Machine learning can deliver major benefits for organizations looking to scale their operations, especially those who continue to experience disruptions from the COVID-19 pandemic. However, there are plenty of challenges that come along with implementing this type of artificial intelligence. Before organizations jump headfirst, there are four major machine learning questions they should consider.
1) What is the specific problem you want your machine learning model to solve?
According to Alyssa Simpson Rochwerger and Wilson Pang, authors of the book “Real World AI: A Practical Guide for Responsible Machine Learning,” just 20% of artificial intelligence system pilots deployed by major companies advance to production. Furthermore, those that do often fail to efficiently serve their customers. This is generally either because “they’re trying to solve the wrong problem,” or because “they fail to account for all the variables (or latent biases) that are crucial to a model’s success or failure.”
Some problems require more advanced machine learning models than others. To avoid wasting time and money, you need to identify the specific problem you want your model to solve before moving into development.
Let’s say you need a machine learning model that can label files in a vast image archive. Such a task would require a deep learning model. However, let’s say you instead need a system that can spot weeds in a large field using a camera-equipped drone. For this, you would need a much more advanced system, such as a specialized neural network.
Besides identifying your system’s purpose, you also need to determine the necessary level of accuracy. For instance, if you are developing a machine learning model to spot cancer, the system needs to be highly accurate or people may lose their lives.
2) How can you make your training dataset as accurate as possible?
Because the data will dictate accuracy, the dataset you use to train your model is even more important than the model. Depending on what you are trying to achieve, you may want to train your model on publicly available datasets, such as ImageNet, an online image database with hundreds of thousands of images, or data you collect from your own sources.
Consider the image archive scenario. For this, a public dataset may work just fine. However, if you are building a weed detection system, where the model needs to be able to identify weeds under different lighting and environmental conditions in a specific agricultural setting, a large scale dataset is unlikely to work. This scenario indicates that it would be best to create your own dataset with images specifically labeled “weed” vs “non-weed.” This would require you to take photos of weeds and non-weeds under different lighting and environmental conditions, then label the images.
In other scenarios, depending solely on a public dataset or your own data may be equally insufficient. If this is the case, consider a mixed approach known as “transfer learning”. In this approach, you develop semi-trained models based on large-scale datasets, but which you then train on your own data to make them more accurate.
3) How can you maintain and update your model to ensure it’s adaptable?
Simply designing and implementing a machine learning model based on your current needs won’t serve your long term goals. Your system needs to be able to adapt to rapidly shifting environments. For example, the COVID-19 pandemic created a massive shift in online shopping habits. As a result, many online retailers had to quickly retrain their machine learning models.
Whether it’s determining a mechanism for labeling new data or allowing users to provide feedback on the accuracy of your model, you need to designate resources for the continuous training of your model. Otherwise, it will “become less accurate over time as the real world changes” around it, according to Rochwerger and Pang.
4) Who are the experts who will oversee the development of your machine learning model?
Developing an effective machine learning model means going beyond the technical aspects of what you want to build. Hiring a diverse team of product managers, ethics experts, and user-feedback researchers is just as important as selecting the right engineers to build the model. This will help ensure the model will be technically sound while also meeting the human needs of your end users.
Developing effective machine learning models can help scale your operations, but it also requires quite a bit of strategizing. However, carefully considering these questions before moving forward can prevent mistakes and save you resources in the long run.
Understanding Machine Learning
By providing AI with the ability to learn from its experiences without needing explicit programming, machine learning is important to developing the technology. Covering machine learning models, algorithms, and platforms, Machine Learning: Predictive Analysis for Business Decisions, is a five-course program from IEEE.
Connect with an IEEE Content Specialist today to learn more about this program and how to get access to it for your organization.
Interested in the program for yourself? Visit the IEEE Learning Network.
Enhancing Business Operations with Machine Learning
Machine learning is one of the most exciting aspects of artificial intelligence (AI). As the key for the future of AI technology, machine learning allows AI to adapt based on its experiences. When applied in a business setting, this type of data-driven business operations and decision-making becomes more accurate and accessible.
Join IEEE on 19 May 2021 at 12pm ET to hear from subject matter expert Grant Scott and learn about the business applications of machine learning, specifically related to decision making and automation. Also learn how collecting, analyzing, and interpreting data, using machine learning, can benefit your business.
Resources
Conzelmann, Anke. (22 April 2021). Prioritizing Artificial Intelligence and Machine Learning in a Pandemic. IoT for All.
Dickson, Ben. (19 April 2021). The challenges of applied machine learning. TechTalks.
[…] Many organizations today are focused on implementing ML, but few have adopted appropriate best practices and policies to make the technology work for them operationally. […]