0

I want to hire a freelancer to develop a machine learning algorithm. In order for the algorithm to provide any value I need it to work correct in over 95% of cases, which is a very difficult requirement to be met.

From my side, I cannot pay for something that works worse than that as it will have no value for me. And I fear people will not accept these hard requirements because any time and effort invested on their end might not produce the required results.

Is there a way I can mitigate this such as offering a very low rate for trying but not achieving the results? How should I handle this? Is there any best practice/strategy on requesting difficult RnD(research and development) requirements while minimizing my risk?

2
  • Please keep in mind it could be impossible to achieve such accuracy, not difficult or costly, but plain impossible. It could also be impossible to judge without a preliminary study and even then you could lose money. Your best bet (not from a freelancing point of view) would be to have a look at the literature or market and see if you find any similar application to your specific problem and gauge the results.
    – Cris
    Mar 7, 2023 at 10:01
  • What is the success rate of a human on this task ? Does it even reach 95% ?
    – user4521
    Mar 10, 2023 at 13:15

2 Answers 2

2

From my side, I cannot pay for something that works worse than that as it will have no value for me

Understandable. Viewed from the other end besides a student or an enthusiast you most likely won‘t find anybody …

A step back, ML is a filter, which depends on the history of its data offered. More generalized there‘s a lot of noise, so your are in the business of designing robust products. This can be done systematically, including identifying and improving limited concepts.

Is there a way I can mitigate this such as offering a very low rate for trying but not achieving the results?

Consider your employer offers you just that … and you know the answer.

Is there any best practice/strategy on requesting difficult RnD(research and development) requirements while minimizing my risk?

As I lined out above, evaluating and improving robustness of your system offers this. It won‘t be a one step process. Here are some things that will and have to happen on this path:

  • You have a goal in mind and a couple of ideas or concept to achieve it
  • That‘s a good starting point
  • These concepts will demonstrate shortcomings (either by expensive market introduction with redesigns, or quick with robust …)
  • Which means: your beloved idea will/has to die to make room for the more capable one
  • Which will require at least one to-the-point invention
  • Repeat
  • While your system will perform so good, you can‘t believe it … if you dare taking this route.

To indicate roughly 2 out of 3 products have a hidden performance, which you simply can use afterwards. Roughly 1 out of 3 products „fail“ optimization attempts … and those are your gold mine. Why? Because their concept imposes limitations. How to overcome it on-the-fly? By said invention(s). Literally in hours. So you want to torture your ML to fail quickly, to invent, improve and learn fast … beyond books, conferences, working groups.

To sum it up, if you don‘t invest into a guide on this route, you won‘t get returns. If you do, you are on a turbo-learning-link, with a small team. Consider investing in this kind of know-how and capability, not into exchangeable capacity. Then review your risk concerns.

Best

1

IMO, the simplest is to state the things as they are: you are looking for someone willing to accept the risk. He will tell you his/her rate for accepting the risk.

You can also suggest a smaller compensation in case of a failure. This is just a matter of finding someone foolish enough to accept. But if things are said upfront, nothing goes wrong.


Keep in mind that the risk is there and there is no reason that someone else should carry it for you.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.