Virtualization Technology News and Information
Article
RSS
Scalable AI starts with Pre-configured Models
No longer can firms categorize AI as "nice to have". The past couple of years have proven that embedding AI across the enterprise to tap its full business value clearly puts progressive organizations on that much sought-after accelerated trajectory to success. This, however, requires making the shift from bespoke builds to an industrialized AI factory for the enterprise. Leaders who have been able to quickly commit to that transformation rigor have achieved a level of performance that peers envy and the market seeks to emulate.

Scaling and Democratizing AI is not without its challenges

Much of the AI revolution is already upon us. New use-cases - be it autonomous retail, smart manufacturing, intelligent supply chains or accelerated drug discovery - continue to show us the potential of all that's possible. But the promised land, assured by AI adoption, is still out of the grasp of several organizations. My experience, working with enterprises - both local majors and transcontinental behemoths - points to two major inter-related impediments that need to be addressed before AI can truly deliver sustainable value.

The first; For any kind of transformation to materialize and make a lasting difference, AI must be accessible to everyone - ground level up, and herein comes the need to democratize AI. Since the individual learning curve for complex AI models is very steep, there is a barrier that filters out the select few who can grapple with the arduous process of AI training and model building. This restricts the innovations that otherwise would have naturally emerged - with widening scope and scale.

This relates to the second aspect of the impediment deterring scalability. Fewer people with access to make-with-AI means building isolated silos of AI innovation. These silos are obviously not conducive when the organization is looking to fire on all cylinders to gain competitive edge - and the shrinking timeframes and soaring expectations in the market only exacerbate the situation. Enterprises need to scale fast and expedite the adoption process across the value chain.

Not only is it imperative for enterprises to get the first-mover advantage to reap the exponential benefits of AI, but they can accelerate the journey by being cautious and avoiding the potential missteps associated with the journey.

The way forward

There is a simple conceptual solution that can alleviate this problem and that's the use of pre-configured and ready to use trained AI models.

Developing, training, and optimizing production-quality models is expensive, requiring numerous iterations, domain expertise, and countless hours of computation. Pretrained models can prove to be effective because they "learn" from representative datasets and can be fine-tuned over time with appropriate weightages associated with select data along with checks and balances for biases. These models can be easily customized in a fraction of the effort that it would take to develop new AI-based solutions from scratch.

A repository of pre-built and trained models can enable organizations to readily use and operationalize their models, without undertaking green-field development each time. This will also enable citizen developers and relatively unskilled talent to participate in the build and adopt process. Templatizing AI solution building, in this manner, can significantly reduce the time to market of what would have otherwise been a lengthy deployment process for AI-powered business innovations.

A glimpse into some applications

The applications for computer vision, object detection, speech processing, or any kind of use for AI-based solutions, that use AI models, can become a simple matter of customization.

For example, in a retail firm, one can readily deploy computer vision models for inventory management, planogram compliance, and other asset protection solutions by simply using pre-trained object detection models and then customizing it for their specific SKUs. Similarly, in manufacturing, one can use pre-trained vision models and algorithms for preventive maintenance and fault detection, by just customizing it for the equipment in question, while the underlying model remains the same.

An idea that can prove to be a great asset on the journey

Instituting an AI store with pre-curated and a continuously growing number of assets including datasets, AI models and AI services across different domains and verticals can prove useful. Cross-functional users can then have ready access to a growing portfolio of indigenous AI solutions to solve their business problems. For example, an American bank used one of Infosys' solutions from our own AI-store to build out their NLP-based expense claims management mobile app.

Other solutions like image analytics, video analytics, document digitization platforms are some of the inbuilt accelerators that can truly benefit from such prebuild AI models and are already being used widely across industries to transform and do it at scale - often while meeting aggressive timelines.

In summary

Embracing pre-configured AI models is one of the pivotal steps that enterprises can commit to and invest in to accelerate AI democratization, incentivize adoption, and achieve cross-functional collaboration across levels. After all, for AI to make a sizable contribution to a company's bottom line, the organization must scale the technology in a way that it infuses in core business processes, workflows, and customer journeys. Achieving such scale requires a highly efficient AI production line - and pre-configured AI models are decidedly a smart bet.

##

ABOUT THE AUTHOR

Balakrishna DR 

Balakrishna DR, popularly known as Bali, is a Senior Vice President and heads the delivery for Energy, Communications, Services (ECS) business unit of Infosys catering to Energy, Utilities, Telecommunication, Media, Entertainment and Services industries.

He also heads the AI and Automation unit and is responsible to drive both internal Automation for Infosys and for providing independent automation services leveraging market leading products for clients.

Bali has been with Infosys for more than 25 years and has played sales, program management and delivery roles across different geographies and industry verticals.

Published Wednesday, August 24, 2022 7:35 AM by David Marshall
Filed under: , ,
Comments
There are no comments for this post.
To post a comment, you must be a registered user. Registration is free and easy! Sign up now!
Calendar
<August 2022>
SuMoTuWeThFrSa
31123456
78910111213
14151617181920
21222324252627
28293031123
45678910