Deep learning skills are often portrayed as a dark art. There’s a common misconception that you need a PhD or a high level of technical expertise to use deep learning—and that’s no longer the case.
While deep learning can involve complex algorithms and computational processes, developers and scientists are creating user-friendly applications and high-level methods to help your organization get started with this technology. These developments have made it easier than ever for professionals with basic mathematical aptitude and coding skills to learn and develop deep learning skills.
Your organization can use, adapt, and hone deep neural networks to detect complex data patterns by adjusting parameters in pre-trained networks to find answers to questions at hand with a very high level of accuracy. Below are some real-life deep learning use cases for inspiration on how to get your organization started.
Convolutional neural net (CNN) architectures use a series of filters to trace different transformations in an image. These filters in tandem with a fully connected neural network help the deep learning system predict the probability of an image depicting an item from a given category, like cats, trees, chairs, etc.
The network grows more sophisticated as more images in more categories are fed into it. Once a model is created, by “seeing,” for example, millions of samples in a thousand different categories, it can then be reused and adjusted in different domains. For instance, a pretrained CNN model can have especially useful applications in fields like medicine: by feeding the network MRI images and adjusting its parameters, it is, possible to detect particular disease markers or anomalies.
CNNs also have the potential to help with facial recognition. The detection process might include characteristics like sex and age or emotions like anger and surprise. While facial identification has security uses, it also has useful business applications, including helping firms better target potential customers with personalized ads.
As a second use case, consider recurrent neural nets (RNN) for text and language processing. In RNNs, a neural network has connections that loop back to itself, allowing it to recollect what it has seen before. This is very useful when predicting the next item based on what has occurred previously. Similarly to CNNs, you can train RNNs by feeding the network data sets. In this case, the sets are text-based rather than image-based.
RNN processing helps users analyze the sentiment of a sentence or a block of text. Applications might include enabling an e-commerce firm to review hundreds of customer reviews for similar sentiments or to trawl social media data sets for positive or negative statements.
You can start working with CNNs and RNNs in your own business anytime. It’s now possible to download freely available models that are based around pretrained networks. As a user, you can search the web, find a model that roughly corresponds with your requirements—like image search analysis—and tweak the model to create a more effective match with your business problem.
A great starting place is DAWNBench, a tool that provides a benchmark of CNNs and RNNs. Many RNNs will already have been pretrained by using significant amounts of text in existing databases, such as the written knowledge available freely through Wikipedia. It’s possible to take one of these pretrained models and tweak it for your own business case, such as searching for patterns in a social feed or legal document.
The best part is that many deep learning data models are freely available. That means your users can download the model, adjust the parameters, and then feed the model data. As you adjust the parameters, the deep learning model will specialize for your unique use case. To start this process, you just need to ensure your business has the right supporting infrastructure in place.
Once you’re satisfied with how much your network has learned, the trained network and its associated data set can be pushed to a platform in the cloud or an on-premises data center. Either solution can host the trained model and make it accessible to whoever might want to use it via a web-based application.
Look for a partner with an entire suite of products to help support your deep learning demands. As a growing amount of correctly labeled data is brought into your network, you’ll need a scalable platform that can handle terabytes of information. Start with a common service fabric that allows you to manage, provision, and govern data in a secure fashion.
Your partner should also help you deal with workload and application requirements (like data ingestion, storage, and processing) and compute requirements (like CPU, GPU, and memory). Once your deep learning network has been tested and is ready to roll out, your partner should help you deploy it in your chosen environment, whether that’s cloud, on-premises, or a hybrid of both.
Now is the time to get involved in deep learning. While there’s a great deal of hype surrounding emerging technology, there is still an opportunity for early movers to gain a significant advantage. Even more crucially, there are freely available courses to help your users start building their deep learning skills. Although deep learning might seem complex, it’s easy for interested individuals to use free resources to quickly build a prototype and test its applicability across a range of use cases.
Learn more about how to start applying machine learning and AI in your organization.