Dunith Dhanushka
Senior Developer Advocate
Redpanda

SESSION OVERVIEW: When I was a child, I had an uncle who was blind. Every day after returning from school, I would describe the sunset to him.

What if, with the help of Generative AI and computer vision technologies, I could make this a reality for many people like him today? In this talk, I’ll share a hobby project I’ve developed to narrate the world in real-time.

The “Be My Eyes” project leverages AI to extend the scope of experience for blind or visually impaired people, utilizing OpenAI’s advanced object detection and computer vision models.

The magic unfolds through a simple yet powerful set-up: a video camera that continuously records the user’s surroundings. These real-time recordings are then fed into an AI model trained to analyze and interpret the visual content of the video.
In a remarkable feat of technological integration, “Be My Eyes” converts the AI model’s textual narration of the scene into an audio description via a text-to-speech model.

Anna Lackey
Cyber and Networks Engineer
Roke

Ben Rawson
Graduate Data Scientist
Roke

SESSION OVERVIEW: Artificial Intelligence and Machine Learning solutions can often be considered as silver bullet solutions by business owners. But as the technology matures, so does the sophistication of potential cyber-attacks against them.

Here we explore how the introduction of Machine Learning models can create new attack vectors that companies may not be aware of- leaving them vulnerable to adversaries. We investigate different kinds of machine learning models, and explore some of the attacks they are vulnerable to.

In this talk we will discuss:

– Techniques causing incorrect classification of computer vision models.
– Methods to manipulate sentiment analysis or translation outputs.
– Poising a model to learn incorrect or hidden patterns.
– How adversaries could recover potentially private data from models.
– Methods for defending against these attacks. From robust training techniques to enhance model resilience, to regularization methods, anomaly detection and encryption

Aldan Creo
Technology Research Specialist
Accenture Labs

SESSION OVERVIEW: Knowledge Graphs (KGs) are data structures that can capture complex relationships between concepts in a very flexible way. They excel at harmonizing huge amounts of heterogeneous data, and we can train AI models to generate embeddings of its nodes and vertices. These embeddings can be used to infer knowledge from the graph. For example, if we have the Facebook graph, we can train a model and ask “What’s the probability that I’m friends with this person?”.

Are you ready to dive into the world of AI and Knowledge Graphs? We’ll learn the technical side of these algorithms, their real-world applications, and how to overcome their limitations. We’ll also explore current lines of research to combine them with LLMs. Join us to learn about the future of AI and how you can use it to your advantage!

Sarah Pengelly
Data Scientist
Naimuri

SESSION OVERVIEW: How important is it for your classification model to tell you when its predictions are wrong?

Calibration is a measure of how well the confidence scores produced by a model match its accuracy (a well calibrated model gives high confidence scores to correct predictions and low confidence scores to incorrect predictions, a badly calibrated model does the opposite). This talk gives a detailed look into this often forgotten metric and investigates when you should (and shouldn’t) care about it.

We will cover a range of different methods that can be used to improve model calibration, their pros and cons, and how to implement them. For anyone interested in ways to measure model performance that go beyond accuracy, this talk will offer insight into all the different ways confidence scores can matter and what you can do to make them better.

Rich House
Tech Lead
Counter

SESSION OVERVIEW:

Jess Sherratt
Head of User Experience (UX)
Code

Colin Preston
Product and Service Design Director
Em Code.

SESSION OVERVIEW: tbc

SESSION OVERVIEW:Discover the transformative power of AI at the forefront of innovation! Join us at the AI Track Day during MTF, where Naimuri presents an exclusive panel of top-tier AI experts.

Uncover the thrilling possibilities and potential pitfalls of artificial intelligence. Our esteemed panel will delve into the critical discussions on how AI can revolutionise our lives for the better, ensuring a healthier and safer future, while also addressing the significant risks and challenges that come with it.

Don’t miss this unique opportunity to gain valuable insights and stay ahead in the ever-evolving world of AI