Machine Learning Porto #1 - One-Shot Learning & Active Learning

Join us in the first meetup of our community!
We will have the pleasure of having Tiago Freitas and Pedro Costa who will be teaching us about One-Shot Learning using Siamese Networks and how to use Active Learning to help labeling your dataset.

------------------------------------------
AGENDA
------------------------------------------
18:30 - ✅ Check-in
18:45 - 🎙 Introduction to One-shot learning using siamese networks with Tiago Freitas
19:30 - ☕️ Coffee break
19:45 - 🎙 Using a model to help you label your dataset with Pedro Costa

------------------------------------------
TALKS & SPEAKERS
------------------------------------------

Title:
Introduction to One-shot learning using siamese networks

Abstract:
In this talk it will be introduced the problem of One-Shot learning in computer vision, with a focus on how can we use Siamese Neural Networks to solve this type of problem. The talk will be focused on the theory and implementation details of Siamese Neural Networks for One-shot Image Recognition by Koch et al.

About the speaker:
Tiago Freitas has a MSc in Biomedical Engineering at FEUP (2016). He did some research in biomedical imaging and multimodal face recognition at INESCTEC (2015 - 2016) and after ending his Masters he worked as a Freelancer for Computer Vision problems. He currently works as a Computer Vision / Machine Learning Engineer at Adapttech where he is helping to develop a smart solution for lower-limb prosthetic fitting.

------------------------------------------

Title:
Using a model to help you label your dataset

Abstract:
Machine Learning models’ performance feeds on data. Large public datasets have been arguably the biggest cause of recent success with Deep Learning. Nowadays, the first step of a new Computer Vision project is the collection of image data and further annotation. Labeling a new image dataset is highly time consuming, and part of that time is spent annotating uninformative and redundant images. What if we could select only the most relevant and informative examples from the entire unlabeled dataset? What if we could minimize the number of required labeled examples to achieve a desired performance? In this presentation we explore how we are using Active Learning to create a dataset to solve the problem of detecting fish in subsea videos.

About the speaker:
Pedro Costa holds a MSc (2015) in Informatics and Computing Engineering at the Faculty of Engineering, University of Porto. Costa started working with INESC TEC in 2014 trying to find adverse drug reactions in biological data using ML. After a brief experience in the industry, Costa came back to INESC TEC to work on medical image processing using Deep Learning methods, having published in top medical imaging journals and conferences. He then spent 3 months working on weakly supervised deep learning methods at Carnegie Mellon University (CMU). Currently, he is Head of Research at Abyssal, working on how to improve Remotely Operated Vehicles’ operational efficiency using ML and Computer Vision techniques. He is also a medical imaging researcher at INESC TEC and keeps cooperating with CMU.