Towards fair ML models – Riccardo Monetti
Fairness is a key pillar of AI trustworthiness. How fairness should be handled, assessed and monitored while developing and deploying a machine learning application? The need to establish a workflow based approach to manage these points is rising rapidly and could soon become a regulatory requirement. This presentation’s objective is to propose a possible workflow solution highlighting pros and cons.
Riccardo Monetti is a Data Scientist with a mathematics background. He has a M.Sc. In Statistics and an internship experience in USA. He joined UniCredit in 2018 and worked on e2e ML and data applications across different group functions.
Topics: bias, E2E workflow, fair machine learning
Cross-Attentional Neural Network to Compare Image and Text – Dario Saccavino
Starting from a simple use case concerning the automatic processing of bank cheques, we study a novel multimodal learning problem: given an image containing a single line of text and a candidate text transcription, the goal is to assess whether the text represented in the image corresponds to the candidate text. The proposed model, via end-to-end training, compares the two inputs by applying a cross-attention mechanism over the embedding representations of image and text. Results show a good performance achieved on a variety of configurations.
Dario Saccavino, born in Milan in 1982, is a mathematician who loves coding. He joined UniCredit in 2019 to work as data scientist. Recently he developed further interest into Quantum Computing topics.
Topics: cross attention, joint embedding learning, multimodal learning, text recognition, text matching