Organizations are using deep learning (DL), a growing branch of artificial intelligence (AI), to streamline their operations and increase productivity.
See below how several organizations in various industries are applying deep learning to deliver business outcomes:
5 deep learning case studies
Zendesk is a software as a service (SaaS) provider that helps companies create strong customer relationships that facilitate productivity and growth.
As their user base grew, Zendesk needed to find a way to keep up with customers wanting to find answers to their questions as fast as possible. However, routing them to talk to a support agent isn’t scalable and would still mandate wait time. Zendesk addressed this challenge by creating Answer Bot using deep learning.
Using neural networks, Zendesk developed a virtual customer assistant that’s able to answer customer questions using content straight from the Zendesk Guide knowledge base.
“For Answer Bot, we liked the idea that a deep learning model could help the application continually fine-tune itself to give customers the best possible answers,” said Soon-Ee Cheah, a data scientist at Zendesk. “We can scale our deep learning models very efficiently using GPU-processing power on AWS, and that will benefit us while we grow our applications to accommodate more customers.”
Deep learning solutions: Amazon Simple Storage Service (S3), Amazon EC2, Amazon Aurora, and Amazon SageMaker.
- Instantaneous answers to customer questions
- Scalable software infrastructure to meet customer demand
- Quick to train and deploy.
Read the full Zendesk and Amazon SageMaker case study.
2. 90 Seconds
90 Seconds is a video creation platform that regularly manages 12,000 video professionals in over 160 countries. While they started as a low-profile business in New Zealand, their growth forced their hand into using more tech in their operation to keep up with the rise in demand.
By working alongside the Google Cloud Platform, they’re able to train deep learning algorithms to analyze videos and provide relevant analysis for brands. The algorithms are also able to identify and extract specific content from videos, like footage of sunsets or people, and analyze how they contribute to the performance of the video in terms of viewer count and social media engagement.
“Google Cloud Platform has played a key role in helping our business grow to this point,” said Dat Le, director of data science and engineering at 90 Seconds. “We see technologies like Cloud Vision API, Cloud Video Intelligence, and Cloud AutoML helping us become a more intelligent, valuable provider to brands in future.”
Industry: Media production
Deep learning solutions: Cloud Vision API, Cloud Video Intelligence, Kubernetes Engine, Compute Engine, Cloud SQL, BigQuery, and Cloud AutoML.
- Scalable solution that supports the growing demand for cloud video production
- Accelerates software development
- Facilitates decision-making by capturing and analyzing data from multiple services
- Supports an online marketplace of 12,000 videos creative professionals and 3,000 brands
Read the full 90 Seconds and Google Cloud Video Intelligence case study.
3. Princeton University: ITER Fusion Energy
Researchers at the international ITER Facility, the world’s largest experimental nuclear fusion reactor, are looking to become pioneers in clean energy by harnessing power generated through nuclear fusion. Unlike nuclear fission, fusion is inherently safe and doesn’t result in radioactive waste.
However, sustaining fusion reactors is challenging as a lot could go wrong. To be successful and safe, ITER needed a way to reliably predict, respond to, and minimize—or entirely avoid—any disruption that can hinder the process.
Researchers at Princeton University developed an advanced deep learning Fusion Recurrent Neural Network (FRNN) predictive code using NVIDIA P100 GPUs.
“Fusion has the exciting potential to be the future of sustainable clean energy, with ITER representing the next major step in this direction,” said William Tang, a professor at Princeton University’s Program in Plasma Physics and the Princeton Plasma Physics Laboratory. “Deep learning predictive capability powered by NVIDIA’s advanced GPUs is helping accelerate progress towards making this vision a reality in our lifetime.”
Industry: Higher education and energy
Deep learning solutions: NVIDIA Tesla P-100 GPU, Google Tensorflow, Theano, NVIDIA CUDA, and NVIDIA cuDNN.
- Achieved 90% accuracy with predictions
- 5% false positives
- Minimum 30 milliseconds before the onset of disruptions
Read the full Princeton University: ITER Fusion Energy and NVIDIA Tesla p-100 GPU case study.
4. Elektronische Fahrwerksysteme
Elektronische Fahrwerksysteme (EFS) is a technology partner of the automotive manufacturer Audi. They’re responsible for examining, developing, and implementing future-forward technologies in Audi’s vehicles, including their latest project of automated driving.
By partnering up with Microsoft Azure and NVIDIA, EFS researchers used tech from both companies to create a deep learning AI solution that’s able to analyze high-resolution, two-dimensional images of roads. Using Microsoft Azure and NVIDIA Tesla p-100 GPUs to build their deep learning architecture, they gamified the deep learning process by taking the images for teaching the algorithm from a video game that generates labeled images.
“As part of our autonomous driving research for Audi, we proved that it’s possible to use deep learning to analyze roads. That is a really big deal,” said Max Jesch, a software developer at EFS. “As far as we know, EFS is the first company to do it on such a large scale.
“We ultimately based our decision to choose Azure on the relationship we have with Microsoft and with the people who represent it and support us. The cooperation we got on this project has been awesome.”
Industry: Technology and automotive
Deep learning solutions: NVIDIA Tesla p-100 GPUs, Microsoft Azure, Microsoft Azure Storage, and Microsoft Azure Virtual Machines.
- Ability to use 2D road images instead of 3D
- Safer driving in low-visibility driving conditions
- Scalability with Azure Virtual Machines
- Saving learning time by compressing high-resolution images
Read the full Audi and EFS, and Microsoft Azure Virtual Machines case studies.
5. NASA: Ames Research Center
The Ames Research Center at NASA needed an efficient way to monitor the climate of the earth in order to track the changes in temperatures, sea levels, and acidity and how they’d affect crop yields and overall vegetation patterns. However, using automated satellite image classification is incredibly challenging as the data provided is highly variable, and there isn’t sufficient training data.
To overcome this obstacle, NASA developed DeepSat, a deep learning framework for processing satellite images through classification and segmentation. Thanks to the high resolution of satellite images DeepSat is able to provide, scientists can use the data in simulators and track the signs of changing landscapes.
“Our best network dataset produced a classification accuracy of 97.95% and outperformed three state-of-the-art object recognition algorithms by 11%,” said Sangram Ganguly, senior research scientist at NASA Ames Research Center.
Industry: Defense and space
Deep learning solutions: NVIDIA Tesla GPUs, NVIDIA DIGITS DevBox, cuDNN, Caffe, Torch, and Tensorflow.
- Average image size of 6000 x 7000 pixels
- Increasing the input size allowed with less noise
- Reducing training time
- 97.95% classification accuracy
- Outperforming previous algorithms by 11%
Read the full NASA: Ames Research Center and NVIDIA Tesla GPUs case study.