Startup Makes Artificial Intelligence Far Easier to Use in M&E Operations

Rejeev Dutt, CEO, DimensionalMechanics, Inc.

Rejeev Dutt, CEO, DimensionalMechanics, Inc.

DimensionalMechanics says AI Expertise Is no Longer Essential to Leveraging Technology

By Fred Dawson

June 22, 2017 – We might not yet have access to genies in bottles, but how about a means by which people with no experience in artificial intelligence can put AI to use in whatever ways are beneficial to their businesses?

This is the vision startup DimensionalMechanics, Inc. (DMI) has brought to fruition for a rapidly expanding base of media and entertainment industry customers who are getting a leg up on competitors by executing tasks that would otherwise be beyond their reach. Exemplifying one area of M&E operations that stands to benefit from use of AI is a recently announced strategic alliance between DMI and GrayMeta, a provider of automated metadata collection, curation and search applications to M&E firms such as ABC and CBS.

AI, which not too long ago was dismissed as a futurist fantasy, is now widely used in robotics and myriad routine applications across the enterprise landscape to guide computers in deep human-like learning processes that continually adapt processes engineered for certain tasks to changing conditions. In M&E operations AI is fast becoming an essential component in a wide range of applications such as content recommendation, service personalization, addressable advertising, voice recognition, network diagnostics and much else.

For GrayMeta the ability to quickly implement AI-based solutions in the many scenarios involving use of metadata across the M&E market is a significant benefit to its business, says GrayMeta CEO Tom Szabo. “DimensionalMechanics will give us the ability to rapidly deploy highly customizable AI solutions to dramatically improve content discovery and recommendations for our customers,” Szabo notes.

One of GrayMeta’s recently announced customers is Levels Beyond, which, as previously reported, provides tools to broadcasters enabling instant access to archived video in live sports coverage and other aspects of production. The GrayMeta Platform creates searchable metadata which can identify faces, people, logos, speech, tags, descriptions, on-screen text and other elements in video streams.

GrayMeta founder and executive chairman Tim Stockhaus has also made known the firm’s use of AI in conjunction with integration of its technologies with Microsoft Azure to enable users to easily search and find content in all Microsoft Office products, including emails held on their networks. “As machine learning technologies rapidly evolve, new capabilities and avenues for productivity are being added every day,” Stockhaus notes.

The process of modeling AI architectures suited to the unique needs of each application category and specific solutions within those categories is, as DMI CEO Rejeev Dutt puts it, “closer to art than science.” Applying AI in highly specialized areas with unique modalities suited to meeting each company’s needs typically requires engagement of a team of machine-learning specialists who not only understand AI but have sufficient knowledge of the application category to ensure development of useful AI architectures.

Moreover, ongoing changes in conditions not accounted for in the initial AI modeling process may require sustained involvement of an expert team to ensure the AI mechanisms remain relevant to the tasks at hand. Experts with the right combination of machine-learning and application-specific skills are hard to come by, even if a company can afford to employ them.

“Our core purpose is to make AI more accessible, reducing the difficulties involved and making it easier for people to bring models up faster,” Dutt says. “The second part is about how you manage and modify those models over time.”

DMI customers can build and continually refine AI architectures without requiring the retention of machine-learning specialists, he notes. And for customers who do have access to such expertise, personnel can utilize DMI resources to achieve optimal results much faster, he adds.

As explained by Dutt, an AI architecture is the blueprint for a “neuro-network” or interconnected network of intelligent processing layers that describes their connectivity and points of communication, the machine-learning rates within and across the layers and the overall capacity for learning. “To try and figure out how to build a neuro-network for specific problems is hard,” he says. “A machine-learning expert can do this, but it’s hard for others, including IT guys.”

But even for the machine-learning experts making AI useful in everyday operations is a major challenge. Not only is it hard to build architectures that are broadly applicable to all conditions; customizing architectures to be more flexible is hard as well.

For example, a highly accurate voice recognition system might be stymied by someone whose voice is outside the pitch ranges the system is designed to work with, or a system adept in identifying certain types of images might draw a blank on others. Ideally, the intrinsic capabilities of machine learning would make the necessary adjustments, but often that’s not the case. “Deep learning models are very vertical,” Dutt notes.

Further complicating matters is the fact that contracting with machine-learning pros to build and adjust AI architectures traditionally has required a willingness on the part of companies to expose data off premises. “A lot of companies are reluctant to use AI if it means comprising data privacy,” Dutt says.

And then there’s the question of payoff. While some AI applications are built solely for internal use, others have marketable potential that companies would like to exploit. But, notes Dutt, there hasn’t been a common marketplace for organizations to share or monetize their AI creations.

DMI has introduced a portfolio of solutions within its NeoPulse Framework that address all these issues, he says. First and foremost, when it comes to building AI models, DMI has demonstrated that IT personnel with no background in machine learning can do the job using DMI’s technology. “You need some coding experience to build sophisticated models,” he says, “but the bigger challenge is getting and curating the data you want to use with the model.”

DMI’s NeoPulse Modeling Language provides customers an efficient and easy-to-use means of automating the process of building AI models. By using AI as an “oracle” to build AI, DMI is continually enhancing customers’ capabilities as its proprietary technology acquires ever more knowhow.

“Oracle gets smarter over time, constantly learning from previous models,” Dutt says. “Very often it’s not easy even for us to decide the right architecture.” In the building process, the platform leverages what’s already been learned to figure out what looks to be the optimal architecture and then begins a testing and refinement process that quickly leads to the best possible starting point, he explains.

“We just built an application involving sentiment detection technology that only took 18 lines of code,” Dutt notes. “If we’d done that with TensorFlow [an AI software library] and Python [the language used with TensorFlow] it would have taken more than 600 lines.”

DMI is preparing a new release of Oracle enabling more sophisticated analyses of video. “I can’t say much about this yet, but it will allow an IT team to address some big problems,” Dutt says.

DMI’s new product release, NeoPulse AI Studio, leverages architectures built with Oracle in six general categories of operations important to M&E companies and other entities. These include image recognition, object recognition, recommendations, identification of unacceptable content, detection of infrastructure anomalies associated with intrusion or other malicious activities and character recognition in conjunction with Japanese and other languages using ideographic symbols.

These application frameworks allow IT teams without AI experience to quickly create optimal AI architectures suited to their specific requirements, meeting expectations for human-like discernment at speeds and often with accuracy beyond what can be expected based on traditional norms. As an example Dutt cites a recent instance where a test of AI-enabled search for prohibited adult content resulted in identification of nudity from a grainy old film displayed on a TV screen in the background of a scene from a TV program that had drawn a fine from regulators after going undetected by the programmer’s monitoring team.

In another test, a DMI-developed AI application assigned the task of ranking the likely appeal of headlines for news articles made choices that closely aligned with choices made by producers. “Digital broadcasters want to generate more clicks by making content as interesting as possible to their audiences,” Dutt notes. “Having confidence they can do this automatically whether it’s with the choice of headlines, images or other elements will save them a lot of time.”

DMI’s solutions also address the previously mentioned data privacy concerns that inhibit wider utilization of AI. By enabling creation and ongoing refinement of AI architectures by in-house staff, AI Studio allows customers to keep high-value data on premises and to ensure they are using the most current data to retrain their models, Dutt says.

At the same time, he adds, the company offers a cloud-based licensing model for companies that can’t afford to build models in house. “For example, you can port your voice recognition to run as an AI application on our platform,” Dutt explains. “If someone builds an app that uses voice recognition, they can call on your model and the voice recognition comes back.”

“If your app does really well, millions of users are calling your AI, and we earn royalties on the run times,” he adds. This license-free royalty-base monetization model also applies to AI applications built in house by customers using AI Studio.

The opportunity to monetize AI-enabled applications in an environment where there’s been no common way for entities to do that is another benefit touted by DMI. DMI’s NeoPulse AI Store operates like a traditional app store enabling organizations to distribute, license and monetize their AI models.

Whether apps are developed on customers’ premises or on the DMI cloud, they can be pushed to the cloud-based AI Store for broader consumption, Dutt says. AI Studio scales exponentially as successful AI models in the store are licensed and used by additional users and organizations, he adds.

While M&E is a great proving ground for building momentum in the AI marketplace, DMI has its sights on many other fields as well. “We’re just at the point where we’ve released our primary product and are heading into the next realm of funding,” Dutt says, noting DMI has raised $6.7 million so far. “We now have 22 pre-orders from Fortune 500 companies.”