A naukri.com initiative
ML News
Arxiv
6d
341
Image Credit: Arxiv
ProtoECGNet: Case-Based Interpretable Deep Learning for Multi-Label ECG Classification with Contrastive Learning
- ProtoECGNet is a prototype-based deep learning model for interpretable, multi-label ECG classification.
- It employs a structured, multi-branch architecture integrating different CNNs with global and time-localized prototypes for rhythm classification, morphology-based reasoning, and diffuse abnormalities.
- The model is trained using a prototype loss designed for multi-label learning, combining clustering, separation, diversity, and contrastive loss.
- ProtoECGNet demonstrates competitive performance compared to state-of-the-art models and provides faithful, case-based explanations, making it a practical solution for transparent and trustworthy deep learning models in clinical decision support.
Read Full Article
20 Likes
Arxiv
6d
50
Image Credit: Arxiv
Surrogate-based optimization of system architectures subject to hidden constraints
- Exploration of novel architectures using physics-based simulation presents challenges for optimization algorithms.
- Surrogate-Based Optimization (SBO) algorithms, specifically Bayesian Optimization (BO) using Gaussian Process (GP) models, address the challenges.
- Strategies are investigated for satisfying hidden constraints in BO algorithms, including rejection of failed points, replacing failed points, and predicting the failure region.
- A mixed-discrete GP is found to achieve the best performance in predicting the Probability of Viability (PoV), demonstrated in solving a jet engine architecture problem.
Read Full Article
2 Likes
Arxiv
6d
12
Image Credit: Arxiv
EquiNO: A Physics-Informed Neural Operator for Multiscale Simulations
- Multiscale problems in physics often require computationally expensive high-resolution simulations.
- Data-driven surrogate models have been used as a faster alternative, but struggle with incorporating microscale physical constraints.
- Equilibrium Neural Operator (EquiNO) is proposed as a complementary physics-informed surrogate that can predict microscale physics.
- The EquiNO framework integrates the finite element method with operator learning and achieves significant speedup compared to traditional methods.
Read Full Article
Like
Arxiv
6d
144
Image Credit: Arxiv
mixEEG: Enhancing EEG Federated Learning for Cross-subject EEG Classification with Tailored mixup
- The cross-subject electroencephalography (EEG) classification presents challenges due to the diversity between subjects.
- To address data sharing limitations, researchers propose mixEEG, a framework for cross-subject EEG classification in federated learning.
- mixEEG tailors the mixup technique to better preserve privacy and offers an averaged label as pseudo-label.
- Experiments show that mixEEG enhances the transferability of the global model for cross-subject EEG classification across different datasets and model architectures.
Read Full Article
8 Likes
Arxiv
6d
242
Image Credit: Arxiv
Comparative analysis of Realistic EMF Exposure Estimation from Low Density Sensor Network by Finite & Infinite Neural Networks
- Understanding the spatial and temporal patterns of environmental exposure to radio-frequency electromagnetic fields is crucial for risk assessments.
- A comparative analysis of finite and infinite-width convolutional network-based methods for estimating and assessing RF-EMF exposure levels was conducted.
- Real-world datasets from 70 sensors in Lille, France, were used for the analysis.
- The evaluation criterion, Root Mean Square Error (RMSE), was used to compare the performance of the deep learning models.
Read Full Article
14 Likes
Arxiv
6d
255
Image Credit: Arxiv
Towards Simple Machine Learning Baselines for GNSS RFI Detection
- Machine learning research in GNSS radio frequency interference (RFI) detection often lacks justification for decisions made in deep learning-based model architectures.
- This paper challenges the status quo in machine learning approaches for GNSS RFI detection and advocates for a shift in focus to simpler and more interpretable machine learning baselines.
- The findings suggest the need for the development of simple and interpretable machine learning methods and demonstrate the effectiveness of a simple baseline for GNSS RFI detection.
- The results show that the simple baseline outperforms complex deep learning architectures with 91% accuracy in detecting potential GNSS RFI.
Read Full Article
15 Likes
Arxiv
6d
267
Image Credit: Arxiv
Fusing Global and Local: Transformer-CNN Synergy for Next-Gen Current Estimation
- This paper presents a hybrid model combining Transformer and CNN to predict current waveforms in signal lines.
- The model does not rely on fixed simplified models and replaces the complex process used in traditional SPICE simulations.
- The hybrid architecture combines the global feature-capturing ability of Transformers with the local feature extraction advantages of CNNs.
- Experimental results demonstrate that the proposed algorithm achieves an error of only 0.0098, improving the accuracy of current waveform predictions.
Read Full Article
16 Likes
Arxiv
6d
91
Image Credit: Arxiv
Neuron-level Balance between Stability and Plasticity in Deep Reinforcement Learning
- A new method called Neuron-level Balance between Stability and Plasticity (NBSP) has been proposed for deep reinforcement learning.
- NBSP focuses on the trade-off between retaining existing skills (stability) and learning new knowledge (plasticity).
- The method identifies RL skill neurons crucial for knowledge retention and introduces a framework to target these neurons for preserving existing skills while enabling adaptation to new tasks.
- Experimental results on Meta-World and Atari benchmarks show that NBSP outperforms existing approaches in balancing stability and plasticity.
Read Full Article
5 Likes
Arxiv
6d
218
Image Credit: Arxiv
Self-Bootstrapping for Versatile Test-Time Adaptation
- This paper introduces a self-bootstrapping scheme for versatile test-time adaptation (TTA) objective in various tasks including image classification, regression, object-level predictions, and pixel-level predictions.
- The scheme optimizes prediction consistency between the test image and its deteriorated view.
- The paper addresses challenges related to preserving geometric information and providing sufficient learning signals for TTA.
- Experiments demonstrate superior results of the proposed method across different tasks.
Read Full Article
13 Likes
Arxiv
6d
345
Image Credit: Arxiv
Vector Quantized-Elites: Unsupervised and Problem-Agnostic Quality-Diversity Optimization
- Vector Quantized-Elites is a novel Quality-Diversity algorithm that constructs a structured behavioral space grid.
- It eliminates the need for prior task-specific knowledge by using unsupervised learning.
- The integration of Vector Quantized Variational Autoencoders enables dynamic learning of behavioral descriptors.
- VQ-Elites demonstrates its ability to efficiently generate diverse, high-quality solutions in robotic arm and mobile robot tasks.
Read Full Article
20 Likes
Arxiv
6d
57
Image Credit: Arxiv
The AI Scientist-v2: Workshop-Level Automated Scientific Discovery via Agentic Tree Search
- The AI Scientist-v2 is an agentic system capable of producing AI-generated peer-review-accepted workshop papers.
- This system formulates hypotheses, conducts experiments, analyzes data, and authors scientific manuscripts autonomously.
- Compared to its predecessor, The AI Scientist-v2 eliminates reliance on human-authored code templates and generalizes effectively.
- One AI-generated manuscript successfully navigated peer review, demonstrating AI's capability in scientific research.
Read Full Article
3 Likes
Arxiv
6d
180
Image Credit: Arxiv
Programs as Singularities
- Researchers have developed a correspondence between the structure of Turing machines and the structure of singularities of real analytic functions.
- The correspondence is based on connecting the Ehrhard-Regnier derivative from linear logic with the role of geometry in Watanabe's singular learning theory.
- By embedding ordinary (discrete) Turing machine codes into a family of noisy codes, a smooth parameter space is formed.
- The structure of the Turing machine and its associated singularity is related to Bayesian inference, implying that the Bayesian posterior can discriminate between different algorithmic implementations.
Read Full Article
10 Likes
Arxiv
6d
350
Image Credit: Arxiv
Multi-view autoencoders for Fake News Detection
- Automatic fake news detection has become a highly important task due to the volume and speed at which fake news spreads across social media.
- This paper proposes using multi-view autoencoders to generate a joint feature representation for fake news detection.
- Experiments on fake news datasets show a significant improvement in classification performance compared to individual views.
- Selecting a subset of views instead of composing a latent space with all views can be advantageous in terms of accuracy and computational effort.
Read Full Article
21 Likes
Arxiv
6d
74
Image Credit: Arxiv
RL-based Control of UAS Subject to Significant Disturbance
- This paper presents a Reinforcement Learning (RL)-based control framework for position and attitude control of Unmanned Aerial Systems (UAS) subjected to significant disturbances.
- The proposed method enables the system to anticipate and counteract disturbances by learning the relationship between the trigger signal and disturbance force.
- Three policies were trained and evaluated: a baseline policy without exposure to disturbances, a reactive policy trained with disturbances but without the trigger signal, and a predictive policy that incorporates the trigger signal as an observation and is exposed to disturbances during training.
- Simulation results demonstrate that the predictive policy outperforms the other policies by proactively correcting position deviations and improving UAS performance.
Read Full Article
4 Likes
Arxiv
6d
78
Image Credit: Arxiv
Benchmarking Suite for Synthetic Aperture Radar Imagery Anomaly Detection (SARIAD) Algorithms
- Anomaly detection is a key research challenge in computer vision and machine learning with applications in many fields from quality control to radar imaging.
- In radar imaging, specifically synthetic aperture radar (SAR), anomaly detection can be used for the classification, detection, and segmentation of objects of interest.
- To address the lack of method for developing and benchmarking SAR imagery anomaly detection methods, the Synthetic Aperture Radar Imagery Anomaly Detection (SARIAD) suite is introduced.
- SARIAD integrates multiple SAR datasets, various anomaly detection algorithms, and provides metric evaluation and visualization tools for benchmarking SAR models and datasets.
Read Full Article
4 Likes
For uninterrupted reading, download the app