The Application of Expressive Machine Learning

In this final episode we conclude our short series on Differentiable Programming by looking at some of its potential use cases and discuss the far reaching consequences it can have throughout a supply chain.

We learn how Differentiable Programming can be used to combine challenges that previously were broached in complete isolation and how the approach varies between different verticals. We investigate the problems that can be improved upon by applying Differentiable Programming techniques and explore the various positive impacts that this exciting new technology can have.

Differentiable Programming can also be used to divide and conquer silos, a key element for many supply chains. By moving away from the classic time-series forecast, Differentiable Programming can take in to account a client’s point of view when they enter a retail point of sale, something that allows much more intelligent and relevant purchasing decisions to be made. We talk about some of the other benefits of this approach and learn how it can be applied throughout a supply chain, from the point of sale to warehousing, through multi-echelon supply chain challenges.

Finally, we compare this approach to previous iterations of Lokad’s technology and learn about the critical insights that Differentiable Programming can deliver through increased expressiveness. But nobody can deny that Differentiable Programming is anything but straightforward. At the end of the day, are we using a sledgehammer to crack a nut when a much simpler tool would do? We explain why we don’t think that’s the case.

Timestamps

00:08 Introduction

00:30 What are some of the problems which can be improved by applying Differentiable Programming techniques?

02:55 What is the game changing property which makes this possible?

06:03 How about we look at things from a warehousing level. Is it mostly by forecasting demand more accurately that Differentiable Drogramming can help us ?

08:10 Was this something we could do previously with the existing techniques?

10:39 How about Differentiable Programming for manufacturing constraints and for those multi-echelon Supply Chain challenges?

14:16 Is this where the idea of modelling outcomes that are not completely deterministic comes in?

16:14 What you are saying is that Differentiable Programming can change depending on the different verticals. Is that right?

19:27 The problem that many people will have it that Differentiable Programming is very complex. Are we not using a sledgehammer to crack a nut and using it to solve problems which could actually be approached with a simple solution?

22:06 What are the critical insights that Differentiable Programming can provide that previously were not possible?