Future IoT TechnologiesLaajuus (5 cr)

Objective

Course objectives
In the course, you will learn how to use advanced new technologies, such as artificial intelligence, in the implementation of IoT systems. You get to know different neural networks and understand their basic parameters.

Competences:
EUR-ACE Knowledge and Understanding
EUR-ACE Engineering Practice


You know the architectures of neural networks and understand their application areas. You know how to make a Q-table of basic neural network design and implementation in practice. You know what the RL agent does and you know how to train the agent to make it work optimally. You understand the difference between On-Policy and Off-Policy algorithms and know how to use information to select an algorithm.

Content

Comparison of neural networks (RL, Supervised, Unsupervised )
Keras, Pytorch and other algorithm-based neural networks
RL agent and environment
Exploitation and exploration
Reward and hyperparameters
Basic neuromath and activation functions
Policy gradient method and value based learning
Bellman equation and its demo (Q-learning)
Pytorch, PPO algorithm and demo

Qualifications

Johdanto IoT-järjestelmiin

Assessment criteria, satisfactory (1)

Avoidance 1: You understand the principles of neural networks and can apply the knowledge to select them

Satisfactory 2: You understand the basics of Q-learning methods and can make a small application with it

Assessment criteria, good (3)

Good 3: Can adjust the parameter of the Bellman algorithm (Exploration, Exploitation, gamma +....) so that the operation of the controlled object is optimized

Excellent 4: You know how to tune a multilayer neural network using the PPO algorithm so that the given actuator behaves as desired.

Assessment criteria, excellent (5)

Excellent 5: In addition to the previous ones, you understand the limitations of artificial intelligence in various practical applications and understand that OpenAI-Gym models require a strong substance control. You also know how to document your work well, using TensorFlow outputs to evaluate hyperparameters.