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

LISA: Learning-Integrated Space Partitioning Framework for Traffic Accident Forecasting on Heterogeneous Spatiotemporal Data

Traffic accident forecasting is an important task for intelligent transportation management and emergency response systems.
Existing data-driven methods mostly focus on studying homogeneous areas with limited size and fail to handle the heterogeneous accident patterns over space at different scales.
This paper proposes a novel Learning-Integrated Space Partition Framework (LISA) that simultaneously learns partitions while training models, guided by prediction accuracy.
Experimental results using real-world datasets show that LISA captures underlying heterogeneous patterns and improves baseline networks by an average of 13.0%.

Read Full Article

3 Likes

Arxiv

350

Image Credit: Arxiv

A Multi-Fidelity Graph U-Net Model for Accelerated Physics Simulations

Physics-based deep learning frameworks are effective in modeling complex physical systems with generalization capability.
A novel GNN architecture, Multi-Fidelity U-Net, utilizes multi-fidelity methods to enhance GNN model performance.
The proposed approach reduces data requirements and performs better in accuracy compared to benchmark multi-fidelity approaches.
The proposed models provide a feasible alternative for addressing computational and accuracy requirements in time-consuming simulations.

Read Full Article

21 Likes

Arxiv

346

Image Credit: Arxiv

Granger Causality Detection with Kolmogorov-Arnold Networks

Discovering causal relationships in time series data is central in many scientific areas.
Granger causality is a powerful tool for causality detection, but its original formulation is limited to linear relationships.
This study explores the use of Kolmogorov-Arnold networks (KANs) in Granger causality detection, comparing them to multilayer perceptrons (MLP).
The findings suggest that KANs have the potential to outperform MLPs in identifying sparse Granger causal relationships, especially in high-dimensional settings.

Read Full Article

20 Likes

Arxiv

Image Credit: Arxiv

LG-Sleep: Local and Global Temporal Dependencies for Mice Sleep Scoring

Efficiently identifying sleep stages is crucial for unraveling the intricacies of sleep in both preclinical and clinical research.
This study introduces LG-Sleep, a novel subject-independent deep neural network architecture designed for mice sleep scoring through electroencephalogram (EEG) signals.
LG-Sleep extracts local and global temporal transitions within EEG signals to categorize sleep data into three stages: wake, rapid eye movement (REM) sleep, and non-rapid eye movement (NREM) sleep.
Experimental findings demonstrate superior performance of LG-Sleep compared to conventional deep neural networks, even with limited training samples.

Read Full Article

3 Likes

Discover more

Arxiv

310

Image Credit: Arxiv

Dimension Reduction with Locally Adjusted Graphs

Dimension reduction algorithms have proven to be useful for analyzing large-scale high-dimensional datasets.
The initial phase of these algorithms involves converting the data into a graph, but this graph is often suboptimal.
LocalMAP is a new dimensionality reduction algorithm that dynamically adjusts the graph to address this challenge.
LocalMAP helps identify and separate real clusters in the data, offering improved accuracy in cluster identification.

Read Full Article

18 Likes

Arxiv

298

Image Credit: Arxiv

Offline Safe Reinforcement Learning Using Trajectory Classification

Offline safe reinforcement learning (RL) is a promising approach for learning safe behaviors without risky online interactions with the environment.
Existing methods in offline safe RL often result in overly conservative policies or safety constraint violations.
This paper proposes a new approach to offline safe RL that learns a policy generating desirable trajectories and avoiding undesirable ones.
The approach involves partitioning a pre-collected dataset into desirable and undesirable subsets, and using a classifier to score the trajectories.

Read Full Article

17 Likes

Arxiv

207

Image Credit: Arxiv

Time Will Tell: Timing Side Channels via Output Token Count in Large Language Models

This paper introduces a new side-channel in large language models (LLMs) that allows an adversary to extract sensitive information about inference inputs.
The side-channel is based on the number of output tokens in the LLM response.
The paper demonstrates attacks utilizing this side-channel in machine translation tasks and text classification tasks.
Proposed mitigations against the output token count side-channel are also discussed.

Read Full Article

12 Likes

Arxiv

211

Image Credit: Arxiv

Non-Uniform Parameter-Wise Model Merging

Combining multiple machine learning models has been a technique for enhancing performance.
Traditional approaches like model ensembles are expensive in terms of memory and compute.
Methods based on averaging model parameters have gained popularity but can yield worse results with differently initialized models.
Non-uniform Parameter-wise Model Merging (NP Merge) is introduced as a novel approach, achieving better results for merging models of various architectures.

Read Full Article

12 Likes

Arxiv

242

Image Credit: Arxiv

Understanding When and Why Graph Attention Mechanisms Work via Node Classification

This paper explores when and why graph attention mechanisms are effective in node classification tasks.
The theoretical analysis reveals that the effectiveness of graph attention mechanisms depends on the relative levels of structure noise and feature noise in graphs.
In situations where structure noise exceeds feature noise, graph attention mechanisms enhance classification performance.
A novel multi-layer Graph Attention Network (GAT) architecture is proposed, which outperforms single-layer GATs in achieving perfect node classification.

Read Full Article

14 Likes

Arxiv

230

Image Credit: Arxiv

A Robust Prototype-Based Network with Interpretable RBF Classifier Foundations

Prototype-based classification learning methods are interpretable but have lower performance compared to deep models.
Deep Prototype-Based Networks (PBNs) aim to combine interpretability with higher performance.
The Classification-by-Components (CBC) approach within PBNs has shortcomings in creating contradicting explanations.
The proposed extension of CBC resolves these issues, improves robustness, and achieves state-of-the-art classification accuracy.

Read Full Article

13 Likes

Arxiv

310

Image Credit: Arxiv

Stylish and Functional: Guided Interpolation Subject to Physical Constraints

Generative AI is revolutionizing engineering design practices by enabling rapid prototyping and manipulation of designs.
This study proposes a zero-shot framework that enforces physical and functional requirements in the generation process of design images.
The case study focuses on generating rotational symmetric wheel designs in the automotive industry.
The proposed approach outperforms existing methods in terms of realism and adherence to physical and functional requirements.

Read Full Article

18 Likes

Arxiv

211

Image Credit: Arxiv

RESQUE: Quantifying Estimator to Task and Distribution Shift for Sustainable Model Reusability

RESQUE is a predictive quantifier that estimates the retraining cost of a model to distributional shifts or change of tasks.
It provides a concise index for estimating the resources required for retraining the model.
Experiments demonstrate that RESQUE has a strong correlation with various retraining measures such as epochs, gradient norms, and parameter changes.
RESQUE enables users to make informed decisions for cost-effective and sustainable retraining options, reducing the model's environmental footprint.

Read Full Article

12 Likes

Arxiv

187

Image Credit: Arxiv

Novelty-Guided Data Reuse for Efficient and Diversified Multi-Agent Reinforcement Learning

Recently, deep Multi-Agent Reinforcement Learning (MARL) has demonstrated its potential to tackle complex cooperative tasks, pushing the boundaries of AI in collaborative environments.
To enhance MARL performance, a novel sample reuse approach called Multi-Agent Novelty-Guided sample Reuse (MANGER) is introduced.
MANGER utilizes a Random Network Distillation (RND) network to measure the novelty of each agent's current state and assigns additional sample update opportunities based on the uniqueness of the data.
Evaluations show significant improvements in MARL effectiveness in scenarios such as Google Research Football and StarCraft II micromanagement tasks.

Read Full Article

11 Likes

Arxiv

179

Image Credit: Arxiv

PreNeT: Leveraging Computational Features to Predict Deep Neural Network Training Time

PreNeT is a predictive framework designed to optimize the training time of deep learning models, particularly Transformer-based architectures.
It integrates comprehensive computational metrics, including layer-specific parameters, arithmetic operations, and memory utilization.
PreNeT accurately predicts training duration on various hardware infrastructures, including novel accelerator architectures.
Experimental results show that PreNeT achieves up to 72% improvement in prediction accuracy compared to contemporary frameworks.

Read Full Article

10 Likes

Arxiv

Image Credit: Arxiv

Generalized Back-Stepping Experience Replay in Sparse-Reward Environments

Back-stepping experience replay (BER) is a reinforcement learning technique that can accelerate learning efficiency in reversible environments.
An enhanced version called Generalized BER (GBER) is proposed, which extends the original algorithm to sparse-reward environments.
GBER improves the performance of BER by introducing relabeling mechanism and applying diverse sampling strategies.
Experimental results show that GBER significantly boosts the performance and stability of the baseline algorithm in various sparse-reward environments.

Read Full Article

4 Likes

For uninterrupted reading, download the app