Parameter estimation in stock assessment modelling: caveats with gradient-based algorithms. (18th April 2018)

• Record Type: Journal Article
• Authors: Subbey, Sam
• Is Part Of: ICES journal of marine science. Volume 75:Number 5(2018)
• Page Start: 1553
• Page End: 1559
• Publication Date: 2018-04-18
• Subjects: automatic differentiation (AD) -- global optimum -- Hessian -- likelihood -- local minima -- objective function -- optimization -- parameters -- positive definite -- uncertainty; Ocean -- Periodicals; Fisheries -- Periodicals; Fishes -- Periodicals; Marine biology -- Bibliography -- Periodicals; 551.4605
• DOI: 10.1093/icesjms/fsy044
• Languages: English
• ISSNs: 1054-3139

Abstract

Abstract: Using simple illustrative examples, this note highlights some of the caveats with gradient-based algorithms. This class of algorithms underpins the state-of-the-art modelling platform in fisheries science. The goal is to sound a cautionary note about an increasing trend in fisheries science, where blind faith is being invested in results obtained from algorithms that are fast, and proven to have machine precision .