Research

Deep learning for time-series classification, applied to human activity recognition from wearable sensors.

My doctoral research developed TSCResNet, a one-dimensional residual convolutional network for time-series classification, and applied it to recognizing human activities from wearable-sensor data. I studied it first in the closed-set setting, where every activity is seen during training, and then in the open-set setting, where the model must also recognize activities it has never seen. Code and details are on the project page.

Publications

Performance Evaluation of Open-Set Classification in Human Activity Recognition via a Residual Neural Network Architecture
D. Rodriguez Gonzalez, Y. Qu · 2023 Congress in Computer Science, Computer Engineering, & Applied Computing (CSCE), IEEE, pp. 470–476, 2023

Applies TSCResNet to open-set activity recognition, where some activities are absent from training, a situation common in real deployments but little studied. Evaluated on the PAMAP2 benchmark with the same preprocessing and protocol as prior open-set work.

DOI

Improving the Performance of Closed-Set Classification in Human Activity Recognition by Applying a Residual Neural Network Architecture
D. Rodriguez Gonzalez, Y. Qu · European Journal of Electrical Engineering and Computer Science 6(2), 59–65, 2022 · Open access (CC BY 4.0)

Introduces TSCResNet and evaluates it on the PAMAP2 benchmark using the standard held-out-subject protocol, reaching 95.6% accuracy and 95.5% mean F1, above the prior results it was compared against (best: 93.7% accuracy).

DOI PDF

Improving the Performance of Open-Set Classification in Human Activity Recognition by Applying a Residual Neural Network Architecture
D. Rodriguez Gonzalez · Doctor of Computer Science dissertation, Colorado Technical University, 2022

ProQuest

Patents

Co-inventor, U.S. provisional patent application (2025).

Cite

BibTeX

@inproceedings{rodriguezgonzalez2023openset,
  author    = {Rodriguez Gonzalez, Daniel and Qu, Yanzhen},
  title     = {Performance Evaluation of Open-Set Classification in Human Activity
               Recognition via a Residual Neural Network Architecture},
  booktitle = {2023 Congress in Computer Science, Computer Engineering,
               \& Applied Computing (CSCE)},
  pages     = {470--476},
  year      = {2023},
  publisher = {IEEE},
  doi       = {10.1109/CSCE60160.2023.00084}
}

@article{rodriguezgonzalez2022closed,
  author  = {Rodriguez Gonzalez, Daniel and Qu, Yanzhen},
  title   = {Improving the Performance of Closed-Set Classification in Human
             Activity Recognition by Applying a Residual Neural Network Architecture},
  journal = {European Journal of Electrical Engineering and Computer Science},
  volume  = {6},
  number  = {2},
  pages   = {59--65},
  year    = {2022},
  doi     = {10.24018/ejece.2022.6.2.433}
}

@phdthesis{rodriguezgonzalez2022dissertation,
  author = {Rodriguez Gonzalez, Daniel},
  title  = {Improving the Performance of Open-Set Classification in Human Activity
            Recognition by Applying a Residual Neural Network Architecture},
  school = {Colorado Technical University},
  type   = {Doctor of Computer Science dissertation},
  year   = {2022},
  note   = {ProQuest Publication No. 28967546}
}