FORESIGHT, Issue 55

Description

Fall 2019

Special Feature

Autonomous or “Lights Out” Supply-Chain Planning: What New Technology Is Required by Niels van Hove
Analogous to the emergence of autonomous vehicles is the momentum toward autonomous supply-chain planning, given names such as “lights out” planning. Niels van Hove describes this development as the third wave of supply-chain planning, following the functionality in ERPs (first wave) and advanced planning software (second wave). But he argues that before the lights can be turned out—i.e., before human input can be eliminated— many technological hurdles must be overcome.

• Commentary: Close the Loop, Stabilize, and Respond, by Stefan de Kok

Articles

1. Forecasting at Scale: The Architecture of a Modern Retail Forecasting System by Phillip Yelland, Zeynep Erkin Baz, and David Serafini
   In this first of a three-part article, the authors in the Data Science/AI team at Target, describe their team’s efforts to construct a demand forecasting system capable of efficiently generating the nearly one billion weekly forecasts required by the Target Corporation. They highlight the interplay of challenges arising in the contexts of statistical modeling, software engineering, and business practice and explain how the team surmounted obstacles in these three fields of knowledge. Subsequent parts of the article will address the process of implementing the forecasting system and its maintenance in production.
2. Open-Source Forecasting Tools in Python by Tim Januschowski, Jan Gasthaus, and Yuyang Wang
   The authors from Amazon Web Services have been heavily engaged in deep learning modeling and forecasting from neural networks. Their two-part tutorial and analysis of neural forecasting appeared in the Summer and Fall 2018 issues of this journal. Here they review the state of open-source software options for forecasting, concentrating on alternative platforms to the R package.
3. Continual Learning: The Next Generation of Artificial Intelligence by Daniel Philps
   An introduction to automated machine learning and its possible next-generation realization, continual learning (CL). CL advances the state of the art by attempting to automatically learn different tasks while retaining knowledge from previous model implementations. This article presents an application of CL to investment decisions. It also offers the interesting perspective that complexity is not simply a technical characteristic of a model formulation, but also a resultant of the application of human judgment.
4. Book Review by Michael Gilliland
   Forecasting: An Essential Introduction, by Jennifer Castle, Michael Clements and David Hendry