficiency of Bayesian optimization.
GSSBO constructs the Gaussian process (GP) model on a selected set of samples using gradient information.
The approach reduces the computational cost of GP fitting while maintaining optimization performance similar to baseline methods.

Read Full Article

21 Likes

Arxiv

Image Credit: Arxiv

Search-contempt: a hybrid MCTS algorithm for training AlphaZero-like engines with better computational efficiency

A new hybrid variant of the Monte Carlo Tree Search (MCTS) algorithm, called search-contempt, has been introduced to improve the computational efficiency of training AlphaZero-like engines.
The search-contempt algorithm alters the distribution of positions generated in self-play, emphasizing more challenging positions.
It has been demonstrated to significantly enhance the strength of engines in Odds Chess.
The use of search-contempt enables the possibility of training self-play based engines with fewer training games, lowering the computational and financial costs.

Read Full Article

5 Likes

Discover more

Arxiv

344

Image Credit: Arxiv

Adaptive Detection of Fast Moving Celestial Objects Using a Mixture of Experts and Physical-Inspired Neural Network

A novel algorithm is presented for detecting fast moving celestial objects within star fields.
The algorithm enhances neural networks by transforming them into physical-inspired neural networks.
It leverages the point spread function and observational mode as priors for accurate detection.
The algorithm is effective in detecting fast moving celestial objects across different observational modes.

Read Full Article

20 Likes

Arxiv

Image Credit: Arxiv

A System for Comprehensive Assessment of RAG Frameworks

A comprehensive evaluation framework called SCARF (System for Comprehensive Assessment of RAG Frameworks) has been introduced to assess the performance of Retrieval Augmented Generation (RAG) systems.
SCARF is designed to provide a black-box approach for evaluating RAG applications in real-world deployment scenarios.
The evaluation framework supports multiple deployment configurations and automated testing across vector databases and Large Language Model (LLM) serving strategies.
SCARF integrates practical considerations such as response coherence, making it a scalable and adaptable solution for researchers and industry professionals.

Read Full Article

5 Likes

Arxiv

340

Image Credit: Arxiv

Smoothed Distance Kernels for MMDs and Applications in Wasserstein Gradient Flows

Negative distance kernels are widely used in maximum mean discrepancies (MMDs) and have shown favorable numerical results in various applications.
However, due to its non-smoothness in x=y, classical theoretical results do not hold true.
A new Lipschitz differentiable kernel is proposed that maintains the favorable properties of the negative distance kernel.
The new kernel performs similarly well as the negative distance kernel in gradient descent methods, but now with theoretical guarantees.

Read Full Article

20 Likes

Arxiv

Image Credit: Arxiv

How do Large Language Models Understand Relevance? A Mechanistic Interpretability Perspective

Recent studies have shown that large language models (LLMs) can assess relevance and support information retrieval (IR) tasks such as document ranking and relevance judgment generation.
In this paper, researchers investigate how different LLM modules contribute to relevance judgment through the lens of mechanistic interpretability.
They analyze the roles of various model components and identify a multi-stage, progressive process in generating relevance judgment.
The findings provide insights into the mechanisms underlying relevance assessment in LLMs, offering implications for leveraging LLMs for information retrieval tasks.

Read Full Article

1 Like

Arxiv

352

Image Credit: Arxiv

The Efficacy of Semantics-Preserving Transformations in Self-Supervised Learning for Medical Ultrasound

A study investigated the impact of data augmentation and preprocessing strategies in self-supervised learning for lung ultrasound.
Semantics-preserving data augmentation resulted in the greatest performance for COVID-19 classification, while cropping-based methods yielded the greatest performance on B-line and pleural effusion object classification tasks.
Increased downstream performance for multiple tasks was observed with semantics-preserving ultrasound image preprocessing.
Guidance regarding data augmentation and preprocessing strategies in self-supervised learning for ultrasound was provided.

Read Full Article

21 Likes

Arxiv

Image Credit: Arxiv

Policy Gradient Converges to the Globally Optimal Policy for Nearly Linear-Quadratic Regulators

Nonlinear control systems with partial information to the decision maker are prevalent in a variety of applications.
This work explores reinforcement learning methods for finding the optimal policy in the nearly linear-quadratic regulator systems.
The cost function of the system is nonconvex, but the study establishes local strong convexity and smoothness in the vicinity of the global optimizer.
A policy gradient algorithm is proposed that is guaranteed to converge to the globally optimal policy with a linear rate.

Read Full Article

2 Likes

Arxiv

323

Image Credit: Arxiv

Understanding and Mitigating the Bias in Sample Selection for Learning with Noisy Labels

The study focuses on understanding and mitigating bias in sample selection for learning with noisy labels.
Existing sample selection methods suffer from both data and training bias.
The research introduces the NoIse-Tolerant Expert Model (ITEM) to address the limitations.
ITEM incorporates a robust network architecture and a mixed sampling strategy to mitigate both training and data bias.

Read Full Article

19 Likes

Arxiv

176

Image Credit: Arxiv

Understanding Contrastive Representation Learning from Positive Unlabeled (PU) Data

Pretext Invariant Representation Learning (PIRL) followed by Supervised Fine-Tuning (SFT) has become a standard paradigm for learning with limited labels.
The Positive Unlabeled (PU) setting involves a small set of labeled positives and a large unlabeled pool containing both positives and negatives.
The Positive Unlabeled Contrastive Learning (puCL) objective integrates weak supervision from labeled positives into the contrastive loss, without access to the class prior.
When the class prior is known, Positive Unlabeled InfoNCE (puNCE) re-weights unlabeled samples as soft positive negative mixtures for better representation learning.

Read Full Article

10 Likes

Arxiv

235

Image Credit: Arxiv

Efficient Heterogeneous Large Language Model Decoding with Model-Attention Disaggregation

Transformer-based large language models (LLMs) can be inefficient in real-world serving due to the expensive accelerators.
To address this, the paper introduces model-attention disaggregation, leveraging cheap, memory-optimized devices for attention operators.
This approach maximizes performance and cost efficiency by tailoring each component to its workload.
Experimental results show that Lamina, an LLM inference system using this approach, can provide higher estimated throughput than existing solutions with similar costs.

Read Full Article

14 Likes

Arxiv

289

Image Credit: Arxiv

POWQMIX: Weighted Value Factorization with Potentially Optimal Joint Actions Recognition for Cooperative Multi-Agent Reinforcement Learning

The Potentially Optimal Joint Actions Weighted QMIX (POWQMIX) algorithm is proposed as an improvement to value function factorization methods in cooperative multi-agent reinforcement learning.
POWQMIX recognizes potentially optimal joint actions and assigns higher weights to corresponding losses during training, increasing the representation capacity of value factorization compared to existing methods.
The algorithm guarantees to recover the optimal policy through its weighted training approach.
Experiments in various environments demonstrate that POWQMIX outperforms state-of-the-art value-based multi-agent reinforcement learning methods.

Read Full Article

17 Likes

Arxiv

100

Image Credit: Arxiv

Demystifying amortized causal discovery with transformers

Supervised learning approaches for causal discovery from observational data often achieve competitive performance despite seemingly avoiding explicit assumptions that traditional methods make for identifiability.
Researchers investigated the transformer-based model CSIvA, exploring its ability to train on synthetic data and transfer to real data.
The study found that constraints on the training data distribution implicitly define a prior on the test observations, and good performance is achieved when there is a good prior on the test data and the underlying model is identifiable.
The study also showed that training on datasets generated from different classes of causal models, even if individually identifiable, improves test generalization, as the ambiguous cases resulting from the mixture of identifiable causal models are unlikely to occur.

Read Full Article

6 Likes

For uninterrupted reading, download the app