A Deep Learning Approach to on-Node Sensor Data Analytics for Mobile or Wearable Devices
read more
Citations
Deep learning for healthcare: review, opportunities and challenges.
Deep learning for sensor-based activity recognition: A survey
Deep Learning for IoT Big Data and Streaming Analytics: A Survey
Deep learning algorithms for human activity recognition using mobile and wearable sensor networks: State of the art and research challenges
Network Intrusion Detection for IoT Security Based on Learning Techniques
References
ImageNet Classification with Deep Convolutional Neural Networks
Gradient-based learning applied to document recognition
ImageNet classification with deep convolutional neural networks
Object recognition from local scale-invariant features
The Design and Implementation of FFTW3
Related Papers (5)
Deep Convolutional and LSTM Recurrent Neural Networks for Multimodal Wearable Activity Recognition
A Survey on Human Activity Recognition using Wearable Sensors
Frequently Asked Questions (15)
Q2. What are the methods used to classify the data?
Methods such as decision trees and support vector machines (SVM) are then trained to classify the data using the given features [21]–[23].
Q3. What are the main challenges of using statistical parameters to describe time-series data?
In other applications, statistical parameters [8], basis transform coding [9], and symbolic representation [10] are often used as “shallow” features to describe time-series data.
Q4. How many filters are needed to perform the performance?
For simple datasets, a filter size of two seems to provide the best performance, while for more complex datasets this size needs to be increased up to three.
Q5. What datasets are the candidates for the combined approach?
Combining both deep learnt and shallow features allows the best of the approaches to be exploited, producing a more generalizable solution that, as shown in their experiments, overcomes the other approaches for all datasets except for the Skoda dataset.
Q6. What is the reason for the lack of a deep learning approach?
Another possibility is that, since the extraction of features through deep learning is driven by data, if the dataset is not well represented in all the possible modalities(i.e., location of the sensor, different sensor’s properties such as amplitude or sampling rate) the deep learning approach is not capable to generalize these data modalities automatically for the classification task.
Q7. How is the proposed method suitable for real-time on-node HAR?
the authors show that the computation time obtained from low-power devices, such as smartphones, wearable devices, and IoT, is suitable for real-time on-node HAR.
Q8. What is the main challenge when designing a classification method for time-series analysis?
One of the main challenges, when designing a classification method for time-series analysis, is selecting a suitable set of features for subsequent classification.
Q9. What is the main advantage of the filters?
These filters are applied repeatedly to the entire spectrogram and the main advantage is that the network contains just a number of neurons equal to a single instance of the filters, which drastically reduces the connections from the typical neural network architecture.
Q10. What is the plausible reason for the lower precision and recall results observed?
For the Daphnet FoG dataset, the under representation of the class“freeze” in the training data is the most plausible reason for the lower precision and recall results observed.
Q11. How many filters can be used to implement the deep learning model?
For both architectures, the authors use theFFTW3 library [33] to extract the spectrogram and the Torch framework [34] to implement the deep learning model.
Q12. What is the largest dataset in terms of number of samples?
It is one of the largest datasets in terms of number of samples with around 30 h of labeled raw data, and it is the first database that groups together data captured using different sensor configurations.
Q13. What are the disadvantages of deep learning?
these disadvantages include the following:1) deep learning modules can contain redundant links between pairs of nodes that connect two consecutive layers of the neural network; 2) correlations in different signal points are usually overlooked; and 3) a large set of layers can be built on top of each other to extract a hierarchy of features from low level to high level.
Q14. What is the sum of the frequency distribution of the activity?
Since each activity has a discriminative distribution of frequencies, as shown in Fig. 2, the sum is performed in correspondence to each frequency.
Q15. How long did the computation take to perform the classification task?
To evaluate if the proposed method could achieve real-time performance on a smartphone or a miniature wearable device, the computation time required to perform the classification task for a 10 s segment of data was measured.