AI-based marketing

AI is great, data is even better

It is often believed that AI models, by default, can produce the desired results. But in reality, it is the quality of data that defines their success

Datasets are essential to AI models. They provide the truth by which we train AI models and measure their success. Engineers generally look to the AI model as the key to delivering highly accurate results, but it is often the data that determines the model’s success. Data flows through every step of the AI workflow, from model training to deployment, and the way they are prepared can be the main driver of accuracy when designing robust AI models.

Engineers can use the following tips to improve their data preparation process and drive success when developing a complete AI system.

Tip 1: Don’t settle for inadequate data

If you don’t have enough data for your AI model, don’t settle for it. There are various techniques you can use to augment and cultivate new data and overcome any shortage of data.

One can generate new data through simulation of a physical model, a common scenario used in predictive maintenance applications. Consider, for example, the case of a hydraulic pump used in oil extraction. You often know what the critical failure causes are, such as a seal leak in the pump. They rarely happen and are destructive, making it very difficult to get actual failure data. With tools that allow you to design and simulate physical systems, you can create a realistic model of the pump and use it to run simulations under various failure scenarios.

If more data doesn’t result in higher accuracy, clean, crop, label and transform the data to provide as many high-quality data samples to the model as possible.

Synthetic data produced can then be used to train AI models, alleviating the issue of lack of data for AI and allowing engineers to continue to focus on building accurate models.

Tip 2: More data doesn’t [necessarily] mean a more successful model

A common reason for frustration when designing AI models is that even with large amounts of data the performance of the model does not increase.

For any problem where more data does not translate to higher accuracy, the solution lies in cleaning, cropping, labeling, and transforming the data to provide as many high-quality data samples to the model as possible. You should look for tools that can help in creating clean data samples and provide automated video labeling and signal labeling capabilities.

Tip 3: Apply your domain expertise to transform your data

Accurate models are never a surprise when they are made with thoughtful, well-prepared data. This is especially important to engineers and scientists using signal data. Raw signal data is rarely added directly to AI models, as signal data tends to be noisy and memory intensive. Instead, time-frequency techniques are incorporated to transform the data to gather the most important features the models will learn.

If you don’t have enough data for your AI model, don’t settle for it. You can generate new data through simulation of a physical model.

Tip 4: Use data to gain insight into your model

Using debugging techniques, engineers can ask the model why a certain category was predicted and what features the model is primarily focused on based on the category. Debugging techniques such as LIME provide insight into the model through data. Therefore, data is equally important to the debugging process – offering insight into both the model and the most important features, with the opportunity to use the debugging information for model improvement.

In summary, it is important to look for the right data and ensure that you clean the data to develop better models. The other critical aspect is the need to apply domain expertise to transform your data and use data debugging techniques to gain insight.





Dr Amod Anandkumar, Application Engineering Manager, MathWorks





Johanna Pingel, Product Marketing Manager, MathWorks

Leave a Reply

Your email address will not be published. Required fields are marked *