A naukri.com initiative
ML News
Arxiv
2h
82
Image Credit: Arxiv
MANGO: Multimodal Acuity traNsformer for intelliGent ICU Outcomes
- MANGO: Multimodal Acuity traNsformer for intelliGent ICU Outcomes is a model designed to improve the prediction of patient acuity in the Intensive Care Unit (ICU).
- The model utilizes a multimodal dataset, incorporating electronic health records (EHR) data, wearable sensor data, video of patient's facial cues, and ambient sensor data.
- MANGO employs a multimodal feature fusion network powered by Transformer masked self-attention method to capture complex interactions across these diverse data modalities.
- The model achieved promising results, with an area under the receiver operating characteristic curve (AUROC) of 0.76-0.82 for predicting acuity status, transitions, and the need for life-sustaining therapy.
Read Full Article
4 Likes
Arxiv
2h
78
Image Credit: Arxiv
Transfer Learning with Active Sampling for Rapid Training and Calibration in BCI-P300 Across Health States and Multi-centre Data
- Machine learning and deep learning advancements in Brain-Computer Interface (BCI) have limitations due to individual health, hardware variations, and cultural differences affecting neural data.
- Transfer learning with active sampling (AS) using a convolutional neural network enhances BCI performance in diverse settings.
- The proposed AS method improves classification accuracy by 5.36% and reduces standard deviation by 12.22%.
- This approach shows better generalizability, computational time, and training efficiency compared to traditional methods.
Read Full Article
4 Likes
Arxiv
2h
75
Image Credit: Arxiv
EEG-GMACN: Interpretable EEG Graph Mutual Attention Convolutional Network
- Electroencephalogram (EEG) is a valuable technique for analyzing brain activity.
- Existing Graph Signal Processing (GSP) studies lack interpretability and prediction confidence.
- EEG-GMACN is introduced to enhance interpretability and credibility of EEG classification.
- The study improves transparency and effectiveness of EEG analysis for clinical and neuroscience research.
Read Full Article
4 Likes
Arxiv
2h
72
Image Credit: Arxiv
SCFNet:A Transferable IIIC EEG Classification Network
- A neural network architecture called SCFNet is proposed for EEG signal classification.
- The feature extractor of the model is an RCNN network with single-channel input.
- EEG-SCFNet improves accuracy by 4% compared to the baseline model.
- EEG-SCFNet maintains performance even with different channel leads.
Read Full Article
4 Likes
Arxiv
2h
308
Image Credit: Arxiv
Cross-Species and Cross-Modality Epileptic Seizure Detection via Multi-Space Alignment
- Researchers propose a multi-space alignment approach for cross-species and cross-modality epileptic seizure detection.
- The approach employs deep learning techniques, including domain adaptation and knowledge distillation.
- Experiments on human and canine EEG datasets show significant improvements in detection accuracy.
- The study highlights the effectiveness of integrating data from different species and modalities for EEG-based seizure detection.
Read Full Article
18 Likes
Arxiv
2h
292
Image Credit: Arxiv
Bridging the Data Provenance Gap Across Text, Speech and Video
- The authors conducted a longitudinal audit of popular text, speech, and video datasets.
- They analyzed nearly 4000 public datasets between 1990-2024, spanning 608 languages, 798 sources, 659 organizations, and 67 countries.
- They found that web-crawled, synthetic, and social media platforms are the primary sources for multimodal machine learning applications.
- They also discovered that a significant portion of widely-used datasets carry non-commercial restrictions and coverage of languages and geographies has not significantly improved in recent years.
Read Full Article
17 Likes
Arxiv
2h
206
Image Credit: Arxiv
Compact Neural Network Algorithm for Electrocardiogram Classification
- A high-performance, compact electrocardiogram (ECG)-based system for automatic classification of arrhythmias is presented in this paper.
- The system integrates machine learning approaches and feature enhancement techniques to capture morphological and time-frequency features from ECG signals.
- It includes 17 newly engineered features to extract significant data and physiological patterns from the ECG signal.
- The system achieves an accuracy of 97.36% on the MIT-BIH arrhythmia database and shows potential for clinical deployment and portable devices in cardiac health monitoring applications.
Read Full Article
12 Likes
Arxiv
2h
196
Image Credit: Arxiv
Foxtsage vs. Adam: Revolution or Evolution in Optimization?
- Foxtsage is a hybrid optimization approach that integrates Hybrid FOX-TSA with Stochastic Gradient Descent for training Multi-Layer Perceptron models.
- Foxtsage achieves a 42.03% reduction in loss mean and a 42.19% improvement in loss standard deviation compared to the widely adopted Adam optimizer.
- There are modest improvements in accuracy, precision, recall, and F1-score with Foxtsage.
- However, Foxtsage has a higher computational cost with a 330.87% increase in time mean compared to Adam.
Read Full Article
11 Likes
Arxiv
2h
183
Image Credit: Arxiv
The Unreasonable Effectiveness of Open Science in AI: A Replication Study
- A systematic replication study was conducted on 30 highly cited AI studies to assess reproducibility.
- Out of the 30 articles, 8 were rejected due to inaccessible data or hardware.
- 6 articles were fully reproduced, while 5 were partially reproduced.
- Reproducibility is strongly correlated with sharing code and data, and the quality of data documentation.
Read Full Article
11 Likes
Arxiv
2h
111
Image Credit: Arxiv
EnhancePPG: Improving PPG-based Heart Rate Estimation with Self-Supervision and Augmentation
- EnhancePPG is a method that improves heart rate estimation from photoplethysmography (PPG) signals.
- It integrates self-supervised learning with data augmentation to enhance the performance of deep learning models.
- By utilizing unsupervised PPG signal reconstruction and data augmentation, the approach improves the best HR estimation by 12.2%.
- The method focuses on training the deep learning model without significantly increasing its inference latency.
Read Full Article
6 Likes
Arxiv
2h
170
Image Credit: Arxiv
A Multimodal Emotion Recognition System: Integrating Facial Expressions, Body Movement, Speech, and Spoken Language
- A multimodal emotion recognition system has been developed to provide standardized and objective emotional assessment.
- The system integrates facial expressions, body movement, speech, and spoken language analysis.
- It aims to capture subtle emotional cues often overlooked in human evaluations and reduce the risk of mis- and overdiagnosis.
- Preliminary testing demonstrates its potential in improving diagnostic accuracy and its application in clinical and therapeutic settings.
Read Full Article
10 Likes
Arxiv
2h
318
Image Credit: Arxiv
Causal Composition Diffusion Model for Closed-loop Traffic Generation
- Simulation is crucial for safety evaluation in autonomous driving, but generating realistic and controllable traffic scenarios is challenging.
- A new framework called Causal Compositional Diffusion Model (CCDiff) is introduced to address the challenges.
- CCDiff maximizes controllability while adhering to realism by injecting causal structures into the diffusion process.
- CCDiff outperforms state-of-the-art approaches in generating realistic and user-preferred trajectories.
Read Full Article
19 Likes
Arxiv
2h
318
Image Credit: Arxiv
ArchComplete: Autoregressive 3D Architectural Design Generation with Hierarchical Diffusion-Based Upsampling
- ArchComplete is a two-stage dense voxel-based 3D generative pipeline for architectural design.
- Stage 1 uses a 3D Voxel VQGAN model with an autoregressive transformer to generate coarse models.
- Stage 2 employs Hierarchical Voxel Upsampling Networks to enhance the coarse shapes with fine geometric details.
- ArchComplete supports various interaction modes and shows notable improvements over state-of-the-art methods.
Read Full Article
19 Likes
Arxiv
2h
219
Image Credit: Arxiv
tuGEMM: Area-Power-Efficient Temporal Unary GEMM Architecture for Low-Precision Edge AI
- A novel GEMM architecture called tuGEMM is proposed for low-precision edge AI.
- tuGEMM is based on temporal-coding and performs exact computation.
- Two variants of tuGEMM, serial and parallel, are introduced with distinct area/power-latency trade-offs.
- The designs show significant advantages in area-power efficiency compared to state-of-the-art stochastic unary systems, especially at low precisions.
Read Full Article
13 Likes
Arxiv
2h
137
Image Credit: Arxiv
CARL-GT: Evaluating Causal Reasoning Capabilities of Large Language Models
- Causal reasoning capabilities of large language models (LLMs) are evaluated using a benchmark named CARL-GT.
- CARL-GT assesses LLMs in areas such as causal graph reasoning, knowledge discovery, and decision-making.
- LLMs are found to be weak in causal reasoning, particularly with tabular data to uncover new insights.
- Different benchmark tasks show varying strengths of LLMs, with performance correlation within categories.
Read Full Article
8 Likes
For uninterrupted reading, download the app