20 Jan 2025
Cristobel
Soares
On Monday, 20 January 2025 the University of Portsmouth hosted the GRADnet Workshop:
Application of Neural Networks to Image Recognition for 25 SEPnet postgraduate and postdoctoral researchers.
Thank you to Dr Michal Gnacik, EPAM Systems, Data Science and Machine Learning Specialist, who kindly delivered this workshop on behalf of SEPnet.
Delegates were introduced to the most common types of neural networks widely used in Image Recognition and details of how neural networks learn; feeding forward; backpropagation and gradient descent type methods.
The focus of the workshop was the implementation (in Python, PyTorch) of these neural networks and applying them to classify images; building some simple networks with a few convolutional layers and then exploiting popular deep neural networks (e.g., ResNet, Inception https://pytorch.org/vision/stable/models.html).
Using a few Kaggle image dataset, in particular, students were able to recognise emotions from the images of people's faces and to discuss the generative models for image generation, in particular, Latent Diffusion model Kadinsky available from Hugging Face.
Michael used PyTorch mainly but showed delegates how this can be done in Keras with TensorFlow backend.
Delegates said:
“It was a great introduction to the field of neural networks and I enjoyed the beginning maths part as well as the hands on stuff. I don't directly use Neural Networks in my research but if I end up doing so I would use his Collab books as a template to build my own stuff from.”
“That was a great event with a good mix of advanced and basic materials about neural networks.”
“Please convey my heartiest gratitude to Michal. The notebooks he prepared are excellent and accessible. The material was practical, and he also suggested a handful of online services to make our lives easier. It was great to attend this workshop”.
For more information on EPAM Systems see below:
• YouTube
