Why is deep learning important today?
Deep learning is one of the foundations of cognitive computing. The current interest in deep learning is due in part to the buzz surrounding cognitive computing, software applications that understand human input and can respond in humanlike form or output. Deep learning techniques have greatly improved our ability to classify, recognize, detect and describe – in one word, understand. Many applications are in fields where these tasks apply to non-numerical data, for example, classification of images, recognizing speech, detecting objects and describing content. Systems such as Siri and Cortana are powered in part by cognitive computing and driven by deep machine learning.
Several developments are now advancing deep learning. Algorithmic improvements have boosted the performance of deep learning methods, and improving the accuracy of machine learning approaches creates massive business value. New classes of neural networks have been developed that fit particularly well for applications like text translation and image classification. Neural networks have existed for almost 50 years. They had fallen out of favor by the late 1990s due to difficulties in getting good accuracy in the absence of large training data. However, two changes within the last decade have revolutionized their use:
- We have a lot more data available to build neural networks with many deep layers, including streaming data from the Internet of Things, textual data from social media, physicians notes and investigative transcripts.
- Computational advances of distributed cloud computing and graphics processing units have put incredible computing power at our disposal. This level of computing power is necessary to train deep algorithms.
At the same time, human-to-machine interfaces have evolved greatly as well. The mouse and the keyboard are being replaced with gesture, swipe, touch and natural language, ushering in a renewed interest in cognitive computing.
Deep learning opportunities and applications
Due to the iterative nature of deep learning algorithms, their complexity as the number of layers increase, and the large volumes of data needed to train the networks, a lot of computational power is needed to solve deep learning problems.
Traditional modeling methods are well understood, and their predictive methods and business rules can be explained. Deep learning methods have been characterized as more of a black-box approach. You can prove that they perform well by testing them on new data. However, it is difficult to explain to decision makers why they produce a particular outcome, due to their nonlinear nature. This can create some resistance to adoption of these techniques, especially in highly regulated industries.
On the other hand, the dynamic nature of learning methods – their ability to continuously improve and adapt to changes in the underlying information pattern – presents a great opportunity to introduce less deterministic, more dynamic behavior into analytics. Greater personalization of customer analytics is one possibility.
Another great opportunity is to improve accuracy and performance in applications where neural networks have been used for a long time. Through better algorithms and more computing power, we can add greater depth.
While the current market focus of deep learning techniques is in applications of cognitive computing, SAS sees great potential in more traditional analytics applications, for example, time series analysis.
Another opportunity is to simply be more efficient and streamlined in existing analytical operations. Recently, SAS experimented with deep neural networks in speech-to-text transcription problems. Compared to the standard techniques, the word-error-rate (WER) decreased by more than 10 percent when deep neural networks were applied. They also eliminated about 10 steps of data preprocessing, feature engineering and modeling. Computationally, it might take longer to train the deep network compared to the traditional modeling flow, but the impressive performance gains and the time savings when compared to feature engineering signify a paradigm shift.