How Often Do Data Scientists Utilize Deep Learning in Their Work?

As a data scientist, the adoption of deep learning (DL) in solving technical problems is far from universal, but it certainly plays a significant role in many projects. In this article, we'll delve into the frequency and effectiveness of using deep learning in data science, based on the experiences of various professionals in this field.

Why Deep Learning Sometimes Isn’t the First Choice

Many data scientists and professionals working in the field emphasize that deep learning is often seen as a last resort or a method employed when similar problems have already been addressed successfully using DL. For example, in domains like computer vision (CV), where there is ample data available, DL is the go-to solution. However, in cases where data is limited, other machine learning techniques such as support vector machines (SVM) may outperform DL.

One professional with experience in both traditional and deep learning methods reports that SVM works better in situations with less data. This highlights the importance of data quantity in determining the suitability of deep learning versus other methods.

The Impact of Data Size on Model Choice

The size of the data set is a critical factor in deciding whether to use deep learning. If there is a large volume of data, DL models are more likely to be employed. Conversely, when data is sparse, simpler and more traditional models might yield better results.

Deep Learning in Action: From Research to Application

As a machine learning (ML) research engineer, many projects involve the application of DL models. However, the scope of deep learning in the broader field of data science is not as universal. While some ML engineers like to explore deep learning frequently, data scientists often utilize a variety of tools to analyze and interpret data, depending on the problem at hand.

A data scientist from another organization reported that deep learning methods are not commonly used in their day-to-day work. They shared that only two instances in their career involved using DL, one for time-series data and the other for consumer behavior analysis. Both experiments yielded disappointing results, outperforming traditional methods like ARIMA and exponential smoothing.

Successful Alternatives to Deep Learning in Data Science

While deep learning might not always be the best choice, other machine learning techniques have proven highly effective. Gradient boosted trees, random forests, and generalized linear models are among the tools that have produced successful results for data scientists. Even conventional approaches like logistic regression can provide surprising accuracy in certain scenarios.

The Role of Data Characteristics

The nature of the data and its quality play a significant role in the choice of modeling methods. In industries where data is noisy, it may be challenging to achieve good results with neural network-based models. For instance, image data is often beyond the scope of current data sets, making image classification using deep learning impractical.

Similarly, text data can also pose challenges. Given the messy and incomplete nature of the text data, methods like word2vec, a type of deep learning model, are currently not applicable. Instead, traditional methods such as gradient boosted trees and random forests are more suitable.

In conclusion, the integration of deep learning in data science is not a one-size-fits-all solution. The choice of model heavily depends on the problem at hand, the size and quality of the data, and the context in which the data is being analyzed. For those looking to specialize in deep learning, there are organizations like Quantiphi that are actively seeking data scientists proficient in DL.

Final Thoughts

While deep learning is a powerful tool, its application in data science is not as ubiquitous as one might think. Data scientists must carefully consider the specifics of each project and choose the most appropriate model for the task. By understanding the strengths and limitations of various models, data scientists can optimize their approaches to achieve better results.