We’re excited to bring Transform 2022 back to life on July 19th and virtually July 20-28. Join AI and data leaders for sensible conversations and exciting networking opportunities. Register today!
Machine learning (ML) is an invaluable asset to modern businesses across the board. However, when it comes to the ML model, both B2C and B2B companies face the problem of delayed time in the market. According to Algorithmia, most companies take at least a month or more to develop and then deploy their ML model.
This is because of a complex and often very expensive two-stage process. Developing an ML model can be a long and potentially expensive process. But what many companies often don’t realize at the beginning is that after the initial phase, another, arguably more challenging phase – should be followed by deployment. This second stage involves moving the finished model into production, testing it and fine-tuning it and then scaling accordingly.
It is estimated that only about 10% of all businesses have sufficient experience, financial resources and technical expertise to deploy a new ML model for production within one week of completion. Many struggle for up to a year, with at least 30% of companies taking at least three months after post-deployment. How long it takes depends on which three popular model types the company chooses.
Off-the-shelf, custom and custom adaptive models
Of the ML models currently available in the market, the following are: Generic model, Custom model and Custom adaptive model.
Normal and custom models are basically anti-polar. The difference is that the general model is low in both price and accuracy, while the custom model is high in both price and accuracy. This is because the generic model is designed to suit virtually every business in that industry. These are usually based on ResNet, BERT / GPT and similar off-the-shelf technologies. As a result, these models are affordable and reliable, but they are also far from perfect fit.
In contrast, custom models are always tailored to the task at hand and are therefore more accurate. However, they also come with a much higher cost due to their high development and maintenance costs. Those who start with a simple solution and then try to modify their ML model often venture beyond the basic architecture of the model. The custom model is the one they end up with. A custom adaptive model that can be quickly adapted to a wide range of business needs and leaves most post-deployment fine-tuning.
The adaptive model is a type of custom model that has some advantages that the general model offers. Like all other custom models, adaptive models are designed with specific business needs in mind. For this reason, they are very accurate. At the same time, they do not require the company to detect MLops after the initial development phase. As a result, they work in some ways similar to the general model in the deployment and post-deployment phases, with relatively low maintenance costs and improved timing for the market.
Choosing the ML model
Which model your business needs – that is, whether the extra pay is worth the stretch – depends on your specific situation. Your business may need something as simple as placing online orders in different warehouses depending on their location. In this case, the typical ML model can only do the trick, especially if you are a small business.
On the other hand, if it’s something specific like content moderation for the online community of doctors discussing medical instruments, the custom model would work better. The generic ML model may be seen as inappropriate language – for example, referring to the genitals – not only appropriate but also necessary in the context of a medical discussion. In this case the training model needs to be tailored to the specific needs of the company. And the model made by this tailor can be either adaptive or not.
Let’s consider the advantages and disadvantages of each model:
Custom adaptive model
Custom ML models are often expensive due to unforeseen pre- and post-deployment costs. Because of these generally high startup costs, some companies tend to shy away from tailor-made options, rather than opting for less accurate but less expensive generic tracks. How expensive a training model really is depends on a number of factors, including the chosen data-labeling method, which is reflected in the flexibility of the model or lack thereof.
The following case demonstrates the crowdsourcing-based custom adaptive model in action, that is, the adaptive model that relies on human-in-the-loop labeling:
A well-known company providing technical acquisition environments wanted to increase the accuracy of its software and reduce the model’s training costs. The engineering team had to come up with a more efficient solution to improve the sentences in English. Any solution should be fully compatible with the manual labeling pipeline that already exists.
The final solution involves using a pre-existing custom model for a linguistic process tailored to the client’s needs. Third-party AutoML was used for text classification in target sentences. Since then, phrase accuracy has increased from 6% – from 76% to 82%. This, in turn, reduced the model’s training costs by 3%. In addition, the client did not need to make additional investments in the model’s infrastructure – financial or otherwise – as is usually the case with most custom models.
Key points to keep in mind
Choosing the right ML model for your business can be a daunting task. Here’s a summary of what you need to consider when making an informed decision:
- Consider how specific your needs are: the more specific the need, the more you should move away from the general model.
- Always think of scalability – if it’s something you know you’ll need, just think of paying extra for something made for you.
- If you do not need high precision but need fast deployment, consider choosing the usual route.
- If accuracy is important to you, consider how much time you can spend in the market.
- If you have less time and need higher precision, consider taking a custom adaptive route; Otherwise, any custom solution could potentially meet your needs as well.
- In terms of overall cost, the generic route is the cheapest – followed by the custom adaptive route that bypasses most MLops costs – and finally by all the other custom solutions whose prices can increase significantly after deployment (exact figures are very different). Case-by-case support).
- Consider that you have in-house data scientists and MLEs at your disposal – if yes, it is possible to go for an internally developed traditional custom option; If not – consider the other two (normal or custom adaptive).
- When choosing between custom vs. Custom adaptive options, consider how accurate and precise the ML model should ultimately be for your customer’s needs. The higher the accuracy and adaptability, the higher the cost and the longer the waiting period for the model to be designed and maintained.
Fedor Zhdanov is head of ML products at Toloka AI,
Welcome to the VentureBeat community!
DataDecisionMakers is a place where experts, including tech people working on data, can share data-related insights and innovations.
If you would like to read about the latest ideas and latest information, best practices and the future of data and data tech, join us at DataDecisionMakers.
You might even consider contributing to your own article!
Read more from DataDecisionMakers