Agnik International is Developing New Scalable Distributed Machine Learning Architecture for Large Language Models and Physical AI Applications

BusinessWire India
Kolkata (West Bengal) [India], January 29: Agnik International, a leading data science company with market-leading analytic products, today announced that they are developing a new distributed machine learning architecture based on decades of peer-reviewed research. This architecture and its underlying algorithms will be used to scale AI applications across various domains, including large language models (LLM), vehicle analytics, and agentic controls for physical AI.
Currently, most popular machine learning systems rely on a specific deep learning architecture known as Transformers. Common GPU-based implementations of these systems make use of synchronous, tightly-coupled computing infrastructure. Despite the remarkable gains Transformers have made in advancing the field, this approach requires massive, special-purpose computing infrastructure and large data centres with energy demands that are difficult to sustain. Furthermore, there remains significant room to improve the computational efficiency for the training and inferencing algorithms. While large investments are being poured into the AI technology based on this approach, it is not yet clear whether the fundamental problems of developing efficient polynomial-complexity algorithms for deep machine learning and the related computability challenges have been adequately addressed.
Agnik International is developing a different architecture based on asynchronous local algorithms using loosely-coupled distributed computing environments. This methodology differs significantly from the traditional synchronous, tightly-coupled Transformer-based approach. The core research-team-members at Agnik International have a long track record of developing scalable distributed machine learning systems across academia, industry, and government. This new architecture will support scaling of common AI tasks, such as learning and inferencing, at a lower computational cost with reduced power consumption and better fault tolerance.
"We have a decades-long history and extensive research experience in developing algorithms, architectures, and systems for distributed machine learning," said Dr. Hillol Kargupta, President of Agnik Group of Companies. "I am excited to see the team making fundamental progress for both science and humanity."
This team's prior work in distributed machine learning--specifically in the fields of connected vehicles and the Internet of Things (IoT)--pioneered edge-analytics-based architecture for mobile and embedded applications. This resulted in a patented architecture that produced several popular connected vehicle products widely adopted in the global market. The dominance of the Agnik Group of Companies in the vehicle analytics market is a direct result of its long history in developing state-of-the-art distributed machine learning applications for its global customers.
"In early 2000s, when sending large volume of raw telematics data from vehicle to the cloud was impractical due to limited wireless bandwidth, we pioneered distributed machine learning technology with onboard data stream mining algorithms running on resource-constrained hardware mounted in vehicles," said Kakali Sarkar, COO of Agnik Group of Companies. "I am happy to see the exciting progress we are making in the field of large language models using scalable distributed algorithms. We look forward to offering a groundbreaking alternative architectural solution for LLMs, just as we did for connected vehicles."
(ADVERTORIAL DISCLAIMER: The above press release has been provided by BusinessWire India. ANI will not be responsible in any way for the content of the same)