A naukri.com initiative
ML News
Arxiv
4d
20
Image Credit: Arxiv
seeBias: A Comprehensive Tool for Assessing and Visualizing AI Fairness
- A new tool called seeBias has been developed for evaluating AI fairness and predictive performance.
- seeBias offers comprehensive evaluation across classification, calibration, and other performance domains.
- The tool includes customizable visualizations to support transparent reporting and responsible AI implementation.
- Available as an R package on GitHub with a Python version currently under development.
Read Full Article
1 Like
Arxiv
4d
267
Image Credit: Arxiv
Customizing Spider Silk: Generative Models with Mechanical Property Conditioning for Protein Engineering
- A novel computational framework has been developed for designing spider silk protein sequences with customizable mechanical properties.
- The framework utilizes a GPT-based generative model, trained on curated subsets of the Spider Silkome dataset, to generate biologically plausible spider silk repeat regions.
- The model is able to tailor the mechanical properties of the generated sequences to specific requirements.
- Validation studies have shown the accuracy of the model in predicting mechanical properties and confirming its potential for engineering spider silk-inspired biomaterials.
Read Full Article
16 Likes
Arxiv
4d
385
Image Credit: Arxiv
A Systematic Evaluation of Knowledge Graph Embeddings for Gene-Disease Association Prediction
- Discovery gene-disease links is important in biology and medicine areas, enabling disease identification and drug repurposing.
- Machine learning approaches accelerate this process by leveraging biological knowledge represented in ontologies and the structure of knowledge graphs.
- A novel framework is introduced for comparing the performance of link prediction versus node-pair classification tasks in gene-disease association prediction.
- Results show that enriching the semantic representation of diseases slightly improves performance, while additional links generate a greater impact. Link prediction methods outperform overall.
Read Full Article
23 Likes
Arxiv
4d
82
Image Credit: Arxiv
LGRPool: Hierarchical Graph Pooling Via Local-Global Regularisation
- Hierarchical Graph Pooling (HGP) methods aim to address the limitations of conventional Graph Neural Networks (GNN) in terms of being inherently flat and lacking multiscale analysis.
- However, most HGP methods fail to consider the global topology of the graph, focus primarily on feature learning, and align local and global features in a multiscale manner.
- The proposed LGRPool method utilizes the expectation maximization framework in machine learning to align local and global aspects of message passing.
- LGRPool introduces a regularizer to ensure that the global topological information is in line with the local message passing at different scales in HGP.
Read Full Article
4 Likes
Arxiv
4d
106
Image Credit: Arxiv
Explainability and Continual Learning meet Federated Learning at the Network Edge
- Leveraging collective compute power of edge devices for distributed learning is gaining interest in wireless networks.
- Critical challenges exist in optimizing learning at the network edge, including the trade-off between predictive accuracy and interpretability.
- Integrating inherently explainable models like decision trees in distributed learning is difficult due to their non-differentiable structure.
- Combining continual learning strategies with federated learning supports adaptive, lifelong learning in resource-limited environments.
Read Full Article
6 Likes
Arxiv
4d
223
Image Credit: Arxiv
Slicing the Gaussian Mixture Wasserstein Distance
- Gaussian mixture models (GMMs) are widely used in machine learning for various tasks.
- A key challenge in working with GMMs is defining a computationally efficient and geometrically meaningful metric.
- The mixture Wasserstein (MW) distance has been applied in various domains, but its high computational cost limits scalability to high-dimensional and large-scale problems.
- To address this, the researchers propose slicing-based approximations to the MW distance that reduce computational complexity while preserving optimal transport properties.
Read Full Article
13 Likes
Arxiv
4d
345
Image Credit: Arxiv
Uncovering the Structure of Explanation Quality with Spectral Analysis
- Effective explanation methods are crucial for ensuring transparency in high-stakes machine learning models.
- A new framework based on spectral analysis of explanation outcomes has been proposed.
- The framework uncovers two factors of explanation quality: stability and target sensitivity.
- Experiments on MNIST and ImageNet demonstrate the trade-offs between these factors in popular evaluation techniques.
Read Full Article
20 Likes
Arxiv
4d
349
Image Credit: Arxiv
Boosting-inspired online learning with transfer for railway maintenance
- The integration of advanced sensor technologies with deep learning algorithms has revolutionized fault diagnosis in railway systems, particularly at the wheel-track interface.
- This paper introduces BOLT-RM (Boosting-inspired Online Learning with Transfer for Railway Maintenance), a model designed to address the challenges in railway maintenance using continual learning for predictive maintenance.
- BOLT-RM overcomes the issue of catastrophic forgetting that often plagues traditional models by allowing the model to continuously learn and adapt as new data become available, retaining past knowledge while improving predictive accuracy.
- The proposed BOLT-RM model demonstrates significant improvements in identifying wheel anomalies, establishing a reliable sequence for maintenance interventions.
Read Full Article
21 Likes
Arxiv
4d
207
Image Credit: Arxiv
MooseAgent: A LLM Based Multi-agent Framework for Automating Moose Simulation
- MooseAgent is an automated solution framework that uses large-scale pre-trained language models (LLMs) and a multi-agent system to automate the simulation process in the MOOSE framework.
- It utilizes LLMs to understand user-described simulation requirements in natural language and generates MOOSE input files through task decomposition and iterative verification strategies.
- MooseAgent improves accuracy and reduces model hallucinations by utilizing a vector database containing annotated MOOSE input cards and function documentation.
- The framework shows promising results in automating the MOOSE simulation process, particularly for relatively simple single-physics problems.
Read Full Article
12 Likes
Arxiv
4d
227
Image Credit: Arxiv
Task-conditioned Ensemble of Expert Models for Continuous Learning
- One of the major challenges in machine learning is maintaining the accuracy of the deployed model in a non-stationary environment.
- The proposed method is a task-conditioned ensemble of expert models for continuous learning of the deployed model with new data.
- The method uses in-domain models and task membership information to dynamically adapt the deployed model to new data while retaining accuracy on old data.
- The experiments conducted on different setups demonstrate the benefits of the proposed method.
Read Full Article
13 Likes
Arxiv
4d
339
Image Credit: Arxiv
Deep Learning Methods for Detecting Thermal Runaway Events in Battery Production Lines
- One of the key safety considerations in battery manufacturing is thermal runaway, which can lead to fires, explosions, and toxic gas emissions.
- Researchers investigated the use of deep learning for detecting thermal runaway in VDL Nedcar's battery production line.
- The study collected data from the production line representing baseline and thermal runaway conditions, using optical and thermal images.
- Deep learning models, including shallow convolutional neural networks, residual neural networks, and vision transformers, were evaluated and found to be a viable approach for thermal runaway detection.
Read Full Article
20 Likes
Arxiv
4d
178
Image Credit: Arxiv
Application of machine learning models to predict the relationship between air pollution, ecosystem degradation, and health disparities and lung cancer in Vietnam
- Lung cancer is a major cause of death in Vietnam, with high disease rates and public health burden.
- The study focuses on understanding the sources of lung cancer in Vietnam, considering environmental features, health state, and the country's socioeconomic and ecological context.
- Large datasets, including patient health records and environmental indicators, are utilized to determine causal correlations and identify cancer risk patterns.
- Machine learning models, such as Decision Tree, Random Forest, and Support Vector Machine, show promising results in identifying disease patterns with high accuracy.
Read Full Article
10 Likes
Arxiv
4d
300
Image Credit: Arxiv
Channel Estimation by Infinite Width Convolutional Networks
- Researchers propose a novel approach to channel estimation in wireless communications using infinite width convolutional networks.
- The traditional channel estimation problem in OFDM systems relies on sparse pilot data, which poses an ill-posed inverse problem.
- The proposed approach uses a convolutional neural tangent kernel (CNTK) derived from an infinitely wide convolutional network, which can accurately estimate the channels with limited training data.
- Numerical results show that the proposed strategy outperforms deep learning methods in terms of speed, accuracy, and computational resources for channel estimation.
Read Full Article
18 Likes
Arxiv
4d
25
Image Credit: Arxiv
Regularized infill criteria for multi-objective Bayesian optimization with application to aircraft design
- Bayesian optimization is an advanced tool for efficient global optimization.
- The proposed method extends the super efficient global optimization with mixture of experts (SEGOMOE) to solve constrained multiobjective problems.
- Regularization techniques are used to cope with the ill-posedness of the multiobjective infill criteria.
- Preliminary results show a significant reduction in total cost compared to the evolutionary algorithm NSGA-II.
Read Full Article
1 Like
Arxiv
4d
332
Image Credit: Arxiv
Beyond Black-Box Predictions: Identifying Marginal Feature Effects in Tabular Transformer Networks
- Deep neural networks have been powerful in predictive tasks across various fields, including tabular data problems.
- The transformer architecture has challenged gradient-based decision trees in handling tabular data.
- However, the black-box nature of deep tabular transformer networks makes it difficult to interpret marginal feature effects.
- A proposed adaptation of tabular transformer networks aims to identify and maintain intelligible marginal feature effects while maintaining predictive power.
Read Full Article
20 Likes
For uninterrupted reading, download the app