DailyHealthcareAI
@aipulserx.bsky.social
Sharing research papers and news on AI applications in radiology, pathology, genetics, protein design and many more. Let's learn together!
Meta is releasing these datasets to encourage broader research community involvement and advance their vision of expanding input methods for computing devices.
ai.meta.com/blog/open-so...
ai.meta.com/blog/open-so...
Advancing Neuromotor Interfaces by Open Sourcing Surface Electromyography (sEMG) Datasets for Pose Estimation and Surface Typing
We’re releasing emg2qwerty and emg2pose—two large datasets and benchmarks for sEMG-based typing and pose estimation, as part of the NeurIPS 2024 Datasets and Benchmarks track.
ai.meta.com
December 7, 2024 at 1:07 AM
Meta is releasing these datasets to encourage broader research community involvement and advance their vision of expanding input methods for computing devices.
ai.meta.com/blog/open-so...
ai.meta.com/blog/open-so...
These are the largest open-source sEMG datasets to date, each being 10 times larger than previous comparable datasets. The technology could enable new ways of interacting with devices, particularly in augmented reality, allowing for text input and hand tracking without physical keyboards.
December 7, 2024 at 1:07 AM
These are the largest open-source sEMG datasets to date, each being 10 times larger than previous comparable datasets. The technology could enable new ways of interacting with devices, particularly in augmented reality, allowing for text input and hand tracking without physical keyboards.
The emg2qwerty dataset focuses on typing without a physical keyboard, containing 346 hours of recordings from 108 participants, while emg2pose focuses on hand pose estimation with 370 hours of data from 193 participants.
December 7, 2024 at 1:07 AM
The emg2qwerty dataset focuses on typing without a physical keyboard, containing 346 hours of recordings from 108 participants, while emg2pose focuses on hand pose estimation with 370 hours of data from 193 participants.
The robot maintained consistent performance across 36 PIN, 26 POU, and 30.3 SNC interactions with approximately 60% success rates across all behaviors.
Link: nature.com/articles/s42...
Link: nature.com/articles/s42...
December 7, 2024 at 1:03 AM
The robot maintained consistent performance across 36 PIN, 26 POU, and 30.3 SNC interactions with approximately 60% success rates across all behaviors.
Link: nature.com/articles/s42...
Link: nature.com/articles/s42...
The system demonstrated successful behavior learning by achieving relative dynamic range index (RDRI) scores higher than the mean dataset trajectories.
December 7, 2024 at 1:03 AM
The system demonstrated successful behavior learning by achieving relative dynamic range index (RDRI) scores higher than the mean dataset trajectories.
• The learned behaviors achieved high dynamic similarity scores across all interaction patterns (>0.92 DSY score). Quantitative evaluation showed the robot maintained average joint tracking errors under 0.1 radians.
December 7, 2024 at 1:03 AM
• The learned behaviors achieved high dynamic similarity scores across all interaction patterns (>0.92 DSY score). Quantitative evaluation showed the robot maintained average joint tracking errors under 0.1 radians.
The ML system maps rat behavior data from Cartesian space to the robot's joint space using multilayer perceptrons. The policy network uses mean squared error loss to generate learned behaviors, with the ability to adjust behavior distribution through initial data selection.
December 7, 2024 at 1:03 AM
The ML system maps rat behavior data from Cartesian space to the robot's joint space using multilayer perceptrons. The policy network uses mean squared error loss to generate learned behaviors, with the ability to adjust behavior distribution through initial data selection.
• The system uses a two-stage machine learning approach: a pretraining prediction block and policy optimization. They collected and labeled 88,218 frames of rat motion capture data to train the model.
December 7, 2024 at 1:03 AM
• The system uses a two-stage machine learning approach: a pretraining prediction block and policy optimization. They collected and labeled 88,218 frames of rat motion capture data to train the model.
The robot needed to learn multiple interaction patterns including pinning (PIN), pouncing (POU), and social nose contact (SNC) to effectively modulate rat emotional states.
December 7, 2024 at 1:03 AM
The robot needed to learn multiple interaction patterns including pinning (PIN), pouncing (POU), and social nose contact (SNC) to effectively modulate rat emotional states.
• Previous robot-rat interaction systems have relied on pre-programmed behaviors or simple motion patterns. This study introduces a novel approach using machine learning to capture and replicate the subtle nuances of rat social interactions.
December 7, 2024 at 1:03 AM
• Previous robot-rat interaction systems have relied on pre-programmed behaviors or simple motion patterns. This study introduces a novel approach using machine learning to capture and replicate the subtle nuances of rat social interactions.
The system aims to bridge the gap between traditional robot-animal interactions and the need for natural social engagement by creating a biohybrid system capable of learning and reproducing complex rat behaviors.
December 7, 2024 at 1:03 AM
The system aims to bridge the gap between traditional robot-animal interactions and the need for natural social engagement by creating a biohybrid system capable of learning and reproducing complex rat behaviors.
• Researchers developed SMuRo, an autonomous rat-like robot that uses imitation learning to reproduce rat social behaviors.
December 7, 2024 at 1:03 AM
• Researchers developed SMuRo, an autonomous rat-like robot that uses imitation learning to reproduce rat social behaviors.
and long-term monitoring over 5 separate nights and a continuous 21-night period, demonstrating median real-time inter-beat interval errors of 26.1ms and 34.1ms in outpatient and daily scenarios respectively, representing a tenfold improvement over existing systems.
Link: nature.com/articles/s41...
Link: nature.com/articles/s41...
December 7, 2024 at 12:55 AM
and long-term monitoring over 5 separate nights and a continuous 21-night period, demonstrating median real-time inter-beat interval errors of 26.1ms and 34.1ms in outpatient and daily scenarios respectively, representing a tenfold improvement over existing systems.
Link: nature.com/articles/s41...
Link: nature.com/articles/s41...
and calculates key metrics like RT-IBI, RMSSD, SDRR, and pNN50 to evaluate heart rate variability with clinical precision.
• The system was validated through extensive testing with 6,222 eligible participants in an outpatient setting achieving 83.4% accuracy in cardiac abnormality detection,
• The system was validated through extensive testing with 6,222 eligible participants in an outpatient setting achieving 83.4% accuracy in cardiac abnormality detection,
December 7, 2024 at 12:55 AM
and calculates key metrics like RT-IBI, RMSSD, SDRR, and pNN50 to evaluate heart rate variability with clinical precision.
• The system was validated through extensive testing with 6,222 eligible participants in an outpatient setting achieving 83.4% accuracy in cardiac abnormality detection,
• The system was validated through extensive testing with 6,222 eligible participants in an outpatient setting achieving 83.4% accuracy in cardiac abnormality detection,
and proving its effectiveness in monitoring cardiac abnormalities.
• The method uses variational mode decomposition algorithm for signal processing, implements a beat frequency pattern extraction technique,
• The method uses variational mode decomposition algorithm for signal processing, implements a beat frequency pattern extraction technique,
December 7, 2024 at 12:55 AM
and proving its effectiveness in monitoring cardiac abnormalities.
• The method uses variational mode decomposition algorithm for signal processing, implements a beat frequency pattern extraction technique,
• The method uses variational mode decomposition algorithm for signal processing, implements a beat frequency pattern extraction technique,
• The researchers developed a novel 60-64 GHz radio frequency sensing system that can monitor heart rate variability (HRV) without skin contact, overcoming respiratory interference using previously undiscovered frequency ranges beyond 10-order heartbeat harmonics,
December 7, 2024 at 12:55 AM
• The researchers developed a novel 60-64 GHz radio frequency sensing system that can monitor heart rate variability (HRV) without skin contact, overcoming respiratory interference using previously undiscovered frequency ranges beyond 10-order heartbeat harmonics,
• Cardiovascular diseases cause 17.9 million deaths annually costing $555 billion, with over 80% of premature cases being preventable through early detection, but current monitoring methods like ECG and wearables have limitations in comfort and accuracy.
December 7, 2024 at 12:55 AM
• Cardiovascular diseases cause 17.9 million deaths annually costing $555 billion, with over 80% of premature cases being preventable through early detection, but current monitoring methods like ECG and wearables have limitations in comfort and accuracy.
December 7, 2024 at 12:42 AM
super-resolution enhancement (0.8mm³ from 3.0mm³), scanner harmonization across Siemens/GE/Philips machines, and downstream tasks like segmentation and diagnosis, with significant improvements in quantitative metrics like tissue contrast t-score (p<0.001).
December 7, 2024 at 12:42 AM
super-resolution enhancement (0.8mm³ from 3.0mm³), scanner harmonization across Siemens/GE/Philips machines, and downstream tasks like segmentation and diagnosis, with significant improvements in quantitative metrics like tissue contrast t-score (p<0.001).
• The model demonstrated superior performance over state-of-the-art methods across 19 public datasets spanning fetal to elderly subjects, effectively handling motion correction,
December 7, 2024 at 12:42 AM
• The model demonstrated superior performance over state-of-the-art methods across 19 public datasets spanning fetal to elderly subjects, effectively handling motion correction,
network to generate high-quality images, using 516 training subjects (52 fetal, 464 aged 0-6 years) and testing on 19 datasets with 13,411 total images (10,963 in vivo, 2,448 synthetic).
December 7, 2024 at 12:42 AM
network to generate high-quality images, using 516 training subjects (52 fetal, 464 aged 0-6 years) and testing on 19 datasets with 13,411 total images (10,963 in vivo, 2,448 synthetic).
highlighting the critical need for automated enhancement solutions that can handle diverse imaging conditions.
• The methodology involves training a tissue-classification neural network to predict tissue labels which guide a "tissue-aware" enhancement
• The methodology involves training a tissue-classification neural network to predict tissue labels which guide a "tissue-aware" enhancement
December 7, 2024 at 12:42 AM
highlighting the critical need for automated enhancement solutions that can handle diverse imaging conditions.
• The methodology involves training a tissue-classification neural network to predict tissue labels which guide a "tissue-aware" enhancement
• The methodology involves training a tissue-classification neural network to predict tissue labels which guide a "tissue-aware" enhancement