Artificial Intelligence-as-a-Service: Nervana Systems Allows Everyone Harness the Power of Big Data

As big data becomes more prevalent and available, Nervana Systems provides deep learning AI for companies that lack the data analytic capabilities to interpret this data in a meaningful way.

The Advent of Big Data

As technology is increasingly integrated into every aspect of daily life, more and more information is available through device connectivity due to the rapid expansion of the internet of things. With growing databases of customer data, market information, consumer trends, and product and manufacturing information, incredible opportunities exist for business models that can successfully harness the power of big data to customize or adapt their operating strategy. Companies have more data than they know what to do with but lack the data scientists or resources to develop analytics departments in order to interpret the data in meaningful and insightful ways [1].

While experts with a full understanding of the data can directly program algorithms to process these large data sets and produce useful insights, explicitly programmed algorithms have the potential for problems with the introduction of bias and mistakes that can impact different aspects of the calculations. Fueled by advances in computing technology, an emerging alternative method of data analysis is deep machine learning through neural networks, where programs are able to train themselves and provide accuracy of results beyond human capability.

What Are Neural Networks?

Neural networks are a computational approach which attempts to mimic the way the human brain solves problems through the inter connectivity of nodes and relational functions between them. These programs differ from traditional computing methods because they are not explicitly programmed, but instead are train themselves on massive data sets, examining inputs and outputs, and adjust their internal algorithms accordingly as they fine tune themselves to recognize trends, spot anomalies and understand the nuances of big data [2].

neuralnet

Figure 1: Neural Network Computational Illustration

Nervana Helps Everyone Take on Big Data

Nervana Systems offers a neural network platform in a software-as-a-service(SaaS) model called the Nervana Cloud that allows clients to develop custom deep learning software [3]. This provides companies unprecedented access to computing ability and analytical firepower previously available to a select handful of companies. Nervana delivers value by providing advanced computing and analytic tools that the vast majority of businesses could not implement themselves and allows them to tackle the business challenges that are too small or mundane to catch the attention of computing goliaths like Google and IBM’s Watson [4]. Nervana’s cloud based model allows Nervana to control and optimize the necessary computing hardware to complement their on-staff data analytics experts to efficiently implement neural nets for client companies. Nervana has previously partnered with agriculture companies to optimize crop yields, drilling companies to process underground imaging data to identify oil reserves more accurately, and many more to solve everyday business problems more efficiently than ever before [4, 5]. Nervana’s self-learning platform is inherently translatable across all industries and as the data sets grow, the capabilities are only limited by the capacity of the hardware performing the calculations.

Nervana has recently expanded its operating model in a quest to improve computing capabilities when it announced in February of 2016 that it would begin producing its own integrated circuits, optimized for neural net style computing [6]. This is operationally a massive risk for the company which up until this point has exclusively produced software. Nervana’s proposed chip is capable of processing up to 2.4 terrabytes of data per second, potentially establishing a key position in many data driven markets including driverless cars, a seemingly ideal neural net application due the control software’s need for adaptability, learning, and high rate of data processing [7]. Following the hardware announcement, Nervana was acquired by IBM in August of 2016 for $350 million. The acquisition by IBM and the support of IBM’s hardware infrastructure will allow quicker scaling of the deep learning as a service model and could potentially provide the manufacturing support required to make the hardware initiatives possible [3].

As the landscape of big data expands and Nervana attempts to scale its neural network SaaS model, more efficient computing technology is required to efficiently process these massive amounts of data. As long as neural networks remain a niche software product, processing hardware will continue to be optimized for mass market traditional computing and continue to limit the value that Nervana can provide. Pursuing custom-designed and optimized hardware is not only a next logical step for Nervana, but an absolute necessity to maximize the value for Nervana’s clients. By introducing a hardware practice, Nervana also protects itself from competitors in a purely software space with relatively low barriers to entry. With expanding connectivity between devices, data aggregation across industries, and increasingly automated world around us, Nervana must integrate hardware development in order to continue to deliver its core value proposition to customers and position itself to be a vital part of the connected world we live in. (768 Words)

[1] “Big Data: What it is and why it matter” http://www.sas.com/en_us/insights/big-data/what-is-big-data.html. Accessed 11/16/16.

[2] “Artificial Neural Network” https://en.wikipedia.org/wiki/Artificial_neural_network. Accessed 11/16/16.

[3] “Nervana Systems” https://en.wikipedia.org/wiki/Nervana_Systems. Accessed 11/16/16.

[4] “Nervana Systems Puts Deep Learning AI in the Cloud” http://spectrum.ieee.org/tech-talk/computing/software/nervana-systems-puts-deep-learning-ai-in-the-cloud. Accessed 11/16/16.

[5] Paradigm: Advanced Science for Everyone http://www.pdgm.com/products/stratimagic/ . Accessed 11/16/16.

[6] Nervana Engine: Hardware optimized for deep learning. https://www.nervanasys.com/technology/engine/. Accessed 11/17/16.

[7] Why Intel Bought Artificial Intelligence Startup Nervana Systems http://fortune.com/2016/08/09/intel-machine-learning-nervana/. Accessed 11/17/16.

Images Source: Artificial neural networks are changing the world. What are they? https://www.extremetech.com/extreme/215170-artificial-neural-networks-are-changing-the-world-what-are-they. Accessed 11/17/2016.

Previous:

NEST: our real-life J.A.R.V.I.S?… Perhaps, but not quite yet.

Next:

Under Armour Wants Your Data

2 thoughts on “Artificial Intelligence-as-a-Service: Nervana Systems Allows Everyone Harness the Power of Big Data

  1. Great post Matt. I think one key factor that Nervana (and by extension) IBM have to think through is how they will relate to the open source ecosystem. As many of the larger pioneers of Deep Neural Nets achieve research breakthroughs, they are beginning to open-source not only their NN training libraries (e.g., Google Tensor Flow [1] ) but also their custom-built ASIC designs (e.g., Google TPU’s [2] ). While this still does not solve the talent-shortage associated with implementing these methods, it does raise the question as to how defensible it will be to sell implementation of analytical techniques that can now be accessed through turnkey libraries and run on specialized processors that could easily be available as part of a cloud infrastructure stack.

    [1] https://www.tensorflow.org/versions/r0.11/get_started/index.html
    [2] https://cloudplatform.googleblog.com/2016/05/Google-supercharges-machine-learning-tasks-with-custom-chip.html

  2. Really interesting piece Matt! This is definitely a challenging space, and knowing how to interpret and effectively use data is something that a lot of tech/software companies are struggling with — even when you have an analytics department and data scientists for instance, it can be really challenging to gain takeaways / actual insights that are actionable for your product. For instance, just because you know the behavior that the users of your product are exhibiting (e.g. clicks, pages visited, etc.) doesn’t mean you know why they are doing what they are doing or how to get them to perform the actions you want them to take.

Leave a comment