Explainability and reliability

Explainability and reliability

Visual Intelligence is developing deep learning methods which provide explainable and reliable predictions, opening the “black box” of deep learning.

Motivation

A limitation of deep learning models is that there is no generally accepted solution for how to open the “black box” of the deep network to provide explainable decisions which can be relied on to be trustworthy. Therefore, there is e a need for explainability, which means that the models should be able to summarize the reasons for their predictions, both to gain the trust of users and to produce insights about the causes of their decisions.

Solving research challenges through new deep learning methodology

Visual Intelligence researchers have proposed new methods that are designed to provide explainable and transparent predictions. These results include methods for:

• content-based CT image retrieval, imbued with a novel representation learning explainability network.

• explainable marine image analysis, providing clearer insights into the decision-making of models designed for marine species detection and classification.

• tackling distribution shifts and adverserial attacks in various federated learning settings involved in images.

• discovering features to spot counterfeit images.

Developing explainable and reliable models is a step towards achieving deep learning models that are transparent, trustworthy, and accountable. Our proposed methods are therefore critical for bridging the gap between technical performance and real-world usage in an ethical and responsible manner.

‍