AI (artificial Intelligence) concept.
During the past decade, deep learning has seen groundbreaking developments in the field of AI (Artificial Intelligence). But what is this technology? And why is it so important?
Well, let’s first get a definition of deep learning. Here’s how Kalyan Kumar, who is the Corporate Vice President & Chief Technology Officer of IT Services at HCL Technologies, describes it: “Have you ever wondered how our brain can recognize the face of a friend whom you had met years ago or can recognize the voice of your mother among so many other voices in a crowded marketplace or how our brain can learn, plan and execute complex day-to-day activities? The human brain has around 100 billion cells called neurons. These build massively parallel and distributed networks, through which we learn and carry out complex activities. Inspired from these biological neural networks, scientists started building artificial neural networks so that computers could eventually learn and exhibit intelligence like humans.”
Think of it this way: You first will start with a huge amount of unstructured data, say videos. Then you will use a sophisticated model that will process this information and try to determine underlying patterns, which are often not detectable by people.
“During training, you define the number of neurons and layers your neural network will be comprised of and expose it to labeled training data,” said Brian Cha, who is a Product Manager and Deep Learning evangelist at FLIR Systems. “With this data, the neural network learns on its own what is ‘good’ or ‘bad.’ For example, if you want the neural network to grade fruits, you would show it images of fruits labeled ‘Grade A,’ ‘Grade B,’ ‘Grade C,’ and so on. The neural network uses this training data to extract and assign weights to features that are unique to fruits labelled good, such as ideal size, shape, color, consistency of color and so on. You don’t need to manually define these characteristics or even program what is too big or too small, the neural network trains itself using the training data. The process of evaluating new images using a neural network to make decisions on is called inference. When you present the trained neural network with a new image, it will provide an inference, such as ‘Grade A with 95% confidence.’”
What about the algorithms? According to Bob Friday, who is the CTO of Mist Systems, a Juniper Networks company, “There are two kinds of popular neural network models for different use cases: the Convolutional Neural Network (CNN) model is used in image related applications, such as autonomous driving, robots and image search. Meanwhile, the Recurrent Neural Network (RNN) model is used in most of the Natural Language Processing-based (NLP) text or voice applications, such as chatbots, virtual home and office assistants and simultaneous interpreters and in networking for anomaly detection.”
Of course, deep learning requires lots of sophisticated tools. But the good news is that there are many available and some are even free like TensorFlow, PyTorch and Keras.
“There are also cloud-based server computer services,” said Ali Osman Örs, who is the Director of AI Strategy and Strategic Partnerships for ADAS at NXP Semiconductors. “These are referred to as Machine Learning as a Service (MLaaS) solutions. The main providers include Amazon AWS, Microsoft Azure, and Google Cloud.”
Because of the enormous data loads and complex algorithms, there is usually a need for sophisticated hardware infrastructure. Keep in mind that it can sometimes take days to train a model
“The unpredictable process of training neural networks requires rapid on-demand scaling of virtual machine pools,” said Brent Schroeder, who is the Chief Technology Officer at SUSE. “Container based deep learning workloads managed by Kubernetes can easily be deployed to different infrastructure depending upon the specific needs. An initial model can be developed on a small local cluster, or even an individual workstation with a Jupyter Notebook. But then as training needs to scale, the workload can be deployed to large, scalable cloud resources for the duration of the training. This makes Kubernetes clusters a flexible, cost-effective option for training different types of deep learning workloads.”
Pros and Cons
Deep learning has been shown to be quite efficient and accurate with models. “Probably the biggest advantage of deep learning over most other machine learning approaches is that the user does not need to worry about trimming down the number of features used,” said Noah Giansiracusa, who is an Assistant Professor of Mathematical Sciences at Bentley University. “With deep learning, since the neurons are being trained to perform conceptual tasks—such as finding edges in a photo, or facial features within a face—the neural network is in essence figuring out on its own which features in the data itself should be used.”
Yet there are some notable drawbacks to deep learning. One is cost. “Deep learning networks may require hundreds of thousands or millions of hand-labeled examples,” said Evan Tann, who is the CTO and co-founder of Thankful. “It is extremely expensive to train in fast timeframes, as serious players will need commercial-grade GPUs from Nvidia that easily exceed $10k each.”
Deep learning is also essentially a “black box.” This means it can be nearly impossible to understand how the model really works!
“This can be particularly problematic in applications that require such documentation like FDA approval of drugs and medical devices,” said Dr. Ingo Mierswa, who is the Founder of RapidMiner.
And yes, there are some ongoing complexities with deep learning models, which can create bad outcomes. “Say a neural network is used to identify cats from images,” said Yuheng Chen, who is the COO of rct studio. “It works perfectly, but when we want it to identify cats and dogs at the same time, its performance collapses.
But then again, there continues to be rapid progress, as companies continue to invest substantial amounts into deep learning. For the most part, things are still very much in the nascent stages.
“The power of deep learning is what allows seamless speech recognition, image recognition, and automation and personalization across every possible industry today, so it’s safe to say that you are already experiencing the benefits of deep learning,” said Sajid Sadi, who is the VP of Research at Samsung and the Head of Think Tank Team.