Electronic systems are going to have so-called smart components. Smart function advanced appears in systems that are capable of taking sensor data and learning from them to perform tasks that recognize, classify, organize, group and determine actions based on new data similar to those previously captured.
The smart sensors integrate the transducer with a data processor that recognizes the data pattern and provides the upper subsystem with a more structured and useful description of the data (for example podometers or fingerprint readers).
In this subject, we teach data analysis methodologies, learning from them and design of intelligent systems based on the recognition and processing of this data.
All of this will be complemented with practices aimed at experimenting with the methodology and operation of the algorithms proposed in practical systems integrated into an embedded system: Smart Health, Smart detection of car breakdown, smart recycling, haptic systems, gesture recognition manuals etc.
The ultimate goal is for students to be able to investigate the application of the most advanced technologies in Advanced in Electronic Systems. In addition, the ability to analyze new and complex systems is sought.
It is engineering within a broader multidisciplinary context; select and apply the most suitable methods already established analytical, computational and experimental methods, as well as innovative methods and interpreting and criticize the results of these analysis.
- Introduction to intelligence in electronic systems: Incomplete data. Data sensors. Sensors
Intelligent data management and decision making. Machine learning. - Introduction to classification systems: Simple classifiers: Rule 0-R, 1-R, Naive Bayes, Trees of
Decision and Random Forest. - Practice: Use of the WEKA tool. Introduction to the Weka interface explorer. Data loading, editor
data, data filtering. Try some simple algorithms. - Training / evaluation: Cross validation. Statistical significance. Overtraining
- Practice: Smart Health. Recognition of physical activities based on 3D accelerometer. Online Assessment
vs Offline. Benchmark of different classification strategies: hit rate, computational cost. - Selection of characteristics. Study and selection of features to improve the system.
- Practice: Intelligent Motors. Intelligent detection of anomalies in engines and transmission systems
automobiles. Study and selection of features to improve the system. Manual selection of attributes. Selection
automatic attributes. - Advanced classification systems: Logistic, Random Forest, Vector support machines, Neighbor’s rule
closer. - Neural Networks I. Multilayer Perceptron. Deep learning. Convolutional networks.
- Neural Networks II. Deep learning. Transfer learning. Data augmentation.
- Practice: Smart Recycling: Recognition of objects with deep learning. Use of compact models specifically developed for resource-limited HW platforms.
- Unsupervised learning, Clustering, K-Means, Deep learning for clustering.
- Joint learning. General rules of system design.
- Competition Presentations: Accelerometer-based gesture recognition.
