Wildly successful in narrow domains recently, where will it take us? The notion of the brittleness of solutions is brought up. Brittleness links to need for retesting and maintenance. Which was our own experience.
On Neural Nets By Vinton G. Cerf
I am only a layman in the neural network space so the ideas and opinions in this column are sure to be refined by comments from more knowledgeable readers. The recent successes of multilayer neural networks have made headlines. ....
Communications of the ACM, July 2018, Vol. 61 No. 7, Page 7
10.1145/3224195
A short piece with cautions about the dependency on neural net architectures ...
Showing posts with label Brittleness. Show all posts
Showing posts with label Brittleness. Show all posts
Tuesday, June 26, 2018
Monday, November 13, 2017
Limits of Machine Learning
Good piece. In particular was struck by the term 'brittleness', which we encountered many times in out exploration of enterprise AI. A solution could be useful in a given context, but if the context changed even slightly, and most do change over time, do we detect it, what is the risk, and how to we prepare for any changes? The challenge in modern machine learning is no different. Every project we worked had an element of this.
The Limits of Machine Learning – Is your ML Solution Viable?
By Ahmed Fattah Executive Analytics Architect at IBM Cognitive Solutions
Abstract
Machine Learning (ML) is an extremely powerful technology that is likely to transform business and society. In some cases, ML is the perfect tool for a given task in others it could be an overkill or utterly inappropriate. As with any technology, ML has limits. Lack of understanding of these limits is likely to result in inappropriate use, undesirable outcomes and eventually a repeat disenchantment with whole of AI. The key tools for understanding and taking account of these limits and for developing effective and viable ML solutions are the proven disciplines of Software Engineering and Architecture.
Introduction
When I was writing my Machine Learning (ML) thesis I came across this quote:
“The statement ‘God created man in His own image’ is recursive.”
I was extremely enchanted, inspired and empowered to develop my intelligent ML algorithm. I believe that the above quote encapsulates the fascination we hold of ML and the deep-seated desire to create intelligent machines. At the same time, the quote conjures some fear and uneasiness. These ambiguous perspectives fuel hype, inflate expectations and polarize people into zealots and alarmists. As an AI practitioner, I am, of course, very enthusiastic about the recent rise of AI and ML and confident that these technologies will transform business and society. However, with my experience in Software Engineering and Architecture, I am aware of the danger of inflated expectations and infatuation with new over-hyped technology.
As with many past over-hyped technologies, there is always a danger that arises from misunderstanding their essence and limits. That is why we should always have a healthy dose of doubt when anything is promoted as a silver bullet. Although ML is not the first technology to be over-hyped, ML has unique characteristics that exacerbate this danger. ..... "
The Limits of Machine Learning – Is your ML Solution Viable?
By Ahmed Fattah Executive Analytics Architect at IBM Cognitive Solutions
Abstract
Machine Learning (ML) is an extremely powerful technology that is likely to transform business and society. In some cases, ML is the perfect tool for a given task in others it could be an overkill or utterly inappropriate. As with any technology, ML has limits. Lack of understanding of these limits is likely to result in inappropriate use, undesirable outcomes and eventually a repeat disenchantment with whole of AI. The key tools for understanding and taking account of these limits and for developing effective and viable ML solutions are the proven disciplines of Software Engineering and Architecture.
Introduction
When I was writing my Machine Learning (ML) thesis I came across this quote:
“The statement ‘God created man in His own image’ is recursive.”
I was extremely enchanted, inspired and empowered to develop my intelligent ML algorithm. I believe that the above quote encapsulates the fascination we hold of ML and the deep-seated desire to create intelligent machines. At the same time, the quote conjures some fear and uneasiness. These ambiguous perspectives fuel hype, inflate expectations and polarize people into zealots and alarmists. As an AI practitioner, I am, of course, very enthusiastic about the recent rise of AI and ML and confident that these technologies will transform business and society. However, with my experience in Software Engineering and Architecture, I am aware of the danger of inflated expectations and infatuation with new over-hyped technology.
As with many past over-hyped technologies, there is always a danger that arises from misunderstanding their essence and limits. That is why we should always have a healthy dose of doubt when anything is promoted as a silver bullet. Although ML is not the first technology to be over-hyped, ML has unique characteristics that exacerbate this danger. ..... "
Subscribe to:
Posts (Atom)