The black box inside the black box: The problem with deep learning.

Mario Verdicchio (Speaker)

Activity: Talk or presentationInvited talk

Description

Deep learning is the latest hot topic in Information Technology: thanks to a significant progress in chip manufacturing techniques, proposals based on neural networks that were only futuristic theory in the 1980s have become the reality of many cutting-edge initiatives around the world, bringing Artificial Intelligence back to the main stage of research, after many "winters" of broken promises and misplaced hopes.
However, despite the enthusiasm of the industry, the press, and many scholars, deep learning comes with the same problem that characterises all neural-network-based approaches: human programmers can train a computational system to perform a task even better than any person, but they do not know how the system works in detail, and they are not able to provide an explicit description of the functions at play when the system acts. This is problematic in several ways, and this lecture will try and unpack the relevant issues.
Period23 Apr 2018
Held atPolitecnico di Milano, Italy
Degree of RecognitionInternational

Keywords

  • Artificial Intelligence
  • Deep Learning
  • Neural Nets
  • Technology
  • Philosophy
  • Ethics