This session is facilitated by Albert Njoroge Kahira
About this session
[5mins] Introduction
[10 mins] I will have a questionnaire that users can take using either from the iPad that will provided or by simply scanning a QR code. The questionnaire will be mostly pictorial presenting the different scenarios of how machine learning is involved in our day to day lives. The objective to gauge the familiarity of participants with uses of ML in their lives.
[30 mins]
I will then take participants through a poster that will be on display. First, explaining the basics ML through games (i.e. guessing game), then I will explain how ML models are attacked/fooled by simple data tricks. I will make a demo of this using say 2 participants or more. Finally, I will demonstrate how scientists/engineers to ensure ML is safe and secure. Even though the topic is very research oriented, the demonstrations will make it easy to understand and relate.
Goals of this session
In the coming digital era, thousands of devices will communicate with each other across the Internet and around us, taking decisions without their owner being even aware of it. Those decisions are taken based in the status of the system and the input data, both of which are subject to errors that can happen either by natural factors or induced by attackers.
The goal of my session is to involve the users in an exploration of different errors and attacks that occur in machine learning systems. It is important for participants to understand this because we constantly engaged in a digital world with almost absolute trust in the systems. However, the presence of such errors/attacks should not be reason to panic, as such, I will engage the users in discussion different ways that scientists are developing to prevent such attacks.