My Ph.D project focuses on the application of Machine Learning and Deep Learning techniques in the field of human activity recognition. In particular, my research focuses on the analysis and on the improvement of classification techniques based on inertial signals recorded by wearable devices.
- Improving traditional machine learning techniques exploiting personaliza- tion methods on metadata: differences within signals could prevent a one to one signal-class association. Intrinsic data grouping could be explored based on un- supervised statistical techniques, such as PCA, Clustering and DLA, or simply on taking into account metadata may highlight data similarities and differences. Based on that, I improved traditional machine learning techniques, such as SVM, AdaBoost, k-NN, by exploiting similarity between users signals data and between users physical attributes.
- Comparison between traditional machine learning and deep learning techniques: differences between machine learning and deep learning techniques are mostly due to the feature extraction procedure. I investigate the performance ef- fects based on different sets of features selection. That is, Convolutional Neural Networ and Residual Neural Network were implemented and compared with SVM and k-NN.
- Homogeneization and integration of public datasets: a plenty of labeled databases for machine learning and deep learning benchmarking are not consistent, both syn- tactically (e.g., different sampling frequency) and semantically (e.g., labels with different meanings). Coherent merging of existing databases would enable the evaluation of generalization capabilities of methods across databases and increase the number of labeled data. I propose a semi-automatic procedure to coherently merge existing databases based on signal and word similarity. Furthermore, I con- struct a platform for integrating and distributing heterogeneous data and models.
- A. Ferrari, D. Micucci, M. Mobilio, P. Napoletano,“Hand-craftedFeaturesvsResidual Networks for Human Activities Recognition using Accelerometer”, IEEE 23° Interna- tional Symposium on Consumer Technologies, 2019.
- A. Ferrari, D. Micucci, M. Mobilio, P. Napoletano, “A Framework for Long-Term Data Collection to Support Automatic Human Activity Recognition”, International Work- shop on the Reliability of Intelligent Environments, 2019.
- A. Ferrari, D. Micucci, M. Mobilio, P. Napoletano, “On the Homogenization of Het- erogeneous Inertial-based Databases for Human Activity Recognition”, IEEE Services Conference, 2019.
- A. Ferrari, D. Micucci, M. Mobilio, P. Napoletano, “A Platform to Collect, Unify, and Distribute Inertial Labeled Signals for Human Activity Recognition”, Italian Confer- ence on ICT for Smart Cities And Communities, 2019.
- A. Ferrari, D. Micucci, M. Mobilio, P. Napoletano, “Human Activities Recognition using Accelerometer and Gyroscope”, European Conference on Ambient Intelligence 2019.
- A. Ferrari, D. Micucci, M. Mobilio, P. Napoletano, “On the Personalization of Classi- fication Models for Human Activity Recognition”, IEEE Access PP(99):1-1, February 2020.
G. Fossati, G. Costigliolo, M. Sironi, A. Colagrossi, S. Erba, A. Ferrari, A. Garlaschi. “Correlazione tra risonanza magnetica ed istopatologia nel planning preoperatorio del carcinoma mammario” accepted as poster at Convegno Nazionale GISMa, Bologna, 12-13 september 2012.
Cultura del Dato – Introduzione a R (2019). Tutoraggio di supporto a studenti e professori di scuole secondarie per l’apprendimento del software R.
anna dot ferrari at disco dot unimib dot it
+39 02 6448 7851
Room 1039 – Sal
Viale Sarca 336